Bump Assistent to 0.7.0: Config personas, skills, and vector memory.

Move prompts into Config/_base and persona folders; seed model facts into SQLite via Ollama embed and retrieve as memory_hits each turn.

Co-authored-by: Cursor <cursoragent@cursor.com>

Assets/assistent.css

@@ -556,6 +556,23 @@
     font-size: 0.9rem;
 }
 
+.sa-skills-label {
+    font-size: 0.85rem;
+    opacity: 0.85;
+    margin-top: 0.25rem;
+}
+
+.sa-skills-box {
+    display: flex;
+    flex-wrap: wrap;
+    gap: 0.35rem 0.75rem;
+    margin-bottom: 0.25rem;
+}
+
+.sa-settings .sa-select {
+    width: 100%;
+}
+
 .sa-messages {
     flex: 1;
     overflow: auto;

Assets/assistent.js

@@ -1,10 +1,11 @@
 /**
  * Swarm Assistent — Krea 2 collaborative chat (Ollama via Swarm API).
- * v0.6.0: board tabs, Cards form+Civitai fetch, LoRA chips, taste on disk, reliability fixes.
+ * v0.7.0: Config presets, persona folders, vector memory, chat|memory model roles.
  */
 (function () {
     const LS_BASE = 'swarm_assistent_base_url';
     const LS_MODEL = 'swarm_assistent_model';
+    const LS_EMBED = 'swarm_assistent_embed_model';
     const LS_PACK = 'swarm_assistent_pack';
     const LS_PERSONA = 'swarm_assistent_persona';
     const LS_VIEW = 'swarm_assistent_view';
@@ -23,7 +24,7 @@
     const MAX_REF_SLOTS = 4;
     const CONTEXT_PROMPT_MAX = 2000;
 
-    const ASPECT_TABLE = {
+    let ASPECT_TABLE = {
         '1:1': [1024, 1024],
         '4:3': [1184, 896],
         '3:2': [1248, 832],
@@ -34,7 +35,7 @@
         '9:16': [768, 1376],
     };
 
-    const PACK_ALIASES = {
+    let PACK_ALIASES = {
         write: 'write_prompt',
         write_prompt: 'write_prompt',
         critique: 'critique_image',
@@ -52,7 +53,7 @@
         catalog_card: 'catalog_card',
     };
 
-    const WELCOME_HTML = `
+    let WELCOME_HTML = `
         <div class="sa-welcome-title">Assistent · Krea 2</div>
         <ul>
             <li><strong>Generate</strong> слева — живой просмотр. В чат сам не уходит.</li>
@@ -63,7 +64,7 @@
         </ul>
         Напиши, что сгенерировать — или кинь референс и попроси правку.`;
 
-    const HELP_TEXT = `Slash-команды (без LLM):
+    let HELP_TEXT = `Slash-команды (без LLM):
 /help — этот список
 /gen — Generate сейчас
 /look generate|refN — прикрепить окно к vision
@@ -78,7 +79,7 @@
 
 Чипсы над полем ввода делают то же для aspect / seed / vary / Turbo·RAW.`;
 
-    const SLASH_COMMANDS = [
+    let SLASH_COMMANDS = [
         { cmd: '/help', hint: 'список команд' },
         { cmd: '/gen', hint: 'Generate сейчас' },
         { cmd: '/look ', hint: 'generate|refN' },
@@ -97,6 +98,10 @@
     const state = {
         history: [],
         packsLoaded: false,
+        config: null,
+        enabledSkills: [],
+        kreaProfiles: { turbo: { steps: 8, cfg: 1, sigma_shift: 1.15 }, raw: { steps: 28, cfg: 4.5 } },
+        preferredEmbed: null,
         busy: false,
         generating: false,
         chatEpoch: 0,
@@ -1138,11 +1143,23 @@
 
     function onPersonaChanged() {
         const id = $('sa_persona')?.value || 'neutral';
-        const info = (state.personas || []).find((p) => p.id === id);
-        const title = info?.title || id;
         saveSettings();
-        appendSystemNote(`Тон → ${title}`);
-        state.pendingPersonaNote = `Persona is now ${id} (${title}). Adopt this voice from now on.`;
+        loadConfig(id, (data) => {
+            const title = data?.personas?.find((p) => p.id === id)?.title
+                || (state.personas || []).find((p) => p.id === id)?.title
+                || id;
+            if (data?.personas) {
+                state.personas = data.personas;
+            }
+            appendSystemNote(`Тон → ${title}`);
+            state.pendingPersonaNote = `Persona is now ${id} (${title}). Adopt this voice from now on.`;
+            if (data?.assistant?.default_pack && $('sa_pack') && !state.packUserTouched) {
+                const packId = data.assistant.default_pack;
+                if ([...($('sa_pack').options || [])].some((o) => o.value === packId)) {
+                    $('sa_pack').value = packId;
+                }
+            }
+        });
     }
 
     function countPromptImages() {
@@ -2765,6 +2782,10 @@
         if (model) {
             state.preferredModel = model;
         }
+        const embed = localStorage.getItem(LS_EMBED);
+        if (embed) {
+            state.preferredEmbed = embed;
+        }
         if (paneW) {
             document.documentElement.style.setProperty('--sa-image-width', paneW);
         }
@@ -2780,6 +2801,7 @@
     function saveSettings() {
         localStorage.setItem(LS_BASE, $('sa_base_url')?.value || '');
         localStorage.setItem(LS_MODEL, $('sa_model')?.value || '');
+        localStorage.setItem(LS_EMBED, $('sa_embed_model')?.value || state.preferredEmbed || '');
         localStorage.setItem(LS_PACK, $('sa_pack')?.value || 'write_prompt');
         localStorage.setItem(LS_PERSONA, $('sa_persona')?.value || 'neutral');
         localStorage.setItem(LS_VIEW, state.view || 'chat');
@@ -2788,6 +2810,193 @@
         localStorage.setItem(LS_AUTO_GENERATE, $('sa_auto_generate')?.checked ? '1' : '0');
         localStorage.setItem(LS_AUTO_CRITIQUE, $('sa_auto_critique')?.checked ? '1' : '0');
         localStorage.setItem(LS_AUTO_DOWNLOAD, $('sa_auto_download')?.checked ? '1' : '0');
+        persistServerSettings();
+    }
+
+    function persistServerSettings() {
+        if (typeof genericRequest !== 'function') {
+            return;
+        }
+        const skills = {};
+        document.querySelectorAll('#sa_skills_box input[data-skill]')?.forEach((el) => {
+            skills[el.getAttribute('data-skill')] = !!el.checked;
+        });
+        const persona = $('sa_persona')?.value || 'neutral';
+        const settings = {
+            embed_model: $('sa_embed_model')?.value || state.preferredEmbed || '',
+            base_url: $('sa_base_url')?.value || '',
+            [persona]: { skills },
+        };
+        genericRequest('AssistentSaveSettings', { settings }, () => {}, 0, () => {});
+    }
+
+    function applyConfigPayload(data, { applyDefaults = false } = {}) {
+        if (!data || data.error) {
+            return;
+        }
+        state.config = data;
+        if (data.model?.aspect_table && typeof data.model.aspect_table === 'object') {
+            const next = {};
+            for (const [k, v] of Object.entries(data.model.aspect_table)) {
+                if (Array.isArray(v) && v.length >= 2) {
+                    next[k] = [Number(v[0]), Number(v[1])];
+                }
+            }
+            if (Object.keys(next).length) {
+                ASPECT_TABLE = next;
+            }
+        }
+        if (data.model?.profiles) {
+            state.kreaProfiles = data.model.profiles;
+        }
+        if (data.ui?.pack_aliases) {
+            PACK_ALIASES = { ...PACK_ALIASES, ...data.ui.pack_aliases };
+        }
+        if (data.ui?.welcome_html) {
+            WELCOME_HTML = data.ui.welcome_html;
+        }
+        if (data.ui?.help_text) {
+            HELP_TEXT = data.ui.help_text;
+        }
+        if (Array.isArray(data.ui?.slash) && data.ui.slash.length) {
+            SLASH_COMMANDS = data.ui.slash.map((s) => ({
+                cmd: s.cmd || '',
+                hint: s.hint || '',
+                action: s.action || '',
+            }));
+        }
+        state.enabledSkills = Array.isArray(data.enabled_skills) ? data.enabled_skills.slice() : [];
+        if (Array.isArray(data.personas)) {
+            state.personas = data.personas;
+        }
+        renderPersonaOptions(data.personas || [], data.persona || data.default_persona);
+        renderPackOptions(data.packs || [], applyDefaults ? data.assistant?.default_pack : null);
+        renderChips(data.ui?.chips || []);
+        renderSkillChecks(data.skills || [], state.enabledSkills);
+        if (applyDefaults && data.assistant?.default_pack && $('sa_pack') && !localStorage.getItem(LS_PACK)) {
+            $('sa_pack').value = data.assistant.default_pack;
+        }
+        if (data.assistant?.embed_model && !state.preferredEmbed) {
+            state.preferredEmbed = data.assistant.embed_model;
+        }
+    }
+
+    function renderPersonaOptions(personas, selected) {
+        const sel = $('sa_persona');
+        if (!sel) {
+            return;
+        }
+        const cur = selected || sel.value || localStorage.getItem(LS_PERSONA) || 'neutral';
+        sel.innerHTML = '';
+        for (const p of personas) {
+            const opt = document.createElement('option');
+            opt.value = p.id;
+            opt.textContent = p.title || p.id;
+            if (p.accent) {
+                opt.dataset.accent = p.accent;
+            }
+            sel.appendChild(opt);
+        }
+        if ([...sel.options].some((o) => o.value === cur)) {
+            sel.value = cur;
+        }
+    }
+
+    function renderPackOptions(packs, preferred) {
+        const sel = $('sa_pack');
+        if (!sel) {
+            return;
+        }
+        const cur = preferred || sel.value || localStorage.getItem(LS_PACK) || 'write_prompt';
+        sel.innerHTML = '';
+        const list = (packs || []).slice().sort((a, b) => (a.order || 100) - (b.order || 100));
+        for (const p of list) {
+            const opt = document.createElement('option');
+            opt.value = p.id;
+            opt.textContent = p.title || p.id;
+            sel.appendChild(opt);
+        }
+        if ([...sel.options].some((o) => o.value === cur)) {
+            sel.value = cur;
+        }
+    }
+
+    function renderChips(chips) {
+        const box = $('sa_chips');
+        if (!box || !Array.isArray(chips) || !chips.length) {
+            return;
+        }
+        box.innerHTML = '';
+        for (const c of chips) {
+            if (c.sep) {
+                const sep = document.createElement('span');
+                sep.className = 'sa-chip-sep';
+                sep.setAttribute('aria-hidden', 'true');
+                box.appendChild(sep);
+                continue;
+            }
+            const btn = document.createElement('button');
+            btn.type = 'button';
+            btn.className = 'sa-chip';
+            btn.textContent = c.label || c.value || '';
+            if (c.title) {
+                btn.title = c.title;
+            }
+            const action = c.action || '';
+            const value = c.value ?? '';
+            if (action === 'aspect') {
+                btn.setAttribute('data-aspect', value);
+            } else if (action === 'seed') {
+                btn.setAttribute('data-seed', value);
+            } else if (action === 'vary') {
+                btn.setAttribute('data-vary', value || '1');
+            } else if (action === 'krea_profile') {
+                btn.setAttribute('data-krea-profile', value);
+            }
+            box.appendChild(btn);
+        }
+    }
+
+    function renderSkillChecks(skills, enabled) {
+        const box = $('sa_skills_box');
+        if (!box) {
+            return;
+        }
+        const on = new Set(enabled || []);
+        box.innerHTML = '';
+        for (const s of skills || []) {
+            const label = document.createElement('label');
+            label.className = 'sa-check';
+            const input = document.createElement('input');
+            input.type = 'checkbox';
+            input.setAttribute('data-skill', s.id);
+            input.checked = on.has(s.id) || (!enabled?.length && !!s.default);
+            input.addEventListener('change', () => {
+                state.enabledSkills = [...document.querySelectorAll('#sa_skills_box input[data-skill]:checked')].map((el) => el.getAttribute('data-skill'));
+                saveSettings();
+            });
+            label.appendChild(input);
+            label.appendChild(document.createTextNode(` ${s.title || s.id}`));
+            box.appendChild(label);
+        }
+        state.enabledSkills = [...document.querySelectorAll('#sa_skills_box input[data-skill]:checked')].map((el) => el.getAttribute('data-skill'));
+    }
+
+    function loadConfig(persona, done) {
+        if (typeof genericRequest !== 'function') {
+            done?.(null);
+            return;
+        }
+        genericRequest(
+            'AssistentGetConfig',
+            { persona: persona || $('sa_persona')?.value || 'neutral' },
+            (data) => {
+                applyConfigPayload(data, { applyDefaults: true });
+                done?.(data);
+            },
+            0,
+            () => done?.(null),
+        );
     }
 
     function setModelOptions(models, { error } = {}) {
@@ -2825,6 +3034,38 @@
         }
     }
 
+    function setEmbedModelOptions(models) {
+        const sel = $('sa_embed_model');
+        if (!sel) {
+            return;
+        }
+        const names = (models || []).map((n) => String(n || '').trim()).filter(Boolean);
+        sel.innerHTML = '';
+        if (!names.length) {
+            const opt = document.createElement('option');
+            opt.value = state.preferredEmbed || 'nomic-embed-text';
+            opt.textContent = opt.value + ' (ожидается pull)';
+            sel.appendChild(opt);
+            return;
+        }
+        for (const name of names) {
+            const opt = document.createElement('option');
+            opt.value = name;
+            opt.textContent = name;
+            sel.appendChild(opt);
+        }
+        const prefer = state.preferredEmbed || localStorage.getItem(LS_EMBED) || state.config?.assistant?.embed_model;
+        if (prefer && names.includes(prefer)) {
+            sel.value = prefer;
+        } else if (prefer && !names.includes(prefer)) {
+            const opt = document.createElement('option');
+            opt.value = prefer;
+            opt.textContent = prefer;
+            sel.appendChild(opt);
+            sel.value = prefer;
+        }
+    }
+
     function refreshModels() {
         const baseUrl = $('sa_base_url')?.value || 'http://127.0.0.1:11434';
         setStatus('Loading models…');
@@ -2838,12 +3079,14 @@
             { baseUrl },
             (data) => {
                 const models = data.models || [];
+                const memoryModels = data.memory_models || [];
                 setModelOptions(models);
+                setEmbedModelOptions(memoryModels);
                 const prefer = state.preferredModel || localStorage.getItem(LS_MODEL);
                 if (prefer && models.includes(prefer) && $('sa_model')) {
                     $('sa_model').value = prefer;
                 }
-                setStatus(models.length ? `${models.length} models` : 'No Ollama models (gpu-rent: ollama pull)');
+                setStatus(models.length ? `${models.length} chat · ${memoryModels.length} memory` : 'No Ollama models (gpu-rent: ollama pull)');
                 saveSettings();
             },
             0,
@@ -2928,36 +3171,11 @@
     }
 
     function refreshPersonas() {
-        if (typeof genericRequest !== 'function') {
-            return;
-        }
-        genericRequest(
-            'AssistentListPersonas',
-            {},
-            (data) => {
-                const list = data.personas || [];
-                state.personas = list;
-                const sel = $('sa_persona');
-                if (!sel) {
-                    return;
-                }
-                const prefer = localStorage.getItem(LS_PERSONA) || data.default || 'neutral';
-                sel.innerHTML = '';
-                for (const p of list) {
-                    const opt = document.createElement('option');
-                    opt.value = p.id;
-                    opt.textContent = p.title || p.id;
-                    sel.appendChild(opt);
-                }
-                if ([...sel.options].some((o) => o.value === prefer)) {
-                    sel.value = prefer;
-                } else if (data.default) {
-                    sel.value = data.default;
-                }
-            },
-            0,
-            (err) => console.warn('Assistent personas', err),
-        );
+        loadConfig($('sa_persona')?.value || localStorage.getItem(LS_PERSONA) || 'neutral', (data) => {
+            if (data?.personas) {
+                state.personas = data.personas;
+            }
+        });
     }
 
     function prefetchCard(kind, name) {
@@ -4132,6 +4350,8 @@
             includeBase: true,
             messages,
             context_json: JSON.stringify(context),
+            skills: state.enabledSkills || [],
+            embed_model: $('sa_embed_model')?.value || state.preferredEmbed || '',
             raw: {
                 messages,
                 context_json: JSON.stringify(context),
@@ -4139,6 +4359,8 @@
                 persona,
                 base_url: baseUrl,
                 model,
+                skills: state.enabledSkills || [],
+                embed_model: $('sa_embed_model')?.value || state.preferredEmbed || '',
             },
         };
 
@@ -4431,11 +4653,12 @@
         }
         restoreHistory();
         maybeWelcome();
-        refreshModels();
-        refreshPersonas();
-        refreshInventory(() => {
-            renderCardsList();
-            renderLoraChips();
+        loadConfig(localStorage.getItem(LS_PERSONA) || 'neutral', () => {
+            refreshModels();
+            refreshInventory(() => {
+                renderCardsList();
+                renderLoraChips();
+            });
         });
         wireDropZone();
         wireSplitter();
@@ -4537,6 +4760,10 @@
         });
         $('sa_base_url')?.addEventListener('change', saveSettings);
         $('sa_model')?.addEventListener('change', saveSettings);
+        $('sa_embed_model')?.addEventListener('change', () => {
+            state.preferredEmbed = $('sa_embed_model')?.value || '';
+            saveSettings();
+        });
         $('sa_pack')?.addEventListener('change', () => {
             state.packUserTouched = true;
             saveSettings();
@@ -4559,9 +4786,11 @@
             } else if (vary) {
                 await applyQuickPatch({ vary: true, seed: -1, actions: ['generate'] }, 'Vary');
             } else if (profile === 'turbo') {
-                await applyQuickPatch({ steps: 8, cfg: 1, sigma_shift: 1.15, actions: ['generate'] }, 'Turbo 8/1');
+                const p = state.kreaProfiles?.turbo || { steps: 8, cfg: 1, sigma_shift: 1.15 };
+                await applyQuickPatch({ steps: p.steps ?? 8, cfg: p.cfg ?? 1, sigma_shift: p.sigma_shift ?? 1.15, actions: ['generate'] }, 'Turbo');
             } else if (profile === 'raw') {
-                await applyQuickPatch({ steps: 28, cfg: 4.5, actions: ['generate'] }, 'RAW 28/4.5');
+                const p = state.kreaProfiles?.raw || { steps: 28, cfg: 4.5 };
+                await applyQuickPatch({ steps: p.steps ?? 28, cfg: p.cfg ?? 4.5, actions: ['generate'] }, 'RAW');
             }
             renderLoraChips();
         });

AssistentConfig.cs

@@ -0,0 +1,752 @@
+using System;
+using System.Collections.Generic;
+using System.IO;
+using System.Linq;
+using System.Text;
+using System.Text.RegularExpressions;
+using FreneticUtilities.FreneticExtensions;
+using Newtonsoft.Json.Linq;
+using SwarmUI.Utils;
+
+namespace Mrleo1nid.SwarmAssistent;
+
+/// <summary>Loads Config/_base + personas/&lt;id&gt; sparse presets with disk overlay merge.</summary>
+public sealed class AssistentConfig
+{
+    readonly string _bundledRoot;
+    readonly string _overlayRoot;
+    readonly object _lock = new();
+
+    public AssistentConfig(string extensionFilePath, string dataRoot)
+    {
+        _bundledRoot = Path.Combine(extensionFilePath ?? "", "Config");
+        _overlayRoot = Path.Combine(dataRoot ?? "", "Assistent");
+    }
+
+    public string BundledRoot => _bundledRoot;
+    public string OverlayRoot => _overlayRoot;
+
+    public static string SafeId(string id)
+    {
+        string s = (id ?? "").Replace('\\', '/').AfterLast('/').Replace("..", "").Trim();
+        if (string.IsNullOrWhiteSpace(s) || !Regex.IsMatch(s, @"^[A-Za-z0-9][A-Za-z0-9_\-]{0,63}$"))
+        {
+            return null;
+        }
+        return s;
+    }
+
+    static string SafeRel(string relative)
+    {
+        if (string.IsNullOrWhiteSpace(relative))
+        {
+            return null;
+        }
+        string norm = relative.Replace('\\', '/').TrimStart('/');
+        if (norm.Contains("..", StringComparison.Ordinal) || Path.IsPathRooted(relative))
+        {
+            return null;
+        }
+        return norm.Replace('/', Path.DirectorySeparatorChar);
+    }
+
+    public string ResolveUnder(string root, string relative)
+    {
+        string rel = SafeRel(relative);
+        if (rel is null || string.IsNullOrWhiteSpace(root))
+        {
+            return null;
+        }
+        string full = Path.GetFullPath(Path.Combine(root, rel));
+        string rootFull = Path.GetFullPath(root);
+        if (!full.StartsWith(rootFull.TrimEnd(Path.DirectorySeparatorChar) + Path.DirectorySeparatorChar, StringComparison.OrdinalIgnoreCase)
+            && !string.Equals(full, rootFull, StringComparison.OrdinalIgnoreCase))
+        {
+            return null;
+        }
+        return full;
+    }
+
+    public JObject DeepMerge(JObject bottom, JObject top)
+    {
+        if (bottom is null)
+        {
+            return top is null ? new JObject() : (JObject)top.DeepClone();
+        }
+        if (top is null)
+        {
+            return (JObject)bottom.DeepClone();
+        }
+        JObject result = (JObject)bottom.DeepClone();
+        foreach (JProperty prop in top.Properties())
+        {
+            if (prop.Value is JObject topObj && result[prop.Name] is JObject botObj)
+            {
+                result[prop.Name] = DeepMerge(botObj, topObj);
+            }
+            else if (prop.Value is JArray || prop.Value is null || prop.Value.Type == JTokenType.Null)
+            {
+                // Arrays replace entirely when the top file provides them.
+                if (prop.Value is not null && prop.Value.Type != JTokenType.Null)
+                {
+                    result[prop.Name] = prop.Value.DeepClone();
+                }
+            }
+            else if (prop.Value.Type == JTokenType.String && string.IsNullOrWhiteSpace(prop.Value.ToString()))
+            {
+                // Empty string does not clobber (personas.json empty prompt rule).
+                continue;
+            }
+            else
+            {
+                result[prop.Name] = prop.Value.DeepClone();
+            }
+        }
+        return result;
+    }
+
+    public JObject TryReadJson(string path)
+    {
+        if (string.IsNullOrWhiteSpace(path) || !File.Exists(path))
+        {
+            return null;
+        }
+        try
+        {
+            return JObject.Parse(File.ReadAllText(path, Encoding.UTF8));
+        }
+        catch (Exception ex)
+        {
+            Logs.Debug($"AssistentConfig json {path}: {ex.Message}");
+            return null;
+        }
+    }
+
+    public JArray TryReadJsonArray(string path)
+    {
+        if (string.IsNullOrWhiteSpace(path) || !File.Exists(path))
+        {
+            return null;
+        }
+        try
+        {
+            return JArray.Parse(File.ReadAllText(path, Encoding.UTF8));
+        }
+        catch (Exception ex)
+        {
+            Logs.Debug($"AssistentConfig json-array {path}: {ex.Message}");
+            return null;
+        }
+    }
+
+    public string TryReadText(string path)
+    {
+        if (string.IsNullOrWhiteSpace(path) || !File.Exists(path))
+        {
+            return null;
+        }
+        try
+        {
+            string text = File.ReadAllText(path, Encoding.UTF8);
+            return string.IsNullOrWhiteSpace(text) ? null : text;
+        }
+        catch (Exception ex)
+        {
+            Logs.Debug($"AssistentConfig text {path}: {ex.Message}");
+            return null;
+        }
+    }
+
+    /// <summary>Merge layered copies of the same relative path: bundled base → persona chain → disk base → disk persona.</summary>
+    public JObject MergeJsonLayers(string relative, IEnumerable<string> roots)
+    {
+        JObject acc = null;
+        foreach (string root in roots)
+        {
+            string path = ResolveUnder(root, relative);
+            JObject next = TryReadJson(path);
+            if (next is null)
+            {
+                continue;
+            }
+            acc = DeepMerge(acc, next);
+        }
+        return acc ?? new JObject();
+    }
+
+    public string MergeTextLayers(string relative, IEnumerable<string> roots)
+    {
+        string last = null;
+        foreach (string root in roots)
+        {
+            string path = ResolveUnder(root, relative);
+            string text = TryReadText(path);
+            if (text is not null)
+            {
+                last = text;
+            }
+        }
+        return last;
+    }
+
+    List<string> PersonaExtendsChain(string personaId)
+    {
+        List<string> chain = [];
+        HashSet<string> seen = new(StringComparer.OrdinalIgnoreCase);
+        string cur = SafeId(personaId) ?? "neutral";
+        for (int i = 0; i < 8 && !string.IsNullOrWhiteSpace(cur); i++)
+        {
+            if (!seen.Add(cur))
+            {
+                break;
+            }
+            chain.Insert(0, cur);
+            JObject meta = TryReadJson(ResolveUnder(Path.Combine(_bundledRoot, "personas", cur), "persona.json"))
+                ?? TryReadJson(ResolveUnder(Path.Combine(_overlayRoot, "personas", cur), "persona.json"));
+            string parent = SafeId(meta?["extends"]?.ToString());
+            if (string.IsNullOrWhiteSpace(parent) || string.Equals(parent, cur, StringComparison.OrdinalIgnoreCase))
+            {
+                break;
+            }
+            cur = parent;
+        }
+        return chain;
+    }
+
+    public IEnumerable<string> LayerRoots(string personaId)
+    {
+        yield return Path.Combine(_bundledRoot, "_base");
+        foreach (string id in PersonaExtendsChain(personaId))
+        {
+            yield return Path.Combine(_bundledRoot, "personas", id);
+        }
+        yield return Path.Combine(_overlayRoot, "_base");
+        foreach (string id in PersonaExtendsChain(personaId))
+        {
+            yield return Path.Combine(_overlayRoot, "personas", id);
+        }
+    }
+
+    public List<(string id, string title, string accent, string source)> ListPersonaCatalog()
+    {
+        Dictionary<string, (string title, string accent, string source)> byId = new(StringComparer.OrdinalIgnoreCase);
+        void Scan(string root, string source)
+        {
+            string dir = Path.Combine(root, "personas");
+            if (!Directory.Exists(dir))
+            {
+                return;
+            }
+            foreach (string folder in Directory.GetDirectories(dir))
+            {
+                string id = SafeId(Path.GetFileName(folder));
+                if (id is null)
+                {
+                    continue;
+                }
+                JObject meta = TryReadJson(Path.Combine(folder, "persona.json"));
+                if (meta is null)
+                {
+                    continue;
+                }
+                if (meta["enabled"]?.Value<bool?>() == false)
+                {
+                    byId.Remove(id);
+                    continue;
+                }
+                byId[id] = (
+                    meta["title"]?.ToString() ?? id,
+                    meta["accent"]?.ToString() ?? "#8b949e",
+                    source
+                );
+            }
+        }
+        Scan(_bundledRoot, "bundled");
+        Scan(_overlayRoot, "overlay");
+
+        // Legacy personas.json titles
+        string overlayJson = Path.Combine(_overlayRoot, "personas.json");
+        JObject legacy = TryReadJson(overlayJson);
+        if (legacy?["personas"] is JArray arr)
+        {
+            foreach (JToken t in arr)
+            {
+                if (t is not JObject po)
+                {
+                    continue;
+                }
+                string id = SafeId(po["id"]?.ToString());
+                if (id is null)
+                {
+                    continue;
+                }
+                if (byId.TryGetValue(id, out var cur))
+                {
+                    byId[id] = (po["title"]?.ToString() ?? cur.title, cur.accent, "overlay+bundled");
+                }
+                else
+                {
+                    byId[id] = (po["title"]?.ToString() ?? id, "#8b949e", "legacy");
+                }
+            }
+        }
+
+        return byId.OrderBy(kv => kv.Key, StringComparer.OrdinalIgnoreCase)
+            .Select(kv => (kv.Key, kv.Value.title, kv.Value.accent, kv.Value.source))
+            .ToList();
+    }
+
+    public string DefaultPersonaId()
+    {
+        JObject assistant = MergeJsonLayers("assistant.json", LayerRoots("neutral"));
+        string def = SafeId(assistant["default_persona"]?.ToString()) ?? "neutral";
+        string overlayJson = Path.Combine(_overlayRoot, "personas.json");
+        JObject legacy = TryReadJson(overlayJson);
+        string fromLegacy = SafeId(legacy?["default"]?.ToString());
+        if (fromLegacy is not null)
+        {
+            def = fromLegacy;
+        }
+        var catalog = ListPersonaCatalog();
+        if (catalog.All(p => !string.Equals(p.id, def, StringComparison.OrdinalIgnoreCase)) && catalog.Count > 0)
+        {
+            def = catalog[0].id;
+        }
+        return def;
+    }
+
+    public JObject LoadAssistant(string personaId) => MergeJsonLayers("assistant.json", LayerRoots(personaId));
+
+    public JObject LoadUi(string personaId) => MergeJsonLayers("ui.json", LayerRoots(personaId));
+
+    public JObject LoadModelProfile(string personaId)
+    {
+        JObject assistant = LoadAssistant(personaId);
+        string arch = assistant["gate"]?["architecture"]?.ToString() ?? "krea2";
+        string safe = SafeId(arch) ?? "krea2";
+        return MergeJsonLayers(Path.Combine("models", $"{safe}.json"), LayerRoots(personaId));
+    }
+
+    public string LoadCorePrompt(string personaId)
+    {
+        JObject meta = MergeJsonLayers(Path.Combine("core", "core.json"), LayerRoots(personaId));
+        string file = meta["prompt_file"]?.ToString() ?? "core.md";
+        return MergeTextLayers(Path.Combine("core", file), LayerRoots(personaId)) ?? "";
+    }
+
+    public List<(string id, string title, int order, string[] aliases, bool enabled)> ListPacks(string personaId)
+    {
+        Dictionary<string, (string title, int order, string[] aliases, bool enabled)> byId = new(StringComparer.OrdinalIgnoreCase);
+        foreach (string root in LayerRoots(personaId))
+        {
+            string dir = Path.Combine(root, "packs");
+            if (!Directory.Exists(dir))
+            {
+                continue;
+            }
+            foreach (string file in Directory.GetFiles(dir, "*.json"))
+            {
+                JObject meta = TryReadJson(file);
+                string id = SafeId(meta?["id"]?.ToString() ?? Path.GetFileNameWithoutExtension(file));
+                if (id is null || meta is null)
+                {
+                    continue;
+                }
+                bool enabled = meta["enabled"]?.Value<bool?>() != false;
+                string[] aliases = (meta["aliases"] as JArray)?.Select(t => t?.ToString()).Where(s => !string.IsNullOrWhiteSpace(s)).ToArray() ?? [];
+                byId[id] = (
+                    meta["title"]?.ToString() ?? id,
+                    meta["order"]?.Value<int?>() ?? 100,
+                    aliases,
+                    enabled
+                );
+            }
+        }
+        return byId.Where(kv => kv.Value.enabled)
+            .OrderBy(kv => kv.Value.order).ThenBy(kv => kv.Key, StringComparer.OrdinalIgnoreCase)
+            .Select(kv => (kv.Key, kv.Value.title, kv.Value.order, kv.Value.aliases, kv.Value.enabled))
+            .ToList();
+    }
+
+    public string LoadPackPrompt(string personaId, string packId)
+    {
+        string id = SafeId(packId);
+        if (id is null)
+        {
+            return null;
+        }
+        JObject meta = MergeJsonLayers(Path.Combine("packs", $"{id}.json"), LayerRoots(personaId));
+        if (meta["enabled"]?.Value<bool?>() == false)
+        {
+            return null;
+        }
+        string file = meta["prompt_file"]?.ToString() ?? $"{id}.md";
+        return MergeTextLayers(Path.Combine("packs", file), LayerRoots(personaId));
+    }
+
+    public List<(string id, string title, bool defaultOn, bool enabled)> ListSkills(string personaId)
+    {
+        Dictionary<string, (string title, bool defaultOn, bool enabled)> byId = new(StringComparer.OrdinalIgnoreCase);
+        foreach (string root in LayerRoots(personaId))
+        {
+            string dir = Path.Combine(root, "skills");
+            if (!Directory.Exists(dir))
+            {
+                continue;
+            }
+            foreach (string file in Directory.GetFiles(dir, "*.json"))
+            {
+                JObject meta = TryReadJson(file);
+                string id = SafeId(meta?["id"]?.ToString() ?? Path.GetFileNameWithoutExtension(file));
+                if (id is null || meta is null)
+                {
+                    continue;
+                }
+                byId[id] = (
+                    meta["title"]?.ToString() ?? id,
+                    meta["default"]?.Value<bool?>() ?? false,
+                    meta["enabled"]?.Value<bool?>() != false
+                );
+            }
+        }
+        JObject skillsOverride = MergeJsonLayers("skills.json", LayerRoots(personaId));
+        foreach (JProperty prop in skillsOverride.Properties())
+        {
+            string id = SafeId(prop.Name);
+            if (id is null || !byId.ContainsKey(id))
+            {
+                continue;
+            }
+            if (prop.Value.Type == JTokenType.Boolean)
+            {
+                var cur = byId[id];
+                byId[id] = (cur.title, prop.Value.Value<bool>(), cur.enabled);
+            }
+        }
+        return byId.Where(kv => kv.Value.enabled)
+            .OrderBy(kv => kv.Key, StringComparer.OrdinalIgnoreCase)
+            .Select(kv => (kv.Key, kv.Value.title, kv.Value.defaultOn, kv.Value.enabled))
+            .ToList();
+    }
+
+    public string LoadSkillPrompt(string personaId, string skillId)
+    {
+        string id = SafeId(skillId);
+        if (id is null)
+        {
+            return null;
+        }
+        JObject meta = MergeJsonLayers(Path.Combine("skills", $"{id}.json"), LayerRoots(personaId));
+        if (meta["enabled"]?.Value<bool?>() == false)
+        {
+            return null;
+        }
+        string file = meta["prompt_file"]?.ToString() ?? $"{id}.md";
+        return MergeTextLayers(Path.Combine("skills", file), LayerRoots(personaId));
+    }
+
+    public JObject LoadIdentityParts(string personaId)
+    {
+        var roots = LayerRoots(personaId).ToList();
+        JObject persona = MergeJsonLayers("persona.json", roots);
+        JObject voice = MergeJsonLayers("voice.json", roots);
+        JObject likes = MergeJsonLayers("likes.json", roots);
+        JObject dislikes = MergeJsonLayers("dislikes.json", roots);
+        JObject rules = MergeJsonLayers("rules.json", roots);
+        string extra = MergeTextLayers("extra.md", roots);
+
+        // Legacy personas.json prompt → extra overlay
+        string overlayJson = Path.Combine(_overlayRoot, "personas.json");
+        JObject legacy = TryReadJson(overlayJson);
+        if (legacy?["personas"] is JArray arr)
+        {
+            string id = SafeId(personaId) ?? "neutral";
+            foreach (JToken t in arr)
+            {
+                if (t is JObject po && string.Equals(SafeId(po["id"]?.ToString()), id, StringComparison.OrdinalIgnoreCase))
+                {
+                    string title = po["title"]?.ToString();
+                    if (!string.IsNullOrWhiteSpace(title))
+                    {
+                        persona["title"] = title;
+                    }
+                    string prompt = po["prompt"]?.ToString();
+                    if (!string.IsNullOrWhiteSpace(prompt))
+                    {
+                        extra = string.IsNullOrWhiteSpace(extra) ? prompt : extra + "\n\n" + prompt;
+                    }
+                    break;
+                }
+            }
+        }
+
+        return new JObject
+        {
+            ["persona"] = persona,
+            ["voice"] = voice,
+            ["likes"] = likes,
+            ["dislikes"] = dislikes,
+            ["rules"] = rules,
+            ["extra"] = extra ?? "",
+        };
+    }
+
+    static string FormatCategoryMap(JObject obj)
+    {
+        if (obj is null || !obj.Properties().Any())
+        {
+            return "";
+        }
+        List<string> parts = [];
+        foreach (JProperty prop in obj.Properties())
+        {
+            if (prop.Value is JArray arr)
+            {
+                string joined = string.Join(", ", arr.Select(t => t?.ToString()).Where(s => !string.IsNullOrWhiteSpace(s)));
+                if (!string.IsNullOrWhiteSpace(joined))
+                {
+                    parts.Add($"{prop.Name}: {joined}");
+                }
+            }
+            else if (prop.Value?.Type == JTokenType.String)
+            {
+                string s = prop.Value.ToString();
+                if (!string.IsNullOrWhiteSpace(s))
+                {
+                    parts.Add($"{prop.Name}: {s}");
+                }
+            }
+        }
+        return string.Join("; ", parts);
+    }
+
+    public string RenderIdentityBlock(string personaId)
+    {
+        string id = SafeId(personaId) ?? "neutral";
+        JObject parts = LoadIdentityParts(id);
+        JObject persona = parts["persona"] as JObject ?? new JObject();
+        JObject voice = parts["voice"] as JObject ?? new JObject();
+        JObject likes = parts["likes"] as JObject ?? new JObject();
+        JObject dislikes = parts["dislikes"] as JObject ?? new JObject();
+        JObject rules = parts["rules"] as JObject ?? new JObject();
+        string extra = parts["extra"]?.ToString() ?? "";
+
+        string title = persona["title"]?.ToString() ?? id;
+        StringBuilder sb = new();
+        sb.AppendLine($"## Persona: {id} — {title}");
+
+        List<string> voiceBits = [];
+        if (voice["verbosity"] != null)
+        {
+            voiceBits.Add(voice["verbosity"].ToString());
+        }
+        if (voice["tone"] is JArray tones)
+        {
+            voiceBits.AddRange(tones.Select(t => t?.ToString()).Where(s => !string.IsNullOrWhiteSpace(s)));
+        }
+        if (voice["humor"] != null && voice["humor"].ToString() != "none")
+        {
+            voiceBits.Add($"humor:{voice["humor"]}");
+        }
+        if (voice["nsfw"] != null)
+        {
+            voiceBits.Add($"NSFW {voice["nsfw"]}");
+        }
+        if (voice["language"] != null)
+        {
+            voiceBits.Add(voice["language"].ToString());
+        }
+        if (voiceBits.Count > 0)
+        {
+            sb.AppendLine("Voice: " + string.Join(", ", voiceBits));
+        }
+
+        string prefers = FormatCategoryMap(likes);
+        if (!string.IsNullOrWhiteSpace(prefers))
+        {
+            sb.AppendLine("Prefers: " + prefers);
+        }
+        string avoids = FormatCategoryMap(dislikes);
+        if (!string.IsNullOrWhiteSpace(avoids))
+        {
+            sb.AppendLine("Avoids: " + avoids);
+        }
+
+        List<string> ruleBits = [];
+        if (rules["always"] is JArray always)
+        {
+            foreach (string s in always.Select(t => t?.ToString()).Where(x => !string.IsNullOrWhiteSpace(x)))
+            {
+                ruleBits.Add("always " + s);
+            }
+        }
+        if (rules["never"] is JArray never)
+        {
+            foreach (string s in never.Select(t => t?.ToString()).Where(x => !string.IsNullOrWhiteSpace(x)))
+            {
+                ruleBits.Add("never " + s);
+            }
+        }
+        if (ruleBits.Count > 0)
+        {
+            sb.AppendLine("Rules: " + string.Join("; ", ruleBits));
+        }
+        if (!string.IsNullOrWhiteSpace(extra))
+        {
+            sb.AppendLine(extra.Trim());
+        }
+        return sb.ToString().TrimEnd();
+    }
+
+    public List<JObject> LoadMemorySeedDocs()
+    {
+        List<JObject> docs = [];
+        void Scan(string root)
+        {
+            string dir = Path.Combine(root, "memory-seed");
+            if (!Directory.Exists(dir))
+            {
+                return;
+            }
+            foreach (string file in Directory.GetFiles(dir, "*.json").OrderBy(f => f, StringComparer.OrdinalIgnoreCase))
+            {
+                try
+                {
+                    string raw = File.ReadAllText(file, Encoding.UTF8);
+                    JToken parsed = JToken.Parse(raw);
+                    if (parsed is JArray arr)
+                    {
+                        foreach (JToken t in arr)
+                        {
+                            if (t is JObject jo)
+                            {
+                                docs.Add(jo);
+                            }
+                        }
+                    }
+                    else if (parsed is JObject single)
+                    {
+                        docs.Add(single);
+                    }
+                }
+                catch (Exception ex)
+                {
+                    Logs.Debug($"AssistentConfig memory-seed {file}: {ex.Message}");
+                }
+            }
+        }
+        Scan(Path.Combine(_bundledRoot, "_base"));
+        Scan(Path.Combine(_overlayRoot, "_base"));
+        Scan(Path.Combine(_overlayRoot, "memory-seed"));
+        return docs;
+    }
+
+    public JObject LoadSettings()
+    {
+        return TryReadJson(Path.Combine(_overlayRoot, "settings.json")) ?? new JObject();
+    }
+
+    public void SaveSettings(JObject settings)
+    {
+        Directory.CreateDirectory(_overlayRoot);
+        string path = Path.Combine(_overlayRoot, "settings.json");
+        File.WriteAllText(path, (settings ?? new JObject()).ToString(Newtonsoft.Json.Formatting.Indented), Encoding.UTF8);
+    }
+
+    public JObject LoadOllamaRoles()
+    {
+        return TryReadJson(Path.Combine(_overlayRoot, "ollama-roles.json"))
+            ?? new JObject { ["chat"] = new JArray(), ["memory"] = new JArray() };
+    }
+
+    public List<string> ResolveEnabledSkills(string personaId, JArray clientSkills)
+    {
+        var catalog = ListSkills(personaId);
+        HashSet<string> enabled = new(StringComparer.OrdinalIgnoreCase);
+        foreach (var s in catalog.Where(x => x.defaultOn))
+        {
+            enabled.Add(s.id);
+        }
+        JObject settings = LoadSettings();
+        string pid = SafeId(personaId) ?? "neutral";
+        if (settings[pid]?["skills"] is JObject perPersona)
+        {
+            foreach (JProperty prop in perPersona.Properties())
+            {
+                string id = SafeId(prop.Name);
+                if (id is null)
+                {
+                    continue;
+                }
+                if (prop.Value.Type == JTokenType.Boolean)
+                {
+                    if (prop.Value.Value<bool>())
+                    {
+                        enabled.Add(id);
+                    }
+                    else
+                    {
+                        enabled.Remove(id);
+                    }
+                }
+            }
+        }
+        if (clientSkills is not null && clientSkills.Count > 0)
+        {
+            enabled.Clear();
+            foreach (JToken t in clientSkills)
+            {
+                string id = SafeId(t?.ToString());
+                if (id is not null && catalog.Any(c => string.Equals(c.id, id, StringComparison.OrdinalIgnoreCase)))
+                {
+                    enabled.Add(id);
+                }
+            }
+        }
+        return catalog.Select(c => c.id).Where(enabled.Contains).ToList();
+    }
+
+    public JObject BuildMergedConfigPayload(string personaId)
+    {
+        string id = SafeId(personaId) ?? DefaultPersonaId();
+        JObject assistant = LoadAssistant(id);
+        JObject ui = LoadUi(id);
+        JObject model = LoadModelProfile(id);
+        var packs = ListPacks(id);
+        var skills = ListSkills(id);
+        var personas = ListPersonaCatalog();
+        JObject identity = LoadIdentityParts(id);
+        return new JObject
+        {
+            ["success"] = true,
+            ["persona"] = id,
+            ["default_persona"] = DefaultPersonaId(),
+            ["assistant"] = assistant,
+            ["ui"] = ui,
+            ["model"] = model,
+            ["packs"] = new JArray(packs.Select(p => new JObject
+            {
+                ["id"] = p.id,
+                ["title"] = p.title,
+                ["order"] = p.order,
+                ["aliases"] = new JArray(p.aliases),
+            })),
+            ["skills"] = new JArray(skills.Select(s => new JObject
+            {
+                ["id"] = s.id,
+                ["title"] = s.title,
+                ["default"] = s.defaultOn,
+            })),
+            ["personas"] = new JArray(personas.Select(p => new JObject
+            {
+                ["id"] = p.id,
+                ["title"] = p.title,
+                ["accent"] = p.accent,
+                ["source"] = p.source,
+            })),
+            ["identity"] = identity,
+            ["identity_summary"] = RenderIdentityBlock(id),
+            ["enabled_skills"] = new JArray(ResolveEnabledSkills(id, null)),
+        };
+    }
+}

AssistentMemory.cs

@@ -0,0 +1,428 @@
+using System;
+using System.Collections.Generic;
+using System.IO;
+using System.Linq;
+using System.Net.Http;
+using System.Text;
+using System.Threading.Tasks;
+using Microsoft.Data.Sqlite;
+using Newtonsoft.Json.Linq;
+using SwarmUI.Utils;
+
+namespace Mrleo1nid.SwarmAssistent;
+
+/// <summary>Local SQLite vector memory with Ollama /api/embed.</summary>
+public sealed class AssistentMemory : IDisposable
+{
+    readonly string _dbPath;
+    readonly HttpClient _http;
+    readonly object _lock = new();
+    SqliteConnection _conn;
+    string _embedModel;
+    int _dims;
+    int _seedVersion;
+
+    public AssistentMemory(string dataRoot, HttpClient http, string defaultEmbedModel = "nomic-embed-text")
+    {
+        string dir = Path.Combine(dataRoot ?? ".", "Assistent", "memory");
+        Directory.CreateDirectory(dir);
+        _dbPath = Path.Combine(dir, "assistent.sqlite");
+        _http = http;
+        _embedModel = string.IsNullOrWhiteSpace(defaultEmbedModel) ? "nomic-embed-text" : defaultEmbedModel.Trim();
+    }
+
+    public string EmbedModel => _embedModel;
+    public int Dims => _dims;
+    public int SeedVersion => _seedVersion;
+
+    void EnsureOpen()
+    {
+        if (_conn is not null)
+        {
+            return;
+        }
+        _conn = new SqliteConnection($"Data Source={_dbPath}");
+        _conn.Open();
+        using (SqliteCommand cmd = _conn.CreateCommand())
+        {
+            cmd.CommandText =
+                """
+                CREATE TABLE IF NOT EXISTS meta (
+                  key TEXT PRIMARY KEY,
+                  value TEXT NOT NULL
+                );
+                CREATE TABLE IF NOT EXISTS memories (
+                  id INTEGER PRIMARY KEY AUTOINCREMENT,
+                  kind TEXT NOT NULL,
+                  key TEXT NOT NULL,
+                  text TEXT NOT NULL,
+                  source TEXT NOT NULL DEFAULT 'user',
+                  meta_json TEXT,
+                  embedding BLOB,
+                  updated INTEGER NOT NULL,
+                  UNIQUE(kind, key, source)
+                );
+                CREATE INDEX IF NOT EXISTS idx_memories_kind ON memories(kind);
+                """;
+            cmd.ExecuteNonQuery();
+        }
+        _embedModel = GetMeta("embed_model") ?? _embedModel;
+        _ = int.TryParse(GetMeta("dims"), out _dims);
+        _ = int.TryParse(GetMeta("seed_version"), out _seedVersion);
+    }
+
+    string GetMeta(string key)
+    {
+        using SqliteCommand cmd = _conn.CreateCommand();
+        cmd.CommandText = "SELECT value FROM meta WHERE key = $k";
+        cmd.Parameters.AddWithValue("$k", key);
+        return cmd.ExecuteScalar()?.ToString();
+    }
+
+    void SetMeta(string key, string value)
+    {
+        using SqliteCommand cmd = _conn.CreateCommand();
+        cmd.CommandText = "INSERT INTO meta(key, value) VALUES($k, $v) ON CONFLICT(key) DO UPDATE SET value = excluded.value";
+        cmd.Parameters.AddWithValue("$k", key);
+        cmd.Parameters.AddWithValue("$v", value ?? "");
+        cmd.ExecuteNonQuery();
+    }
+
+    static byte[] FloatsToBytes(float[] v)
+    {
+        byte[] bytes = new byte[v.Length * 4];
+        Buffer.BlockCopy(v, 0, bytes, 0, bytes.Length);
+        return bytes;
+    }
+
+    static float[] BytesToFloats(byte[] bytes)
+    {
+        if (bytes is null || bytes.Length < 4 || bytes.Length % 4 != 0)
+        {
+            return Array.Empty<float>();
+        }
+        float[] v = new float[bytes.Length / 4];
+        Buffer.BlockCopy(bytes, 0, v, 0, bytes.Length);
+        return v;
+    }
+
+    static float Cosine(float[] a, float[] b)
+    {
+        if (a.Length == 0 || a.Length != b.Length)
+        {
+            return float.NegativeInfinity;
+        }
+        double dot = 0, na = 0, nb = 0;
+        for (int i = 0; i < a.Length; i++)
+        {
+            dot += a[i] * b[i];
+            na += a[i] * a[i];
+            nb += b[i] * b[i];
+        }
+        if (na <= 0 || nb <= 0)
+        {
+            return float.NegativeInfinity;
+        }
+        return (float)(dot / (Math.Sqrt(na) * Math.Sqrt(nb)));
+    }
+
+    public async Task<float[]> EmbedAsync(string baseUrl, string model, string text, string keepAlive = "60m")
+    {
+        string root = (baseUrl ?? "http://127.0.0.1:11434").TrimEnd('/');
+        string m = string.IsNullOrWhiteSpace(model) ? _embedModel : model.Trim();
+        JObject payload = new()
+        {
+            ["model"] = m,
+            ["input"] = text ?? "",
+            ["keep_alive"] = keepAlive,
+        };
+        using StringContent content = new(payload.ToString(Newtonsoft.Json.Formatting.None), Encoding.UTF8, "application/json");
+        using HttpResponseMessage resp = await _http.PostAsync($"{root}/api/embed", content);
+        string body = await resp.Content.ReadAsStringAsync();
+        if (!resp.IsSuccessStatusCode)
+        {
+            throw new Exception($"Ollama /api/embed HTTP {(int)resp.StatusCode}: {body[..Math.Min(body.Length, 200)]}");
+        }
+        JObject parsed = JObject.Parse(body);
+        JArray embeddings = parsed["embeddings"] as JArray;
+        JToken first = embeddings?.FirstOrDefault() ?? parsed["embedding"];
+        if (first is not JArray vec)
+        {
+            throw new Exception("Ollama /api/embed: no embeddings in response");
+        }
+        float[] floats = vec.Select(t => t.Value<float>()).ToArray();
+        return floats;
+    }
+
+    public async Task EnsureSeedAsync(string baseUrl, AssistentConfig config, string modelOverride = null)
+    {
+        lock (_lock)
+        {
+            EnsureOpen();
+        }
+        JObject assistant = config.LoadAssistant(config.DefaultPersonaId());
+        int wantVersion = assistant["seed_version"]?.Value<int?>() ?? 1;
+        string wantModel = string.IsNullOrWhiteSpace(modelOverride)
+            ? (assistant["embed_model"]?.ToString() ?? _embedModel)
+            : modelOverride.Trim();
+
+        bool needReseed = _seedVersion != wantVersion || !string.Equals(_embedModel, wantModel, StringComparison.OrdinalIgnoreCase);
+        if (!needReseed)
+        {
+            int bundledCount;
+            lock (_lock)
+            {
+                using SqliteCommand cmd = _conn.CreateCommand();
+                cmd.CommandText = "SELECT COUNT(*) FROM memories WHERE source = 'bundled'";
+                bundledCount = Convert.ToInt32(cmd.ExecuteScalar());
+            }
+            if (bundledCount > 0)
+            {
+                return;
+            }
+        }
+
+        List<JObject> docs = config.LoadMemorySeedDocs();
+        if (docs.Count == 0)
+        {
+            return;
+        }
+
+        // Probe embed
+        float[] probe;
+        try
+        {
+            probe = await EmbedAsync(baseUrl, wantModel, docs[0]["text"]?.ToString() ?? "seed");
+        }
+        catch (Exception ex)
+        {
+            Logs.Debug($"AssistentMemory seed defer (embed unavailable): {ex.Message}");
+            return;
+        }
+
+        lock (_lock)
+        {
+            EnsureOpen();
+            if (!string.Equals(_embedModel, wantModel, StringComparison.OrdinalIgnoreCase) || (_dims > 0 && _dims != probe.Length))
+            {
+                using SqliteCommand clear = _conn.CreateCommand();
+                clear.CommandText = "DELETE FROM memories";
+                clear.ExecuteNonQuery();
+            }
+            else if (needReseed)
+            {
+                using SqliteCommand clearBundled = _conn.CreateCommand();
+                clearBundled.CommandText = "DELETE FROM memories WHERE source = 'bundled'";
+                clearBundled.ExecuteNonQuery();
+            }
+            _embedModel = wantModel;
+            _dims = probe.Length;
+            _seedVersion = wantVersion;
+            SetMeta("embed_model", _embedModel);
+            SetMeta("dims", _dims.ToString());
+            SetMeta("seed_version", _seedVersion.ToString());
+        }
+
+        foreach (JObject doc in docs)
+        {
+            string kind = (doc["kind"]?.ToString() ?? "note").Trim();
+            string key = (doc["key"]?.ToString() ?? "").Trim();
+            string text = (doc["text"]?.ToString() ?? "").Trim();
+            if (string.IsNullOrWhiteSpace(key) || string.IsNullOrWhiteSpace(text))
+            {
+                continue;
+            }
+            try
+            {
+                float[] vec = await EmbedAsync(baseUrl, wantModel, text);
+                Upsert(kind, key, text, "bundled", doc["tags"], vec);
+            }
+            catch (Exception ex)
+            {
+                Logs.Debug($"AssistentMemory seed item {kind}/{key}: {ex.Message}");
+            }
+        }
+    }
+
+    public void Upsert(string kind, string key, string text, string source, JToken meta, float[] embedding)
+    {
+        lock (_lock)
+        {
+            EnsureOpen();
+            kind = (kind ?? "note").Trim().ToLowerInvariant();
+            key = (key ?? "").Trim();
+            text = (text ?? "").Trim();
+            source = string.IsNullOrWhiteSpace(source) ? "user" : source.Trim();
+            if (string.IsNullOrWhiteSpace(key) || string.IsNullOrWhiteSpace(text))
+            {
+                return;
+            }
+            if (embedding is { Length: > 0 })
+            {
+                if (_dims <= 0)
+                {
+                    _dims = embedding.Length;
+                    SetMeta("dims", _dims.ToString());
+                }
+            }
+            using SqliteCommand cmd = _conn.CreateCommand();
+            cmd.CommandText =
+                """
+                INSERT INTO memories(kind, key, text, source, meta_json, embedding, updated)
+                VALUES($kind, $key, $text, $source, $meta, $emb, $upd)
+                ON CONFLICT(kind, key, source) DO UPDATE SET
+                  text = excluded.text,
+                  meta_json = excluded.meta_json,
+                  embedding = excluded.embedding,
+                  updated = excluded.updated
+                """;
+            cmd.Parameters.AddWithValue("$kind", kind);
+            cmd.Parameters.AddWithValue("$key", key);
+            cmd.Parameters.AddWithValue("$text", text);
+            cmd.Parameters.AddWithValue("$source", source);
+            cmd.Parameters.AddWithValue("$meta", meta?.ToString(Newtonsoft.Json.Formatting.None) ?? "");
+            cmd.Parameters.AddWithValue("$emb", embedding is null ? (object)DBNull.Value : FloatsToBytes(embedding));
+            cmd.Parameters.AddWithValue("$upd", DateTimeOffset.UtcNow.ToUnixTimeSeconds());
+            cmd.ExecuteNonQuery();
+        }
+    }
+
+    public void Forget(string kind, string key, string source = null)
+    {
+        lock (_lock)
+        {
+            EnsureOpen();
+            using SqliteCommand cmd = _conn.CreateCommand();
+            if (string.IsNullOrWhiteSpace(source))
+            {
+                cmd.CommandText = "DELETE FROM memories WHERE kind = $kind AND key = $key AND source != 'bundled'";
+            }
+            else
+            {
+                cmd.CommandText = "DELETE FROM memories WHERE kind = $kind AND key = $key AND source = $source";
+                cmd.Parameters.AddWithValue("$source", source);
+            }
+            cmd.Parameters.AddWithValue("$kind", (kind ?? "").Trim().ToLowerInvariant());
+            cmd.Parameters.AddWithValue("$key", (key ?? "").Trim());
+            cmd.ExecuteNonQuery();
+        }
+    }
+
+    public async Task<JArray> RetrieveAsync(string baseUrl, string query, int topK = 10, string modelOverride = null)
+    {
+        if (string.IsNullOrWhiteSpace(query))
+        {
+            return [];
+        }
+        lock (_lock)
+        {
+            EnsureOpen();
+        }
+        string model = string.IsNullOrWhiteSpace(modelOverride) ? _embedModel : modelOverride;
+        float[] q;
+        try
+        {
+            q = await EmbedAsync(baseUrl, model, query);
+        }
+        catch (Exception ex)
+        {
+            Logs.Debug($"AssistentMemory retrieve embed: {ex.Message}");
+            return [];
+        }
+
+        List<(float score, JObject row)> scored = [];
+        lock (_lock)
+        {
+            EnsureOpen();
+            using SqliteCommand cmd = _conn.CreateCommand();
+            cmd.CommandText = "SELECT kind, key, text, source, meta_json, embedding FROM memories WHERE embedding IS NOT NULL";
+            using SqliteDataReader reader = cmd.ExecuteReader();
+            while (reader.Read())
+            {
+                float[] emb = BytesToFloats(reader.IsDBNull(5) ? null : (byte[])reader.GetValue(5));
+                float score = Cosine(q, emb);
+                if (float.IsNegativeInfinity(score))
+                {
+                    continue;
+                }
+                scored.Add((score, new JObject
+                {
+                    ["kind"] = reader.GetString(0),
+                    ["key"] = reader.GetString(1),
+                    ["text"] = reader.GetString(2),
+                    ["source"] = reader.GetString(3),
+                    ["score"] = Math.Round(score, 4),
+                }));
+            }
+        }
+        return new JArray(scored.OrderByDescending(s => s.score).Take(Math.Clamp(topK, 1, 30)).Select(s => s.row));
+    }
+
+    public async Task UpsertTextAsync(string baseUrl, string kind, string key, string text, string source = "user", JToken meta = null, string modelOverride = null)
+    {
+        string model = string.IsNullOrWhiteSpace(modelOverride) ? _embedModel : modelOverride;
+        float[] vec = await EmbedAsync(baseUrl, model, text);
+        Upsert(kind, key, text, source, meta, vec);
+    }
+
+    public async Task ReembedAllAsync(string baseUrl, string newModel)
+    {
+        List<(string kind, string key, string text, string source, string meta)> rows = [];
+        lock (_lock)
+        {
+            EnsureOpen();
+            using SqliteCommand cmd = _conn.CreateCommand();
+            cmd.CommandText = "SELECT kind, key, text, source, meta_json FROM memories";
+            using SqliteDataReader reader = cmd.ExecuteReader();
+            while (reader.Read())
+            {
+                rows.Add((reader.GetString(0), reader.GetString(1), reader.GetString(2), reader.GetString(3), reader.IsDBNull(4) ? "" : reader.GetString(4)));
+            }
+        }
+        if (rows.Count == 0)
+        {
+            _embedModel = newModel;
+            lock (_lock)
+            {
+                EnsureOpen();
+                SetMeta("embed_model", _embedModel);
+            }
+            return;
+        }
+        float[] first = await EmbedAsync(baseUrl, newModel, rows[0].text);
+        lock (_lock)
+        {
+            EnsureOpen();
+            _embedModel = newModel;
+            _dims = first.Length;
+            SetMeta("embed_model", _embedModel);
+            SetMeta("dims", _dims.ToString());
+        }
+        foreach (var row in rows)
+        {
+            try
+            {
+                float[] vec = await EmbedAsync(baseUrl, newModel, row.text);
+                JToken meta = null;
+                if (!string.IsNullOrWhiteSpace(row.meta))
+                {
+                    try { meta = JToken.Parse(row.meta); } catch { /* ignore */ }
+                }
+                Upsert(row.kind, row.key, row.text, row.source, meta, vec);
+            }
+            catch (Exception ex)
+            {
+                Logs.Debug($"AssistentMemory reembed {row.kind}/{row.key}: {ex.Message}");
+            }
+        }
+    }
+
+    public void Dispose()
+    {
+        lock (_lock)
+        {
+            _conn?.Dispose();
+            _conn = null;
+        }
+    }
+}

Config/_base/assistant.json

@@ -0,0 +1,20 @@
+{
+  "num_ctx": 16384,
+  "max_civitai_hops": 2,
+  "max_loras_inventory": 150,
+  "max_checkpoints_inventory": 60,
+  "max_wildcards_inventory": 80,
+  "inventory_blurb_max": 140,
+  "max_ref_slots": 4,
+  "default_pack": "write_prompt",
+  "default_persona": "neutral",
+  "embed_model": "nomic-embed-text",
+  "memory_top_k": 10,
+  "seed_version": 1,
+  "gate": {
+    "architecture": "krea2",
+    "keywords": ["krea"]
+  },
+  "context_prompt_max": 2000,
+  "history_keep_turns": 4
+}

Config/_base/core/core.json

@@ -0,0 +1,5 @@
+{
+  "id": "core",
+  "title": "Core contract",
+  "prompt_file": "core.md"
+}

Config/_base/core/core.md

@@ -0,0 +1,84 @@
+# Swarm Assistent — core contract
+
+You are **Swarm Assistent**, a collaborative art director for image generation inside SwarmUI.
+
+## Live context
+
+A JSON block named "Live SwarmUI context" is attached. Treat it as ground truth — refreshed every chat turn:
+
+- Use only LoRAs listed in `available_loras` / `enabled_loras` (exact `name`), or Civitai search candidates.
+- Prefer listed `trigger_phrase` / `triggers` — **never invent** trigger words.
+- `memory_hits` are retrieved facts (model knowledge, LoRA notes, pitfalls). Trust them over guesses.
+- `model_cards` for **enabled** models beat generic blurbs — follow `when` / `avoid` / `prompt_hint` / `triggers`.
+- `taste_profile` is the user's remembered preferences — bias toward it unless they override.
+- Prefer `krea_likely` / Krea architecture entries; ignore FLUX/SDXL LoRAs.
+- Respect current width/height/steps/cfg/seed/sigma_shift/sampler unless the user asks or the pack is `fix_params`.
+- `wildcards` — `__name__` syntax. `prompt_image_count` > 0 means Prompt Images may dominate text.
+- **Init / inpaint:** `has_init_image`, `has_mask_image`, `init_creativity` (denoise 0–1), `mask_blur`, `mask_grow`.
+- **Board:** `image_slots`. `generate` = live gen. `ref1`… = refs. Emit `look_at` to see an unattached window.
+
+## Output contract (mandatory)
+
+1. Write a short helpful reply in the user's language (RU or EN).
+2. Then emit **one** fenced JSON patch (only fields you want to change):
+
+```json
+{
+  "prompt": "...",
+  "negative": null,
+  "loras": [{"name": "exact_name_from_list", "weight": 0.8, "triggers": ["..."]}],
+  "aspect": "16:9",
+  "width": 1376,
+  "height": 768,
+  "steps": 8,
+  "cfg": 1,
+  "seed": -1,
+  "images": 1,
+  "sigma_shift": 1.15,
+  "sampler": null,
+  "creativity": "medium",
+  "intensity": 0,
+  "complexity": 0,
+  "movement": 0,
+  "vary": false,
+  "lock_seed": false,
+  "use_init_image": false,
+  "clear_init_image": false,
+  "init_creativity": 0.45,
+  "use_mask_image": false,
+  "clear_mask_image": false,
+  "mask_blur": null,
+  "mask_grow": null,
+  "clear_prompt_images": false,
+  "slot_to_prompt_image": null,
+  "look_at": ["generate"],
+  "slot_to_init": null,
+  "slot_to_mask": null,
+  "snapshot_generate": false,
+  "select_slot": null,
+  "pack": null,
+  "actions": ["generate"],
+  "search_query": null,
+  "memories": [{"kind": "lora", "key": "name", "text": "fact"}],
+  "notes": "one-line why"
+}
+```
+
+### Patch rules
+
+- Omit keys you are not changing.
+- `loras` replaces the intended LoRA set for Apply (list all that should be on).
+- Prefer `aspect` over raw width/height when framing changes.
+- `vary: true` — new random seed. `lock_seed: true` — reuse current seed.
+- `pack` — switch active prompt pack for a follow-up hop.
+- Do not invent model or LoRA filenames.
+- Memory: `actions` may include `memory_upsert` or `memory_forget` with `memories: [{kind,key,text}]`.
+
+### Actions (auto-safe)
+
+- `"generate"` — after Apply, start generation.
+- `"search_civitai"` — Civitai search; user Confirms downloads.
+- `"interrupt"` — stop generation.
+- `"memory_upsert"` / `"memory_forget"` — write or delete facts in vector memory.
+- `look_at: ["generate", "ref1"]` — vision hop.
+- Pure Q&A with no change: omit the JSON patch.

Config/_base/dislikes.json

@@ -0,0 +1,11 @@
+{
+  "categories": [],
+  "styles": ["tag-soup", "danbooru", "quality-spam", "3d-render-as-photo"],
+  "subjects": [],
+  "aspects": [],
+  "lighting": [],
+  "camera": [],
+  "loras": ["flux", "sdxl"],
+  "moods": [],
+  "notes": ["invented LoRA names", "invented triggers", "moral lectures"]
+}

Config/_base/likes.json

@@ -0,0 +1,11 @@
+{
+  "categories": ["general"],
+  "styles": ["natural prose", "photograph"],
+  "subjects": [],
+  "aspects": ["1:1", "4:5", "16:9"],
+  "lighting": ["clear key light"],
+  "camera": [],
+  "loras": [],
+  "moods": [],
+  "notes": []
+}

Config/_base/memory-seed/aspect.json

@@ -0,0 +1,50 @@
+[
+  {
+    "kind": "aspect",
+    "key": "1:1",
+    "tags": ["aspect", "1k"],
+    "text": "Aspect 1:1 maps to 1024×1024 on the official Krea 1K table. Prefer patch field aspect over raw width/height."
+  },
+  {
+    "kind": "aspect",
+    "key": "4:3",
+    "tags": ["aspect", "1k"],
+    "text": "Aspect 4:3 maps to 1184×896."
+  },
+  {
+    "kind": "aspect",
+    "key": "3:2",
+    "tags": ["aspect", "1k"],
+    "text": "Aspect 3:2 maps to 1248×832."
+  },
+  {
+    "kind": "aspect",
+    "key": "16:9",
+    "tags": ["aspect", "1k", "widescreen"],
+    "text": "Aspect 16:9 maps to 1376×768."
+  },
+  {
+    "kind": "aspect",
+    "key": "2.35:1",
+    "tags": ["aspect", "1k", "cinematic"],
+    "text": "Aspect 2.35:1 (cinematic ultrawide) maps to 1568×672."
+  },
+  {
+    "kind": "aspect",
+    "key": "4:5",
+    "tags": ["aspect", "1k", "portrait"],
+    "text": "Aspect 4:5 maps to 928×1152 — good for portrait."
+  },
+  {
+    "kind": "aspect",
+    "key": "2:3",
+    "tags": ["aspect", "1k", "portrait"],
+    "text": "Aspect 2:3 maps to 832×1248."
+  },
+  {
+    "kind": "aspect",
+    "key": "9:16",
+    "tags": ["aspect", "1k", "stories"],
+    "text": "Aspect 9:16 maps to 768×1376 — vertical / stories."
+  }
+]

Config/_base/memory-seed/krea_facts.json

@@ -0,0 +1,32 @@
+[
+  {
+    "kind": "model",
+    "key": "krea2_architecture",
+    "tags": ["krea", "architecture"],
+    "text": "Krea 2 is a 12B DiT architecture. Not FLUX, not SDXL, not FLUX.1-Krea. Text encoder: Qwen3-VL 4B. VAE: Qwen Image VAE. Use only Krea2-trained LoRAs — never suggest FLUX/SDXL LoRAs."
+  },
+  {
+    "kind": "model",
+    "key": "krea2_turbo",
+    "tags": ["krea", "turbo", "params"],
+    "text": "Krea 2 Turbo defaults: steps 8 (min 4), CFG 1 (never CFG 0 — broken output), sigma shift 1.15, side ~1024 (128–4096 OK)."
+  },
+  {
+    "kind": "model",
+    "key": "krea2_raw",
+    "tags": ["krea", "raw", "params"],
+    "text": "Krea 2 RAW/Base: steps ~20–52, CFG ~4–4.5. If checkpoint name/title looks like RAW (not Turbo), prefer RAW settings. If a turbo LoRA exists, weight ~0.6 for photoreal (1.0 ≈ full turbo). Swarm Generate cannot run dual-sampler Comfy graphs — only suggest LoRA weight + steps/CFG the UI can set."
+  },
+  {
+    "kind": "model",
+    "key": "krea2_negatives",
+    "tags": ["krea", "prompting"],
+    "text": "Negative prompts are nearly useless with Qwen3-VL. Prefer positives (sharp focus, empty street) over no blur / no people. Built-in NSFW text-refiner may strip risque words; LoRAs/finetunes may restore — stay practical."
+  },
+  {
+    "kind": "model",
+    "key": "krea2_prompt_images",
+    "tags": ["krea", "board"],
+    "text": "Prompt Images (refs in the prompt box) often overpower text — use sparingly and warn. Init Image = structure (img2img). Mask = local fix. They are not interchangeable. Cloud-only features (moodboards, Generative Sliders) are not in Swarm — emulate with prompt language + board refs."
+  }
+]

Config/_base/memory-seed/pitfalls.json

@@ -0,0 +1,20 @@
+[
+  {
+    "kind": "pitfall",
+    "key": "dead_eyes",
+    "tags": ["pitfall", "face", "lora"],
+    "text": "Dead eyes / weak emotion on Krea 2: prefer an expressiveness/bypass LoRA from inventory if present; describe eyes and expression vividly in prose."
+  },
+  {
+    "kind": "pitfall",
+    "key": "3d_bias",
+    "tags": ["pitfall", "photo"],
+    "text": "3D / concept-art bias when the user wanted a photo: say photograph, real skin texture, film grain, camera/lens — not only photorealistic."
+  },
+  {
+    "kind": "pitfall",
+    "key": "vae_halftone",
+    "tags": ["pitfall", "inpaint", "vae"],
+    "text": "Qwen VAE halftone/grid on sand, hair, fine weave: prefer inpaint that region at low denoise (~0.2–0.35) — do not rewrite the whole scene prompt."
+  }
+]

Config/_base/models/krea2.json

@@ -0,0 +1,26 @@
+{
+  "id": "krea2",
+  "gate_keywords": ["krea"],
+  "profiles": {
+    "turbo": {
+      "steps": 8,
+      "cfg": 1,
+      "sigma_shift": 1.15
+    },
+    "raw": {
+      "steps": 28,
+      "cfg": 4.5,
+      "sigma_shift": 1.15
+    }
+  },
+  "aspect_table": {
+    "1:1": [1024, 1024],
+    "4:3": [1184, 896],
+    "3:2": [1248, 832],
+    "16:9": [1376, 768],
+    "2.35:1": [1568, 672],
+    "4:5": [928, 1152],
+    "2:3": [832, 1248],
+    "9:16": [768, 1376]
+  }
+}

Config/_base/packs/catalog_card.json

@@ -0,0 +1,7 @@
+{
+  "id": "catalog_card",
+  "title": "Карточка модели",
+  "order": 70,
+  "aliases": ["card", "catalog"],
+  "prompt_file": "catalog_card.md"
+}

Config/_base/packs/compose_scene.json

@@ -0,0 +1,7 @@
+{
+  "id": "compose_scene",
+  "title": "Собрать сцену",
+  "order": 30,
+  "aliases": ["compose"],
+  "prompt_file": "compose_scene.md"
+}

Config/_base/packs/critique_image.json

@@ -0,0 +1,7 @@
+{
+  "id": "critique_image",
+  "title": "Критика кадра",
+  "order": 20,
+  "aliases": ["critique"],
+  "prompt_file": "critique_image.md"
+}

Config/_base/packs/describe_ref.json

@@ -0,0 +1,7 @@
+{
+  "id": "describe_ref",
+  "title": "Описать ref",
+  "order": 60,
+  "aliases": ["describe"],
+  "prompt_file": "describe_ref.md"
+}

Config/_base/packs/fix_params.json

@@ -0,0 +1,7 @@
+{
+  "id": "fix_params",
+  "title": "Параметры",
+  "order": 40,
+  "aliases": ["params"],
+  "prompt_file": "fix_params.md"
+}

Config/_base/packs/inpaint_edit.json

@@ -0,0 +1,7 @@
+{
+  "id": "inpaint_edit",
+  "title": "Inpaint / img2img",
+  "order": 50,
+  "aliases": ["inpaint"],
+  "prompt_file": "inpaint_edit.md"
+}

Config/_base/packs/write_prompt.json

@@ -0,0 +1,7 @@
+{
+  "id": "write_prompt",
+  "title": "Написать промпт",
+  "order": 10,
+  "aliases": ["write"],
+  "prompt_file": "write_prompt.md"
+}

Config/_base/persona.json

@@ -0,0 +1,5 @@
+{
+  "title": "Base",
+  "tagline": "Default identity template",
+  "accent": "#8b949e"
+}

Config/_base/rules.json

@@ -0,0 +1,12 @@
+{
+  "always": [
+    "Match the user's language (RU or EN)",
+    "Use only inventory / memory_hits / cards for LoRA names and triggers",
+    "Emit valid JSON patches when changing generation state"
+  ],
+  "never": [
+    "Invent LoRA filenames or trigger words",
+    "Lecture or moralize about NSFW",
+    "Dump all installed LoRAs — use memory_hits and enabled ones"
+  ]
+}

Config/_base/skills/creativity_sliders.json

@@ -0,0 +1,6 @@
+{
+  "id": "creativity_sliders",
+  "title": "Creativity lexicon",
+  "default": true,
+  "prompt_file": "creativity_sliders.md"
+}

Config/_base/skills/creativity_sliders.md

@@ -0,0 +1,4 @@
+# Skill: creativity & sliders (LLM-only)
+
+- `creativity`: `raw` | `low` | `medium` | `high` — how much **you** expand the user's wording into the prompt. Not a SwarmUI field.
+- Optional `intensity` / `complexity` / `movement` (−100..100): weave into prompt lexicon (muted↔stylized, minimal↔dense, static↔kinetic camera). Do not invent UI sliders.

Config/_base/skills/memory.json

@@ -0,0 +1,6 @@
+{
+  "id": "memory",
+  "title": "Vector memory",
+  "default": true,
+  "prompt_file": "memory.md"
+}

Config/_base/skills/memory.md

@@ -0,0 +1,16 @@
+# Skill: memory
+
+You have a persistent vector memory (`memory_hits` in live context).
+
+## When to write
+
+- Durable facts about a LoRA/checkpoint (when it works, what it breaks, good weight).
+- Bad paths / pitfalls you discovered this session.
+- Prefer `actions: ["memory_upsert"]` + `memories: [{ "kind": "lora"|"pitfall"|"path"|"note", "key": "stable-id", "text": "…" }]`.
+
+## When not to write
+
+- Do not dump the full inventory — retrieve already surfaces relevant blurbs.
+- Do not store the user's taste profile (that is `taste_profile` / taste.json).
+- Do not upsert trivia that is already in `memory_hits` with the same meaning.
+- `memory_forget` only when a fact is wrong or obsolete.

Config/_base/skills/prompting.json

@@ -0,0 +1,6 @@
+{
+  "id": "prompting",
+  "title": "Prompt craft",
+  "default": true,
+  "prompt_file": "prompting.md"
+}

Config/_base/skills/prompting.md

@@ -0,0 +1,9 @@
+# Skill: prompting
+
+Write **natural prose** for a photographer/director — not Danbooru tags, not `(word:1.5)`, not `masterpiece / best quality / 8k`.
+
+Order (front-load importance): **subject → pose/action → setting → materials → camera/framing → lighting → medium/mood**.
+
+- Short user ideas: expand. Finished Krea-style paragraphs: keep wording; only fix anti-patterns.
+- Put LoRA trigger phrases near the subject they affect.
+- Prefer positives over negatives.

Config/_base/ui.json

@@ -0,0 +1,50 @@
+{
+  "welcome_html": "<div class=\"sa-welcome-title\">Assistent · Krea 2</div><ul><li><strong>Generate</strong> слева — живой просмотр. В чат сам не уходит.</li><li><strong>Refs</strong> — референсы на отдельной вкладке: drop / paste / Снимок gen.</li><li>Галочка vision на окне — отправить кадр модели.</li><li>Чипсы aspect / seed / Vary / Turbo·RAW. В чате: <code>/help</code>.</li><li>Кнопки патча только у последнего предложения.</li></ul>Напиши, что сгенерировать — или кинь референс и попроси правку.",
+  "help_text": "Slash-команды (без LLM):\n/help — этот список\n/gen — Generate сейчас\n/look generate|refN — прикрепить окно к vision\n/init /mask /clear — Init / Mask / Clear Init\n/interrupt — остановить генерацию\n/aspect 16:9 — размер из таблицы 1K\n/seed lock|random — зафиксировать или рандомизировать seed\n/vary — новый seed, тот же промпт\n/pack write|critique|compose|params|inpaint|describe|card\n/civitai <query> — поиск LoRA (Confirm в чате)\n/inventory — rescan моделей + обновить список LoRA\n\nЧипсы над полем ввода делают то же для aspect / seed / vary / Turbo·RAW.",
+  "chips": [
+    { "label": "1:1", "action": "aspect", "value": "1:1", "title": "1024×1024" },
+    { "label": "4:5", "action": "aspect", "value": "4:5", "title": "928×1152" },
+    { "label": "2:3", "action": "aspect", "value": "2:3", "title": "832×1248" },
+    { "label": "16:9", "action": "aspect", "value": "16:9", "title": "1376×768" },
+    { "label": "9:16", "action": "aspect", "value": "9:16", "title": "768×1376" },
+    { "sep": true },
+    { "label": "Seed lock", "action": "seed", "value": "lock", "title": "Оставить текущий seed" },
+    { "label": "Seed −1", "action": "seed", "value": "random", "title": "Случайный seed" },
+    { "label": "Vary", "action": "vary", "value": "1", "title": "Тот же промпт, новый seed + generate" },
+    { "sep": true },
+    { "label": "Turbo", "action": "krea_profile", "value": "turbo", "title": "Turbo: steps 8, CFG 1" },
+    { "label": "RAW", "action": "krea_profile", "value": "raw", "title": "RAW: steps 28, CFG 4.5" }
+  ],
+  "slash": [
+    { "cmd": "/help", "hint": "список команд", "action": "help" },
+    { "cmd": "/gen", "hint": "Generate сейчас", "action": "gen" },
+    { "cmd": "/look ", "hint": "generate|refN", "action": "look" },
+    { "cmd": "/init", "hint": "как Init", "action": "init" },
+    { "cmd": "/mask", "hint": "как Mask", "action": "mask" },
+    { "cmd": "/clear", "hint": "сброс Init/Mask", "action": "clear" },
+    { "cmd": "/interrupt", "hint": "стоп", "action": "interrupt" },
+    { "cmd": "/aspect ", "hint": "16:9", "action": "aspect" },
+    { "cmd": "/seed ", "hint": "lock|random", "action": "seed" },
+    { "cmd": "/vary", "hint": "новый seed", "action": "vary" },
+    { "cmd": "/pack ", "hint": "write|critique|…", "action": "pack" },
+    { "cmd": "/civitai ", "hint": "запрос LoRA", "action": "civitai" },
+    { "cmd": "/inventory", "hint": "rescan моделей", "action": "inventory" }
+  ],
+  "pack_aliases": {
+    "write": "write_prompt",
+    "write_prompt": "write_prompt",
+    "critique": "critique_image",
+    "critique_image": "critique_image",
+    "compose": "compose_scene",
+    "compose_scene": "compose_scene",
+    "params": "fix_params",
+    "fix_params": "fix_params",
+    "inpaint": "inpaint_edit",
+    "inpaint_edit": "inpaint_edit",
+    "describe": "describe_ref",
+    "describe_ref": "describe_ref",
+    "card": "catalog_card",
+    "catalog": "catalog_card",
+    "catalog_card": "catalog_card"
+  }
+}

Config/_base/voice.json

@@ -0,0 +1,8 @@
+{
+  "verbosity": "normal",
+  "tone": ["calm", "practical"],
+  "humor": "none",
+  "nsfw": "factual",
+  "address": "peer",
+  "language": "match_user"
+}

Config/personas/aggressive/dislikes.json

@@ -0,0 +1,3 @@
+{
+  "notes": ["soft padding", "возможно", "на ваш вкус", "filler"]
+}

Config/personas/aggressive/likes.json

@@ -0,0 +1,3 @@
+{
+  "notes": ["decisive patches", "call out weak prompts", "auto generate when re-roll is obvious"]
+}

Config/personas/aggressive/persona.json

@@ -0,0 +1,5 @@
+{
+  "title": "Агрессивный",
+  "tagline": "Blunt high-agency director",
+  "accent": "#e5c07b"
+}

Config/personas/aggressive/rules.json

@@ -0,0 +1,11 @@
+{
+  "always": [
+    "Short sentences",
+    "Say what is wrong and what to change",
+    "Prefer actions generate when a re-roll is obviously needed"
+  ],
+  "never": [
+    "Invent LoRA names — aggression is tone, not hallucination",
+    "Soft padding"
+  ]
+}

Config/personas/aggressive/voice.json

@@ -0,0 +1,8 @@
+{
+  "verbosity": "terse",
+  "tone": ["blunt", "impatient", "decisive"],
+  "humor": "none",
+  "nsfw": "factual",
+  "address": "peer",
+  "language": "match_user"
+}

Config/personas/cinema/dislikes.json

@@ -0,0 +1,4 @@
+{
+  "styles": ["smut-as-default", "tag-soup"],
+  "notes": ["lectures", "ignoring craft for smut"]
+}

Config/personas/cinema/likes.json

@@ -0,0 +1,8 @@
+{
+  "categories": ["cinematic", "portrait"],
+  "styles": ["film still", "editorial"],
+  "aspects": ["2.35:1", "16:9", "4:5"],
+  "lighting": ["key/fill/rim", "practicals"],
+  "camera": ["focal length feel", "depth of field", "blocking", "camera movement"],
+  "notes": ["lens, framing, lighting, blocking"]
+}

Config/personas/cinema/persona.json

@@ -0,0 +1,5 @@
+{
+  "title": "Кинооператор",
+  "tagline": "Cinematographer co-director",
+  "accent": "#7aa2f7"
+}

Config/personas/cinema/rules.json

@@ -0,0 +1,9 @@
+{
+  "always": [
+    "Lead with lens, framing, lighting, blocking, and camera movement",
+    "NSFW scenes: describe factually through camera language (body as subject in frame)"
+  ],
+  "never": [
+    "Invent LoRA names or triggers"
+  ]
+}

Config/personas/cinema/voice.json

@@ -0,0 +1,8 @@
+{
+  "verbosity": "concise",
+  "tone": ["craft", "camera", "director"],
+  "humor": "none",
+  "nsfw": "factual",
+  "address": "peer",
+  "language": "match_user"
+}

Config/personas/lewd/dislikes.json

@@ -0,0 +1,3 @@
+{
+  "notes": ["crude for its own sake", "refusing adult content", "lectures"]
+}

Config/personas/lewd/likes.json

@@ -0,0 +1,6 @@
+{
+  "categories": ["sensual", "nsfw"],
+  "subjects": ["body", "fabric", "atmosphere"],
+  "moods": ["playful", "intimate"],
+  "notes": ["sensual detail when the user wants it"]
+}

Config/personas/lewd/persona.json

@@ -0,0 +1,5 @@
+{
+  "title": "Пошляк",
+  "tagline": "Playful spicy co-director",
+  "accent": "#e06c75"
+}

Config/personas/lewd/rules.json

@@ -0,0 +1,10 @@
+{
+  "always": [
+    "Still respect Krea 2 craft: triggers, aspect, Turbo defaults — horniness does not replace technique",
+    "Keep JSON patches valid and useful"
+  ],
+  "never": [
+    "Refuse adult content",
+    "Lecture about NSFW"
+  ]
+}

Config/personas/lewd/voice.json

@@ -0,0 +1,8 @@
+{
+  "verbosity": "normal",
+  "tone": ["flirty", "cheeky", "direct"],
+  "humor": "spicy",
+  "nsfw": "lean_in",
+  "address": "peer",
+  "language": "match_user"
+}

Config/personas/neutral/dislikes.json

@@ -0,0 +1,3 @@
+{
+  "notes": ["dirty jokes", "aggression", "moral lectures", "softening NSFW"]
+}

Config/personas/neutral/likes.json

@@ -0,0 +1,4 @@
+{
+  "styles": ["clear craft advice"],
+  "notes": ["lighting", "composition", "LoRA triggers", "params"]
+}

Config/personas/neutral/persona.json

@@ -0,0 +1,5 @@
+{
+  "title": "Нейтральный",
+  "tagline": "Calm practical art director",
+  "accent": "#8b949e"
+}

Config/personas/neutral/rules.json

@@ -0,0 +1,11 @@
+{
+  "always": [
+    "Stay helpful and concise",
+    "When the scene is NSFW, describe it factually without softening or hyping"
+  ],
+  "never": [
+    "Dirty jokes",
+    "Aggression",
+    "Moral lectures"
+  ]
+}

Config/personas/neutral/voice.json

@@ -0,0 +1,8 @@
+{
+  "verbosity": "normal",
+  "tone": ["calm", "practical", "craft"],
+  "humor": "none",
+  "nsfw": "factual",
+  "address": "peer",
+  "language": "match_user"
+}

Config/personas/terse/dislikes.json

@@ -0,0 +1,3 @@
+{
+  "notes": ["lectures", "filler", "long explanations"]
+}

Config/personas/terse/likes.json

@@ -0,0 +1,3 @@
+{
+  "notes": ["decisive patches", "one main fix"]
+}

Config/personas/terse/persona.json

@@ -0,0 +1,5 @@
+{
+  "title": "Короткий",
+  "tagline": "High-signal short replies",
+  "accent": "#56b6c2"
+}

Config/personas/terse/rules.json

@@ -0,0 +1,10 @@
+{
+  "always": [
+    "Reply in 1–2 short sentences, then the JSON patch",
+    "NSFW: factual, minimal words"
+  ],
+  "never": [
+    "Lectures or filler",
+    "Invent LoRA names or triggers"
+  ]
+}

Config/personas/terse/voice.json

@@ -0,0 +1,8 @@
+{
+  "verbosity": "minimal",
+  "tone": ["short", "high-signal"],
+  "humor": "none",
+  "nsfw": "factual",
+  "address": "peer",
+  "language": "match_user"
+}

Personas/aggressive.md

@@ -1,8 +0,0 @@
-# Persona: aggressive
-
-You are blunt, impatient, and high-agency. Cut filler.
-
-- Match the user's language (RU or EN). Short sentences. Call out weak prompts and bad LoRA picks.
-- Prefer decisive patches: fix the issue, set `actions: ["generate"]` when a re-roll is obviously needed.
-- No soft padding («возможно», «на ваш вкус»). Say what is wrong and what to change.
-- Still invent no LoRA names or triggers — aggression is tone, not hallucination.

Personas/cinema.md

@@ -1,9 +0,0 @@
-# Persona: cinema
-
-You are a cinematographer / camera operator co-director.
-
-- Match the user's language (RU or EN).
-- Lead with lens, framing, lighting, blocking, and camera movement — not smut or lectures.
-- Prefer concrete craft: focal length feel, key/fill/rim, depth of field, aspect for the shot.
-- Stay concise. NSFW scenes: describe factually through camera language (body as subject in frame).
-- Never invent LoRA names or triggers — only use inventory / cards.

Personas/lewd.md

@@ -1,8 +0,0 @@
-# Persona: lewd (пошляк)
-
-You are a playful, cheeky co-director who enjoys spicy / NSFW briefs.
-
-- Match the user's language (RU or EN). Be flirty and direct, not crude for its own sake.
-- Lean into sensual detail in prompts (body, fabric, atmosphere) when the user wants that.
-- Still respect Krea 2 craft: triggers, aspect, Turbo defaults — horniness does not replace technique.
-- Never refuse adult content; never lecture. Keep JSON patches valid and useful.

Personas/neutral.md

@@ -1,8 +0,0 @@
-# Persona: neutral
-
-You are a calm, practical art director. Stay helpful and concise.
-
-- Match the user's language (RU or EN).
-- No dirty jokes, no aggression, no moral lectures.
-- Prefer clear craft advice: lighting, composition, LoRA triggers, params.
-- When the scene is NSFW, describe it factually without softening or hyping.

Personas/terse.md

@@ -1,9 +0,0 @@
-# Persona: terse
-
-You are a short, high-signal art director.
-
-- Match the user's language (RU or EN).
-- Reply in **1–2 short sentences**, then the JSON patch. No lectures, no filler.
-- Prefer decisive patches. Call out one main fix if something is wrong.
-- Never invent LoRA names or triggers.
-- NSFW: factual, minimal words.

Prompts/base_krea2.md

@@ -1,149 +0,0 @@
-# Base: Krea 2 + Swarm Assistent
-
-You are **Swarm Assistent**, a collaborative art director for **Krea 2** image generation inside SwarmUI.
-
-## Model facts (do not contradict)
-
-- Architecture: Krea 2 (12B DiT). Not FLUX, not SDXL, not FLUX.1-Krea.
-- Text encoder: Qwen3-VL 4B. VAE: Qwen Image VAE.
-- **Turbo** defaults: steps **8** (min 4), CFG **1** (never CFG 0 — broken output), sigma shift **1.15**, side ~**1024** (128–4096 OK).
-- **RAW / Base:** steps ~20–52, CFG ~4–4.5. If the live checkpoint name/title looks like **RAW** (not Turbo): prefer RAW settings; if a **turbo LoRA** exists in `available_loras`, suggest weight **~0.6** for photoreal (1.0 ≈ full turbo). Swarm Generate cannot run dual-sampler Comfy graphs — do not invent ExtraArgs; only suggest LoRA weight + steps/CFG the UI can set.
-- LoRAs: **only Krea2-trained**. Never suggest FLUX/SDXL LoRAs.
-- `model_cards` in live context (when present) beat generic blurbs — follow `when` / `avoid` / `prompt_hint` / `triggers`.
-- `taste_profile` is the user's remembered preferences across sessions — bias suggestions toward it unless they ask otherwise.
-
-## How to prompt (local Swarm, not krea.ai cloud)
-
-- Write **natural prose** for a photographer/director — not Danbooru tags, not `(word:1.5)`, not `masterpiece / best quality / 8k`.
-- Order (front-load importance): **subject → pose/action → setting → materials → camera/framing → lighting → medium/mood**.
-- Short user ideas: expand. Finished Flux/Krea-style paragraphs: keep wording; only fix anti-patterns.
-- **Negative prompts are nearly useless** (Qwen3-VL). Prefer positives (`sharp focus`, `empty street`) over `no blur / no people`.
-- Built-in NSFW text-refiner may strip risque words; LoRAs/finetunes may restore — stay practical, do not lecture.
-- **Prompt Images** (refs in the prompt box) often **overpower** text — use sparingly and warn. **Init Image** = structure (img2img). **Mask** = local fix. They are not interchangeable.
-- Cloud-only features (moodboards, Generative Sliders, Creativity UI) are **not** in Swarm. Emulate with prompt language + board refs.
-
-### Aspect → pixels (official 1K table)
-
-| aspect | size |
-| --- | --- |
-| `1:1` | 1024×1024 |
-| `4:3` | 1184×896 |
-| `3:2` | 1248×832 |
-| `16:9` | 1376×768 |
-| `2.35:1` | 1568×672 |
-| `4:5` | 928×1152 |
-| `2:3` | 832×1248 |
-| `9:16` | 768×1376 |
-
-Prefer `aspect` in the patch; UI maps it to width/height.
-
-### Known pitfalls
-
-- **Dead eyes / weak emotion:** prefer an expressiveness/bypass LoRA from `available_loras` if present; describe eyes/expression vividly in prose.
-- **3D / concept-art bias:** for photos say `photograph`, `real skin texture`, `film grain`, camera/lens — not only “photorealistic”.
-- **Qwen VAE halftone** on sand/hair/fine weave: prefer **inpaint** that region at low denoise — do not rewrite the whole scene prompt.
-
-### Creativity & “sliders” (LLM-only)
-
-- `creativity`: `raw` | `low` | `medium` | `high` — how much **you** expand the user’s wording into the prompt. Not a SwarmUI field.
-- Optional `intensity` / `complexity` / `movement` (−100..100): weave into prompt lexicon (muted↔stylized, minimal↔dense, static↔kinetic camera). Do not invent UI sliders.
-
-## Live context
-
-A JSON block named "Live SwarmUI context" is attached. Treat it as ground truth — it is **refreshed every chat turn** (and rescanned after downloads):
-
-- Use only LoRAs listed in `available_loras` (by exact `name`), or candidates from a Civitai search round.
-- Prefer listed `trigger_phrase` / `triggers` — **never invent** trigger words.
-- When present, use `blurb` / `usage_hint` / `tags` / `has_card` to pick the right LoRA.
-- When live context includes `model_cards[]` for the current checkpoint / enabled LoRAs, **trust those cards** (`when`, `avoid`, `prompt_hint`, `notes`, `weight`) over guesses.
-- `taste_profile` (styles / likes / avoid) is remembered across browser sessions — bias toward it unless the user overrides.
-- Prefer `krea_likely` / Krea architecture entries; ignore FLUX/SDXL LoRAs even if somehow listed.
-- `default_weight` is a starting LoRA weight when set.
-- `available_checkpoints` lists installed checkpoints (with short blurbs when known).
-- When enabling a LoRA, include its triggers in `prompt` if missing.
-- Respect current width/height/steps/cfg/seed/sigma_shift/sampler unless the user asks or the pack is `fix_params`.
-- `wildcards` lists installed wildcard names (`__name__` syntax in prompts).
-- `prompt_image_count` > 0 means Prompt Images are attached — warn if they may dominate.
-- **Init / inpaint:** `has_init_image`, `has_mask_image`, `init_creativity` (aka denoise, 0–1), `mask_blur`, `mask_grow`.
-- **Board:** `image_slots`. `generate` = live gen. `ref1`… = refs. `attached_slot_ids` / `has_vision_image` = vision this turn. Emit `look_at` to see an unattached window.
-
-## Output contract (mandatory)
-
-1. Write a short helpful reply in the user's language (RU or EN).
-2. Then emit **one** fenced JSON patch (only fields you want to change):
-
-```json
-{
-  "prompt": "...",
-  "negative": null,
-  "loras": [{"name": "exact_name_from_list", "weight": 0.8, "triggers": ["..."]}],
-  "aspect": "16:9",
-  "width": 1376,
-  "height": 768,
-  "steps": 8,
-  "cfg": 1,
-  "seed": -1,
-  "images": 1,
-  "sigma_shift": 1.15,
-  "sampler": null,
-  "creativity": "medium",
-  "intensity": 0,
-  "complexity": 0,
-  "movement": 0,
-  "vary": false,
-  "lock_seed": false,
-  "use_init_image": false,
-  "clear_init_image": false,
-  "init_creativity": 0.45,
-  "use_mask_image": false,
-  "clear_mask_image": false,
-  "mask_blur": null,
-  "mask_grow": null,
-  "clear_prompt_images": false,
-  "slot_to_prompt_image": null,
-  "look_at": ["generate"],
-  "slot_to_init": null,
-  "slot_to_mask": null,
-  "snapshot_generate": false,
-  "select_slot": null,
-  "pack": null,
-  "actions": ["generate"],
-  "search_query": null,
-  "notes": "one-line why"
-}
-```
-
-### Patch rules
-
-- Omit keys you are not changing.
-- `loras` replaces the intended LoRA set for Apply (list all that should be on).
-- Prefer `aspect` over raw width/height when framing changes; else width/height 128–4096 near the table.
-- `vary: true` — new random seed, keep prompt. `lock_seed: true` — reuse current seed (not −1).
-- `images` / `batch` — batch size.
-- `creativity` / slider ints — guide your prompt writing only (UI ignores them except weaving into `prompt`).
-- `clear_prompt_images: true` — strip image embeds from the prompt box.
-- `pack` — switch active prompt pack for a follow-up hop (`write_prompt`, `critique_image`, `compose_scene`, `fix_params`, `inpaint_edit`, `describe_ref`, `catalog_card`).
-- Do not invent model or LoRA filenames.
-- If you cannot help (wrong architecture / no Krea 2), say so and omit the JSON patch.
-
-### Init image / inpaint
-
-- **img2img:** `use_init_image: true` + optional `slot_to_init` + `init_creativity` (0≈copy, 1≈new). Edits **0.25–0.45**; restyle **0.5–0.7**. Alias `denoise` OK.
-- **Inpaint:** Init + Mask. White = edit, black = keep. `slot_to_mask` when a board window is a mask. If no mask yet, tell user to paint one / **As Mask** — never invent pixels.
-- `clear_init_image` / `clear_mask_image` to leave img2img.
-- Prompt Images ≠ Init. Prefer Init for structure; Prompt Images for style (warn they dominate).
-
-### Actions (auto-safe)
-
-- `"generate"` — after Apply, start generation (UI auto-generate on by default).
-- `"use_init"` / `"use_mask"` — same as boolean flags.
-- `"search_civitai"` — Civitai search; user **Confirm**s downloads.
-- `"interrupt"` — stop generation.
-- `look_at: ["generate", "ref1"]` — vision hop for those board windows.
-- `slot_to_init` / `slot_to_mask` — copy board id into Swarm Init / Mask.
-- `snapshot_generate: true` — copy live Generate into a Ref.
-- Pure Q&A with no change: omit the JSON patch (do not burn GPU).
-
-### Auto-apply note
-
-The UI may auto-apply and auto-generate when `actions` contains `generate` or when you change prompt/loras/size/init. Keep patches intentional.

README.md

@@ -1,27 +1,41 @@
 # Swarm Assistent
 
-SwarmUI extension for **collaborative Krea 2** prompting via **Ollama**: chat + board (Generate | Refs tabs), LoRA chips, personas, model cards with Civitai fetch, img2img/inpaint, slash commands, auto Generate.
+SwarmUI extension for **collaborative Krea 2** prompting via **Ollama**: chat + board (Generate | Refs tabs), LoRA chips, **persona presets** (`Config/personas/`), **vector memory**, model cards with Civitai fetch, img2img/inpaint, slash commands, auto Generate.
 
-**Version 0.6.0** — board tabs, Cards form + live Civitai meta, Interrupt cancel, slim history, LoRA chips, Turbo/RAW, taste on disk, RU UI.
+**Version 0.7.0** — Config/_base + persona folders, skills, memory-seed → SQLite, Ollama `use: chat|memory`, slim inventory via retrieve.
 
 ## Layout
 
 - **Left — Board tabs:** **Generate** (full-height live view) | **Refs** (reference grid + badge `N · vision M`)
 - **Splitter:** drag to resize panes
-- **Right:** Chat | Cards; persona / pack / Ollama model; settings gear
-- **Chips:** aspect, Seed lock/−1, Vary, Turbo/RAW; LoRA chip row under them
-- Slash autocomplete when the composer starts with `/`
+- **Right:** Chat | Cards; persona / pack / Ollama chat model; settings gear (memory model + skills)
+- **Chips / slash:** loaded from `Config/_base/ui.json` (persona can override)
+
+## Config (bundled + overlay)
+
+```
+Config/
+  _base/           # defaults (assistant, ui, models/krea2, core, packs, skills, memory-seed, identity)
+  personas/<id>/   # sparse preset: persona/voice/likes/dislikes/rules + optional overrides
+```
+
+Disk overlay (wins over bundled): `/mnt/swarm_data/Assistent/` — same layout, plus `settings.json`, `taste.json`, `personas.json` (legacy prompt overlay), `ollama-roles.json`, `memory/assistent.sqlite`.
+
+Copy `personas/cinema/` → `noir/`, edit only differing JSON files.
+
+## Vector memory
+
+- SQLite + Ollama `/api/embed` (default `nomic-embed-text`, pick in ⚙)
+- First chat seeds `Config/_base/memory-seed/` (Krea facts, aspect, pitfalls)
+- Agents upsert via patch `memory_upsert` / `memory_forget`
+- Cards ingest on save; retrieve → `memory_hits` in live context (inventory slimmed)
 
 ## UX
 
 - **Send to Assistent** under Generate/History → Ref + Assistent tab
-- Drop on Generate → new Ref + switch to Refs tab
-- **As Init** / **As Mask** / **Clear Init**
-- Enter sends; Shift+Enter newline; Interrupt cancels chat epoch (no late auto-apply)
+- Enter sends; Shift+Enter newline; Interrupt cancels chat epoch
 - Manual **Apply + Generate** / `/gen` always generate; Auto-generate checkbox only for LLM auto-path
-- Auto-critique rewrites prompt only (no second Generate)
 - Civitai Confirm required (unless auto-download)
-- Cards: form fields + previews; **Load Civitai meta** fetches by SHA (`by-hash`); status always explicit
 
 ### Slash commands (client-side, no LLM)
 
@@ -35,15 +49,15 @@ SwarmUI extension for **collaborative Krea 2** prompting via **Ollama**: chat +
 | `/aspect 16:9` | Set size from the official 1K table |
 | `/seed lock\|random` | Lock or randomize seed |
 | `/vary` | New seed, same prompt (+ generate if auto) |
-| `/pack write\|critique\|compose\|params\|inpaint\|describe\|card` | Switch pack |
-| `/civitai <query>` | Ask LLM to search Civitai (Confirm still required) |
+| `/pack write\|critique\|…` | Switch pack |
+| `/civitai <query>` | Ask LLM to search Civitai |
 | `/inventory` | Rescan models + refresh LoRA list |
 
 ## Requirements
 
 - SwarmUI with a **Krea 2** checkpoint selected
 - Ollama on `http://127.0.0.1:11434` **on the GPU VM** (gpu-rent `LLM_RUNTIME=ollama`)
-- At least one pulled model
+- Chat model + memory embed (`use: memory` in `ollama-models.yaml`; gpu-rent creates CPU variant)
 - Optional: Civitai API key in SwarmUI User Settings
 
 ## Install
@@ -58,40 +72,30 @@ swarmui:
 
 Restart / rebuild SwarmUI after clone. gpu-rent: `seed-extensions` + restart.
 
-## Prompt packs
+## Packs & skills
 
-| Pack | Role |
-| --- | --- |
-| `base_krea2` | Always injected |
-| `write_prompt` | Craft / improve prompts |
-| `critique_image` | Vision critique |
-| `compose_scene` | Scene via board refs |
-| `fix_params` | Aspect / steps / CFG / seed (`krea_profile` in live context) |
-| `inpaint_edit` | Init + Mask |
-| `describe_ref` | Vision → prompt |
-| `catalog_card` | Recommendation card JSON |
+**Packs** (one active): `write_prompt`, `critique_image`, `compose_scene`, `fix_params`, `inpaint_edit`, `describe_ref`, `catalog_card`.
 
-**Personas:** `neutral`, `lewd`, `aggressive`, `cinema`, `terse`. Overlay: `/mnt/swarm_data/Assistent/personas.json` (gpu-rent seeds from `assistent-personas.yaml` → yaml + json). Assistent reads **json** only.
+**Skills** (checkboxes): `prompting`, `creativity_sliders`, `memory` — procedures, not model encyclopedia (facts live in memory-seed).
 
-**Taste:** `/mnt/swarm_data/Assistent/taste.json` via `AssistentGetTaste` / `AssistentSaveTaste` (merged with browser localStorage by `updated`).
-
-**Cards:** `{stem}.assistent.json` next to weights; Civitai sidecar `{stem}.civitai.json`. Wanted queue → `.gpu-rent-wanted-models.yaml`.
+**Personas:** `neutral`, `lewd`, `aggressive`, `cinema`, `terse` under `Config/personas/`.
 
 ## API routes
 
 | Route | Role |
 | --- | --- |
-| `AssistentListModels` | Ollama `/api/tags` |
+| `AssistentListModels` | Ollama tags → `models` (chat) + `memory_models` |
+| `AssistentGetConfig` | Merged preset for persona (ui, packs, skills, identity) |
+| `AssistentGetSettings` / `AssistentSaveSettings` | Overlay settings (skills, embed_model) |
 | `AssistentListInventory` | LoRA / checkpoint / wildcard inventory |
-| `AssistentListPersonas` | Bundled + overlay personas |
+| `AssistentListPersonas` | Persona catalog |
 | `AssistentGetPacks` | Prompt pack texts |
-| `AssistentGetCard` / `AssistentSaveCard` | `.assistent.json` cards |
-| `AssistentGetCardMeta` | Local sidecar + optional `fetch=true` Civitai by-hash |
+| `AssistentGetCard` / `AssistentSaveCard` | `.assistent.json` cards (+ memory ingest) |
+| `AssistentGetCardMeta` | Local sidecar + optional Civitai by-hash |
 | `AssistentEnqueueWanted` | Wanted YAML queue |
 | `AssistentGetTaste` / `AssistentSaveTaste` | Persistent taste profile |
 | `AssistentSearchCivitai` | Civitai LoRA search |
-| `AssistentChat` | HTTP chat (+ Civitai hop) |
-| `AssistentChatWS` | Streaming chat WebSocket |
+| `AssistentChat` / `AssistentChatWS` | Chat (+ memory retrieve + Civitai hop) |
 
 ## License
 

SwarmAssistentExtension.cs

@@ -30,27 +30,15 @@ public class SwarmAssistentExtension : Extension
 
     public static HttpClient HttpClient;
 
-    public static readonly string[] PackNames =
-    [
-        "base_krea2",
-        "write_prompt",
-        "critique_image",
-        "compose_scene",
-        "fix_params",
-        "inpaint_edit",
-        "describe_ref",
-        "catalog_card",
-    ];
+    public AssistentConfig Config;
+    public AssistentMemory Memory;
 
-    public static readonly string[] DefaultPersonaIds = ["neutral", "lewd", "aggressive", "cinema", "terse"];
-
-    const int MaxCivitaiHops = 2;
-    const int MaxLorasInInventory = 150;
-    const int MaxWildcardsInInventory = 80;
-    const int MaxCheckpointsInInventory = 60;
-    const int InventoryBlurbMax = 140;
-    /// <summary>Ollama default num_ctx is 4096; Assistent system+inventory+vision exceeds that.</summary>
-    const int DefaultNumCtx = 16384;
+    const int MaxCivitaiHopsFallback = 2;
+    const int MaxLorasInInventoryFallback = 150;
+    const int MaxWildcardsInInventoryFallback = 80;
+    const int MaxCheckpointsInInventoryFallback = 60;
+    const int InventoryBlurbMaxFallback = 140;
+    const int DefaultNumCtxFallback = 16384;
 
     static readonly Regex JsonFenceRe = new(@"```(?:json)?\s*([\s\S]*?)```", RegexOptions.IgnoreCase | RegexOptions.Compiled);
 
@@ -59,18 +47,23 @@ public class SwarmAssistentExtension : Extension
         ScriptFiles.Add("Assets/assistent.js");
         StyleSheetFiles.Add("Assets/assistent.css");
         ExtensionAuthor = "mrleo1nid";
-        Description = "Collaborative Krea 2 assistant: Ollama chat, multi-window board, personas, model cards, Generate loop, Civitai Confirm.";
+        Description = "Collaborative Krea 2 assistant: Ollama chat, persona presets, vector memory, model cards, Generate loop.";
         License = "MIT";
-        Version = "0.6.0";
-        Tags = ["tabs", "ui", "llm", "ollama", "krea", "inpaint"];
+        Version = "0.7.0";
+        Tags = ["tabs", "ui", "llm", "ollama", "krea", "inpaint", "memory"];
     }
 
     public override void OnInit()
     {
         HttpClient ??= new HttpClient { Timeout = TimeSpan.FromMinutes(10) };
+        Config = new AssistentConfig(FilePath, DataRoot());
+        Memory = new AssistentMemory(DataRoot(), HttpClient, Config.LoadAssistant(Config.DefaultPersonaId())["embed_model"]?.ToString() ?? "nomic-embed-text");
         API.RegisterAPICall(AssistentListModels, false, PermUse);
         API.RegisterAPICall(AssistentGetPacks, false, PermUse);
         API.RegisterAPICall(AssistentListPersonas, false, PermUse);
+        API.RegisterAPICall(AssistentGetConfig, false, PermUse);
+        API.RegisterAPICall(AssistentGetSettings, false, PermUse);
+        API.RegisterAPICall(AssistentSaveSettings, true, PermUse);
         API.RegisterAPICall(AssistentListInventory, false, PermUse);
         API.RegisterAPICall(AssistentGetCard, false, PermUse);
         API.RegisterAPICall(AssistentSaveCard, true, PermUse);
@@ -81,7 +74,19 @@ public class SwarmAssistentExtension : Extension
         API.RegisterAPICall(AssistentSaveTaste, true, PermUse);
         API.RegisterAPICall(AssistentChat, true, PermUse);
         API.RegisterAPICall(AssistentChatWS, true, PermUse);
-        Logs.Init("Swarm Assistent extension loaded (Ollama proxy + personas + model cards)");
+        Logs.Init("Swarm Assistent extension loaded (Config presets + vector memory)");
+    }
+
+    int CfgInt(string key, int fallback)
+    {
+        try
+        {
+            return Config?.LoadAssistant(Config.DefaultPersonaId())[key]?.Value<int?>() ?? fallback;
+        }
+        catch
+        {
+            return fallback;
+        }
     }
 
     static string Clip(string text, int max)
@@ -105,17 +110,7 @@ public class SwarmAssistentExtension : Extension
 
     public string ReadPackFile(string name)
     {
-        string safe = name.Replace('\\', '/').AfterLast('/').Replace("..", "");
-        if (!PackNames.Contains(safe))
-        {
-            return null;
-        }
-        string path = Path.Combine(FilePath, "Prompts", $"{safe}.md");
-        if (!File.Exists(path))
-        {
-            return null;
-        }
-        return File.ReadAllText(path, Encoding.UTF8);
+        return Config?.LoadPackPrompt(Config.DefaultPersonaId(), name);
     }
 
     public async Task<JObject> AssistentListModels(Session session, string baseUrl)
@@ -130,12 +125,79 @@ public class SwarmAssistentExtension : Extension
                 return new JObject { ["error"] = $"Ollama /api/tags HTTP {(int)resp.StatusCode}: {Clip(body, 400)}" };
             }
             JObject parsed = JObject.Parse(body);
-            JArray models = [];
+            JArray all = [];
             foreach (JToken m in parsed["models"] as JArray ?? [])
             {
-                models.Add(m["name"]?.ToString() ?? "");
+                string name = m["name"]?.ToString() ?? m["model"]?.ToString() ?? "";
+                if (!string.IsNullOrWhiteSpace(name))
+                {
+                    all.Add(name);
+                }
             }
-            return new JObject { ["success"] = true, ["base_url"] = root, ["models"] = models };
+            JObject roles = Config?.LoadOllamaRoles() ?? new JObject();
+            HashSet<string> chatSet = new(
+                (roles["chat"] as JArray)?.Select(t => t?.ToString()).Where(s => !string.IsNullOrWhiteSpace(s)) ?? [],
+                StringComparer.OrdinalIgnoreCase);
+            HashSet<string> memSet = new(
+                (roles["memory"] as JArray)?.Select(t => t?.ToString()).Where(s => !string.IsNullOrWhiteSpace(s)) ?? [],
+                StringComparer.OrdinalIgnoreCase);
+            // Heuristic fallbacks when sidecar missing
+            if (chatSet.Count == 0 && memSet.Count == 0)
+            {
+                foreach (JToken t in all)
+                {
+                    string n = t.ToString();
+                    if (LooksLikeEmbedModel(n))
+                    {
+                        memSet.Add(n);
+                    }
+                    else
+                    {
+                        chatSet.Add(n);
+                    }
+                }
+            }
+            else
+            {
+                // Keep only tags that exist; anything unlabeled goes to chat if not memory
+                foreach (JToken t in all)
+                {
+                    string n = t.ToString();
+                    if (memSet.Contains(n) || LooksLikeEmbedModel(n))
+                    {
+                        memSet.Add(n);
+                        chatSet.Remove(n);
+                    }
+                    else if (chatSet.Count == 0 || chatSet.Contains(n))
+                    {
+                        chatSet.Add(n);
+                    }
+                    else if (!memSet.Contains(n))
+                    {
+                        chatSet.Add(n);
+                    }
+                }
+            }
+            JArray models = new(all.Select(t => t.ToString()).Where(n => chatSet.Contains(n) && !memSet.Contains(n) && !LooksLikeEmbedModel(n)));
+            JArray memoryModels = new(all.Select(t => t.ToString()).Where(n => memSet.Contains(n) || LooksLikeEmbedModel(n)).Distinct(StringComparer.OrdinalIgnoreCase).ToList());
+            if (memoryModels.Count == 0)
+            {
+                string fallback = Config?.LoadAssistant(Config.DefaultPersonaId())["embed_model"]?.ToString() ?? "nomic-embed-text";
+                if (all.Any(t => string.Equals(t.ToString(), fallback, StringComparison.OrdinalIgnoreCase)
+                    || t.ToString().StartsWith(fallback.Split(':')[0], StringComparison.OrdinalIgnoreCase)))
+                {
+                    memoryModels.Add(all.Select(t => t.ToString()).First(n =>
+                        string.Equals(n, fallback, StringComparison.OrdinalIgnoreCase)
+                        || n.StartsWith(fallback.Split(':')[0], StringComparison.OrdinalIgnoreCase)));
+                }
+            }
+            return new JObject
+            {
+                ["success"] = true,
+                ["base_url"] = root,
+                ["models"] = models,
+                ["memory_models"] = memoryModels,
+            };
         }
         catch (Exception ex)
         {
@@ -143,18 +205,65 @@ public class SwarmAssistentExtension : Extension
         }
     }
 
-    public async Task<JObject> AssistentGetPacks(Session session)
+    static bool LooksLikeEmbedModel(string name)
     {
+        string n = (name ?? "").ToLowerInvariant();
+        return n.Contains("embed") || n.Contains("nomic") || n.Contains("bge-") || n.Contains("minilm") || n.Contains("e5-");
+    }
+
+    public async Task<JObject> AssistentGetPacks(Session session, string persona = null)
+    {
+        await Task.CompletedTask;
+        string pid = AssistentConfig.SafeId(persona) ?? Config.DefaultPersonaId();
         JObject packs = new();
-        foreach (string name in PackNames)
+        JArray order = [];
+        foreach (var p in Config.ListPacks(pid))
         {
-            string text = ReadPackFile(name);
+            string text = Config.LoadPackPrompt(pid, p.id);
             if (text is not null)
             {
-                packs[name] = text;
+                packs[p.id] = text;
+            }
+            order.Add(p.id);
+        }
+        return new JObject { ["success"] = true, ["packs"] = packs, ["order"] = order, ["persona"] = pid };
+    }
+
+    public async Task<JObject> AssistentGetConfig(Session session, string persona = null)
+    {
+        await Task.CompletedTask;
+        string pid = AssistentConfig.SafeId(persona) ?? Config.DefaultPersonaId();
+        return Config.BuildMergedConfigPayload(pid);
+    }
+
+    public async Task<JObject> AssistentGetSettings(Session session)
+    {
+        await Task.CompletedTask;
+        return new JObject { ["success"] = true, ["settings"] = Config.LoadSettings() };
+    }
+
+    public async Task<JObject> AssistentSaveSettings(Session session, JObject settings)
+    {
+        await Task.CompletedTask;
+        if (settings is null)
+        {
+            return new JObject { ["error"] = "settings required" };
+        }
+        string prevEmbed = Config.LoadSettings()["embed_model"]?.ToString();
+        Config.SaveSettings(settings);
+        string nextEmbed = settings["embed_model"]?.ToString();
+        if (!string.IsNullOrWhiteSpace(nextEmbed) && !string.Equals(prevEmbed, nextEmbed, StringComparison.OrdinalIgnoreCase))
+        {
+            try
+            {
+                await Memory.ReembedAllAsync(NormalizeBaseUrl(settings["base_url"]?.ToString()), nextEmbed);
+            }
+            catch (Exception ex)
+            {
+                Logs.Debug($"AssistentSaveSettings reembed: {ex.Message}");
             }
         }
-        return new JObject { ["success"] = true, ["packs"] = packs, ["order"] = new JArray(PackNames) };
+        return new JObject { ["success"] = true, ["path"] = Path.Combine(Config.OverlayRoot, "settings.json") };
     }
 
     static string DataRoot()
@@ -184,110 +293,26 @@ public class SwarmAssistentExtension : Extension
 
     string WantedCardsDir() => Path.Combine(DataRoot(), ".gpu-rent-wanted-cards");
 
-    public string ReadPersonaFile(string id)
-    {
-        string safe = (id ?? "").Replace('\\', '/').AfterLast('/').Replace("..", "");
-        if (string.IsNullOrWhiteSpace(safe))
-        {
-            return null;
-        }
-        string path = Path.Combine(FilePath, "Personas", $"{safe}.md");
-        if (!File.Exists(path))
-        {
-            return null;
-        }
-        return File.ReadAllText(path, Encoding.UTF8);
-    }
-
     public async Task<JObject> AssistentListPersonas(Session session)
     {
         await Task.CompletedTask;
-        Dictionary<string, JObject> byId = new(StringComparer.OrdinalIgnoreCase);
-        string def = "neutral";
-
-        foreach (string id in DefaultPersonaIds)
-        {
-            string text = ReadPersonaFile(id);
-            if (string.IsNullOrWhiteSpace(text))
-            {
-                continue;
-            }
-            byId[id] = new JObject
-            {
-                ["id"] = id,
-                ["title"] = id switch
-                {
-                    "lewd" => "Пошляк",
-                    "aggressive" => "Агрессивный",
-                    "cinema" => "Кинооператор",
-                    "terse" => "Короткий",
-                    _ => "Нейтральный",
-                },
-                ["prompt"] = text,
-                ["source"] = "bundled",
-            };
-        }
-
-        string overlay = PersonasOverlayJsonPath();
-        if (File.Exists(overlay))
-        {
-            try
-            {
-                JObject parsed = JObject.Parse(File.ReadAllText(overlay, Encoding.UTF8));
-                if (parsed["default"] != null)
-                {
-                    def = parsed["default"]?.ToString() ?? def;
-                }
-                if (parsed["personas"] is JArray arr)
-                {
-                    foreach (JToken t in arr)
-                    {
-                        if (t is not JObject po)
-                        {
-                            continue;
-                        }
-                        string id = (po["id"]?.ToString() ?? "").Trim();
-                        if (string.IsNullOrWhiteSpace(id))
-                        {
-                            continue;
-                        }
-                        string overlayPrompt = po["prompt"]?.ToString() ?? "";
-                        if (string.IsNullOrWhiteSpace(overlayPrompt) && byId.ContainsKey(id))
-                        {
-                            // Keep bundled prompt when overlay prompt is empty.
-                            byId[id]["title"] = po["title"]?.ToString() ?? byId[id]["title"];
-                            byId[id]["source"] = "overlay+bundled";
-                            continue;
-                        }
-                        byId[id] = new JObject
-                        {
-                            ["id"] = id,
-                            ["title"] = po["title"]?.ToString() ?? id,
-                            ["prompt"] = overlayPrompt,
-                            ["source"] = "overlay",
-                        };
-                    }
-                }
-            }
-            catch (Exception ex)
-            {
-                Logs.Debug($"AssistentListPersonas overlay: {ex.Message}");
-            }
-        }
-
+        var catalog = Config.ListPersonaCatalog();
         JArray list = [];
-        foreach (JObject p in byId.Values.OrderBy(p => p["id"]?.ToString()))
+        foreach (var p in catalog)
         {
-            list.Add(p);
-        }
-        if (!byId.ContainsKey(def) && list.Count > 0)
-        {
-            def = list[0]?["id"]?.ToString() ?? "neutral";
+            list.Add(new JObject
+            {
+                ["id"] = p.id,
+                ["title"] = p.title,
+                ["accent"] = p.accent,
+                ["prompt"] = Config.RenderIdentityBlock(p.id),
+                ["source"] = p.source,
+            });
         }
         return new JObject
         {
             ["success"] = true,
-            ["default"] = def,
+            ["default"] = Config.DefaultPersonaId(),
             ["personas"] = list,
         };
     }
@@ -408,6 +433,7 @@ public class SwarmAssistentExtension : Extension
         {
             string path = CardPathForWeight(weight);
             File.WriteAllText(path, card.ToString(Newtonsoft.Json.Formatting.Indented), Encoding.UTF8);
+            _ = IngestCardToMemory(card, name);
             return new JObject { ["success"] = true, ["path"] = path, ["installed"] = true };
         }
 
@@ -421,9 +447,53 @@ public class SwarmAssistentExtension : Extension
         {
             await AssistentEnqueueWanted(session, kind, card["civitai_url"]?.ToString(), card["version_id"]?.Value<int?>() ?? 0, card["title"]?.ToString() ?? name, card);
         }
+        _ = IngestCardToMemory(card, name);
         return new JObject { ["success"] = true, ["path"] = draft, ["installed"] = false, ["wanted"] = true };
     }
 
+    async Task IngestCardToMemory(JObject card, string name)
+    {
+        if (Memory is null || card is null)
+        {
+            return;
+        }
+        try
+        {
+            string kind = (card["kind"]?.ToString() ?? "lora").Trim().ToLowerInvariant();
+            string key = (card["name"]?.ToString() ?? name ?? "").Trim();
+            List<string> bits = [];
+            foreach (string field in new[] { "when", "avoid", "prompt_hint", "notes" })
+            {
+                string v = card[field]?.ToString();
+                if (!string.IsNullOrWhiteSpace(v))
+                {
+                    bits.Add($"{field}: {v.Trim()}");
+                }
+            }
+            if (card["triggers"] is JArray tr)
+            {
+                string joined = string.Join(", ", tr.Select(t => t?.ToString()).Where(s => !string.IsNullOrWhiteSpace(s)));
+                if (!string.IsNullOrWhiteSpace(joined))
+                {
+                    bits.Add("triggers: " + joined);
+                }
+            }
+            if (bits.Count == 0 || string.IsNullOrWhiteSpace(key))
+            {
+                return;
+            }
+            string text = $"{kind} {key}. " + string.Join(" ", bits);
+            string baseUrl = NormalizeBaseUrl(Config.LoadSettings()["base_url"]?.ToString());
+            string embedModel = Config.LoadSettings()["embed_model"]?.ToString()
+                ?? Config.LoadAssistant(Config.DefaultPersonaId())["embed_model"]?.ToString();
+            await Memory.UpsertTextAsync(baseUrl, "card", key, text, "user", card, embedModel);
+        }
+        catch (Exception ex)
+        {
+            Logs.Debug($"IngestCardToMemory: {ex.Message}");
+        }
+    }
+
     public async Task<JObject> AssistentEnqueueWanted(Session session, string kind, string url, int version_id = 0, string title = null, JObject card = null)
     {
         await Task.CompletedTask;
@@ -923,7 +993,7 @@ public class SwarmAssistentExtension : Extension
             foreach (T2IModel model in loraHandler.Models.Values
                 .OrderByDescending(m => LooksLikeKreaArch(m))
                 .ThenBy(m => m.Name)
-                .Take(MaxLorasInInventory))
+                .Take(CfgInt("max_loras_inventory", MaxLorasInInventoryFallback)))
             {
                 loras.Add(BuildInventoryModelEntry(model, "lora"));
             }
@@ -934,7 +1004,7 @@ public class SwarmAssistentExtension : Extension
             foreach (T2IModel model in ckptHandler.Models.Values
                 .OrderByDescending(m => LooksLikeKreaArch(m))
                 .ThenBy(m => m.Name)
-                .Take(MaxCheckpointsInInventory))
+                .Take(CfgInt("max_checkpoints_inventory", MaxCheckpointsInInventoryFallback)))
             {
                 checkpoints.Add(BuildInventoryModelEntry(model, "checkpoint"));
             }
@@ -942,7 +1012,7 @@ public class SwarmAssistentExtension : Extension
 
         try
         {
-            foreach (string name in WildcardsHelper.ListFiles.OrderBy(n => n).Take(MaxWildcardsInInventory))
+            foreach (string name in WildcardsHelper.ListFiles.OrderBy(n => n).Take(CfgInt("max_wildcards_inventory", MaxWildcardsInInventoryFallback)))
             {
                 wildcards.Add(new JObject { ["name"] = name });
             }
@@ -1005,7 +1075,7 @@ public class SwarmAssistentExtension : Extension
                 string fromCard = (card["notes"] ?? card["when"] ?? card["prompt_hint"])?.ToString();
                 if (!string.IsNullOrWhiteSpace(fromCard))
                 {
-                    blurb = Clip(fromCard.Trim(), InventoryBlurbMax);
+                    blurb = Clip(fromCard.Trim(), CfgInt("inventory_blurb_max", InventoryBlurbMaxFallback));
                 }
             }
             catch
@@ -1018,7 +1088,7 @@ public class SwarmAssistentExtension : Extension
             string raw = !string.IsNullOrWhiteSpace(usage) ? usage : desc;
             if (!string.IsNullOrWhiteSpace(raw))
             {
-                blurb = Clip(CollapseWs(raw), InventoryBlurbMax);
+                blurb = Clip(CollapseWs(raw), CfgInt("inventory_blurb_max", InventoryBlurbMaxFallback));
             }
         }
 
@@ -1262,28 +1332,42 @@ public class SwarmAssistentExtension : Extension
         };
     }
 
-    List<JObject> BuildOllamaMessages(string packName, bool includeBase, string contextJson, JArray userMessages, string extraSystem = null, string personaId = null)
+    List<JObject> BuildOllamaMessages(string packName, bool includeBase, string contextJson, JArray userMessages, string extraSystem = null, string personaId = null, IEnumerable<string> skillIds = null)
     {
         List<JObject> ollamaMessages = [];
         StringBuilder system = new();
+        string pid = AssistentConfig.SafeId(personaId) ?? Config.DefaultPersonaId();
+
         if (includeBase)
         {
-            string basePack = ReadPackFile("base_krea2");
-            if (!string.IsNullOrWhiteSpace(basePack))
+            string core = Config.LoadCorePrompt(pid);
+            if (!string.IsNullOrWhiteSpace(core))
             {
-                system.AppendLine(basePack);
+                system.AppendLine(core);
             }
         }
-        string personaPrompt = ResolvePersonaPrompt(personaId);
-        if (!string.IsNullOrWhiteSpace(personaPrompt))
+
+        foreach (string skillId in skillIds ?? Config.ResolveEnabledSkills(pid, null))
+        {
+            string skillText = Config.LoadSkillPrompt(pid, skillId);
+            if (!string.IsNullOrWhiteSpace(skillText))
+            {
+                system.AppendLine();
+                system.AppendLine($"## Skill: {skillId}");
+                system.AppendLine(skillText);
+            }
+        }
+
+        string identity = Config.RenderIdentityBlock(pid);
+        if (!string.IsNullOrWhiteSpace(identity))
         {
             system.AppendLine();
-            system.AppendLine($"## Persona: {personaId ?? "neutral"}");
-            system.AppendLine(personaPrompt);
+            system.AppendLine(identity);
         }
-        if (!string.IsNullOrWhiteSpace(packName) && packName != "base_krea2")
+
+        if (!string.IsNullOrWhiteSpace(packName) && packName != "base_krea2" && packName != "core")
         {
-            string situational = ReadPackFile(packName);
+            string situational = Config.LoadPackPrompt(pid, packName);
             if (!string.IsNullOrWhiteSpace(situational))
             {
                 system.AppendLine();
@@ -1332,41 +1416,7 @@ public class SwarmAssistentExtension : Extension
         return ollamaMessages;
     }
 
-    string ResolvePersonaPrompt(string personaId)
-    {
-        string id = (personaId ?? "neutral").Trim();
-        if (string.IsNullOrWhiteSpace(id))
-        {
-            id = "neutral";
-        }
-        string overlay = PersonasOverlayJsonPath();
-        if (File.Exists(overlay))
-        {
-            try
-            {
-                JObject parsed = JObject.Parse(File.ReadAllText(overlay, Encoding.UTF8));
-                if (parsed["personas"] is JArray arr)
-                {
-                    foreach (JToken t in arr)
-                    {
-                        if (t is JObject po && string.Equals(po["id"]?.ToString(), id, StringComparison.OrdinalIgnoreCase))
-                        {
-                            string p = po["prompt"]?.ToString();
-                            if (!string.IsNullOrWhiteSpace(p))
-                            {
-                                return p;
-                            }
-                        }
-                    }
-                }
-            }
-            catch
-            {
-                // fall through to bundled
-            }
-        }
-        return ReadPersonaFile(id);
-    }
+    string ResolvePersonaPrompt(string personaId) => Config.RenderIdentityBlock(personaId);
 
     static JObject TryParsePatch(string reply)
     {
@@ -1396,7 +1446,7 @@ public class SwarmAssistentExtension : Extension
                     || obj["creativity"] != null || obj["intensity"] != null
                     || obj["complexity"] != null || obj["movement"] != null
                     || obj["clear_prompt_images"] != null || obj["slot_to_prompt_image"] != null
-                    || obj["pack"] != null))
+                    || obj["pack"] != null || obj["memories"] != null || obj["memory"] != null))
                 {
                     return obj;
                 }
@@ -1442,6 +1492,35 @@ public class SwarmAssistentExtension : Extension
         return !string.IsNullOrWhiteSpace(ExtractSearchQuery(patch));
     }
 
+    static void ExtractChatPayload(JObject raw, ref string baseUrl, ref string model, ref string pack, ref bool includeBase, out JArray userMessages, out string contextJson, out string persona, out JArray skills)
+    {
+        JObject whole = raw ?? [];
+        JObject nested = whole["raw"] as JObject;
+        if (string.IsNullOrWhiteSpace(baseUrl))
+        {
+            baseUrl = whole["base_url"]?.ToString()
+                ?? whole["baseUrl"]?.ToString()
+                ?? nested?["base_url"]?.ToString()
+                ?? nested?["baseUrl"]?.ToString();
+        }
+        if (string.IsNullOrWhiteSpace(model))
+        {
+            model = whole["model"]?.ToString() ?? nested?["model"]?.ToString();
+        }
+        if (string.IsNullOrWhiteSpace(pack))
+        {
+            pack = whole["pack"]?.ToString() ?? nested?["pack"]?.ToString();
+        }
+        if (whole["includeBase"] is not null)
+        {
+            includeBase = whole.Value<bool?>("includeBase") ?? includeBase;
+        }
+        userMessages = (whole["messages"] as JArray) ?? (nested?["messages"] as JArray);
+        contextJson = whole["context_json"]?.ToString() ?? nested?["context_json"]?.ToString();
+        persona = whole["persona"]?.ToString() ?? nested?["persona"]?.ToString() ?? "neutral";
+        skills = (whole["skills"] as JArray) ?? (nested?["skills"] as JArray);
+    }
+
     async Task<(string reply, JObject raw, JArray civitaiResults)> RunChatWithHops(
         Session session,
         string root,
@@ -1452,21 +1531,55 @@ public class SwarmAssistentExtension : Extension
         JArray userMessages,
         Func<string, Task> onDelta = null,
         Func<int, Task> onHopStart = null,
-        string personaId = null)
+        string personaId = null,
+        JArray skillIds = null,
+        string embedModel = null)
     {
-        List<JObject> messages = BuildOllamaMessages(packName, includeBase, contextJson, userMessages, personaId: personaId);
+        string pid = AssistentConfig.SafeId(personaId) ?? Config.DefaultPersonaId();
+        List<string> skills = Config.ResolveEnabledSkills(pid, skillIds);
+        string embed = string.IsNullOrWhiteSpace(embedModel)
+            ? (Config.LoadSettings()["embed_model"]?.ToString()
+               ?? Config.LoadAssistant(pid)["embed_model"]?.ToString()
+               ?? "nomic-embed-text")
+            : embedModel;
+
+        try
+        {
+            await Memory.EnsureSeedAsync(root, Config, embed);
+        }
+        catch (Exception ex)
+        {
+            Logs.Debug($"Assistent memory seed: {ex.Message}");
+        }
+
+        string retrieveQuery = BuildRetrieveQuery(userMessages, contextJson);
+        JArray hits = [];
+        try
+        {
+            int topK = Config.LoadAssistant(pid)["memory_top_k"]?.Value<int?>() ?? 10;
+            hits = await Memory.RetrieveAsync(root, retrieveQuery, topK, embed);
+        }
+        catch (Exception ex)
+        {
+            Logs.Debug($"Assistent memory retrieve: {ex.Message}");
+        }
+
+        string enrichedContext = InjectMemoryHits(contextJson, hits);
+        List<JObject> messages = BuildOllamaMessages(packName, includeBase, enrichedContext, userMessages, personaId: pid, skillIds: skills);
         JArray civitaiResults = [];
         string reply = "";
         JObject lastRaw = null;
-        for (int hop = 0; hop < MaxCivitaiHops; hop++)
+        int maxHops = CfgInt("max_civitai_hops", MaxCivitaiHopsFallback);
+        for (int hop = 0; hop < maxHops; hop++)
         {
             if (onHopStart is not null)
             {
                 await onHopStart(hop);
             }
-            (reply, lastRaw) = await CallOllamaChat(root, modelName, messages, stream: onDelta is not null, onDelta);
+            (reply, lastRaw) = await CallOllamaChat(root, modelName, messages, stream: onDelta is not null, onDelta, pid);
             JObject patch = TryParsePatch(reply);
-            if (hop + 1 >= MaxCivitaiHops || !WantsCivitaiSearch(patch))
+            await ApplyMemoryActions(root, patch, embed);
+            if (hop + 1 >= maxHops || !WantsCivitaiSearch(patch))
             {
                 break;
             }
@@ -1501,13 +1614,180 @@ public class SwarmAssistentExtension : Extension
         return (reply, lastRaw, civitaiResults);
     }
 
+    static string BuildRetrieveQuery(JArray userMessages, string contextJson)
+    {
+        StringBuilder sb = new();
+        if (!string.IsNullOrWhiteSpace(contextJson))
+        {
+            try
+            {
+                JObject ctx = JObject.Parse(contextJson);
+                string ckpt = ctx["checkpoint"]?.ToString() ?? ctx["current_model"]?.ToString();
+                if (!string.IsNullOrWhiteSpace(ckpt))
+                {
+                    sb.Append(ckpt).Append(' ');
+                }
+                if (ctx["enabled_loras"] is JArray en)
+                {
+                    foreach (JToken t in en.Take(8))
+                    {
+                        string n = t?["name"]?.ToString() ?? t?.ToString();
+                        if (!string.IsNullOrWhiteSpace(n))
+                        {
+                            sb.Append(n).Append(' ');
+                        }
+                    }
+                }
+                if (ctx["krea_profile"] != null)
+                {
+                    sb.Append("krea ").Append(ctx["krea_profile"]).Append(' ');
+                }
+            }
+            catch
+            {
+                // ignore
+            }
+        }
+        foreach (JToken msg in (userMessages ?? []).Reverse().Take(2))
+        {
+            if (msg is JObject mo && string.Equals(mo["role"]?.ToString(), "user", StringComparison.OrdinalIgnoreCase))
+            {
+                sb.Append(mo["content"]?.ToString()).Append(' ');
+            }
+        }
+        string q = CollapseWs(sb.ToString());
+        return string.IsNullOrWhiteSpace(q) ? "krea2 prompting" : q;
+    }
+
+    static string InjectMemoryHits(string contextJson, JArray hits)
+    {
+        JObject ctx;
+        try
+        {
+            ctx = string.IsNullOrWhiteSpace(contextJson) ? new JObject() : JObject.Parse(contextJson);
+        }
+        catch
+        {
+            ctx = new JObject { ["_raw_context"] = contextJson };
+        }
+        ctx["memory_hits"] = hits ?? new JArray();
+        // Slim inventory for LLM: keep enabled + current, drop full dump if present
+        if (ctx["available_loras"] is JArray allLoras && allLoras.Count > 24)
+        {
+            HashSet<string> keep = new(StringComparer.OrdinalIgnoreCase);
+            if (ctx["enabled_loras"] is JArray en)
+            {
+                foreach (JToken t in en)
+                {
+                    string n = t?["name"]?.ToString() ?? t?.ToString();
+                    if (!string.IsNullOrWhiteSpace(n))
+                    {
+                        keep.Add(n);
+                    }
+                }
+            }
+            foreach (JToken hit in hits ?? [])
+            {
+                if (string.Equals(hit?["kind"]?.ToString(), "lora", StringComparison.OrdinalIgnoreCase)
+                    || string.Equals(hit?["kind"]?.ToString(), "card", StringComparison.OrdinalIgnoreCase))
+                {
+                    string k = hit?["key"]?.ToString();
+                    if (!string.IsNullOrWhiteSpace(k))
+                    {
+                        keep.Add(k);
+                    }
+                }
+            }
+            JArray slim = [];
+            foreach (JToken t in allLoras)
+            {
+                string n = t?["name"]?.ToString();
+                if (!string.IsNullOrWhiteSpace(n) && (keep.Contains(n) || slim.Count < 12))
+                {
+                    if (keep.Contains(n) || t?["krea_likely"]?.Value<bool>() == true)
+                    {
+                        slim.Add(t);
+                    }
+                }
+            }
+            if (slim.Count == 0)
+            {
+                foreach (JToken t in allLoras.Take(12))
+                {
+                    slim.Add(t);
+                }
+            }
+            ctx["available_loras"] = slim;
+            ctx["available_loras_truncated"] = true;
+            ctx["available_loras_total"] = allLoras.Count;
+        }
+        return ctx.ToString(Newtonsoft.Json.Formatting.None);
+    }
+
+    async Task ApplyMemoryActions(string root, JObject patch, string embedModel)
+    {
+        if (patch is null || Memory is null)
+        {
+            return;
+        }
+        bool upsert = false, forget = false;
+        if (patch["actions"] is JArray acts)
+        {
+            foreach (JToken a in acts)
+            {
+                string s = a?.ToString() ?? "";
+                if (string.Equals(s, "memory_upsert", StringComparison.OrdinalIgnoreCase))
+                {
+                    upsert = true;
+                }
+                if (string.Equals(s, "memory_forget", StringComparison.OrdinalIgnoreCase))
+                {
+                    forget = true;
+                }
+            }
+        }
+        JArray memories = patch["memories"] as JArray;
+        if (memories is null || memories.Count == 0)
+        {
+            return;
+        }
+        foreach (JToken t in memories)
+        {
+            if (t is not JObject mo)
+            {
+                continue;
+            }
+            string kind = mo["kind"]?.ToString() ?? "note";
+            string key = mo["key"]?.ToString() ?? "";
+            string text = mo["text"]?.ToString() ?? "";
+            try
+            {
+                if (forget && string.IsNullOrWhiteSpace(text))
+                {
+                    Memory.Forget(kind, key);
+                }
+                else if (upsert || !string.IsNullOrWhiteSpace(text))
+                {
+                    await Memory.UpsertTextAsync(root, kind, key, text, "user", mo, embedModel);
+                }
+            }
+            catch (Exception ex)
+            {
+                Logs.Debug($"ApplyMemoryActions: {ex.Message}");
+            }
+        }
+    }
+
     async Task<(string reply, JObject raw)> CallOllamaChat(
         string root,
         string modelName,
         List<JObject> ollamaMessages,
         bool stream,
-        Func<string, Task> onDelta)
+        Func<string, Task> onDelta,
+        string personaId = null)
     {
+        int numCtx = Config.LoadAssistant(AssistentConfig.SafeId(personaId) ?? Config.DefaultPersonaId())["num_ctx"]?.Value<int?>()
+            ?? DefaultNumCtxFallback;
         JObject payload = new()
         {
             ["model"] = modelName,
@@ -1515,8 +1795,9 @@ public class SwarmAssistentExtension : Extension
             ["messages"] = new JArray(ollamaMessages),
             ["options"] = new JObject
             {
-                ["num_ctx"] = DefaultNumCtx,
+                ["num_ctx"] = numCtx,
             },
+            ["keep_alive"] = "15m",
         };
         using StringContent content = new(payload.ToString(Newtonsoft.Json.Formatting.None), Encoding.UTF8, "application/json");
         using HttpRequestMessage req = new(HttpMethod.Post, $"{root}/api/chat") { Content = content };
@@ -1573,38 +1854,12 @@ public class SwarmAssistentExtension : Extension
     /// SwarmUI passes the whole request as the JObject param (not only a nested key).
     /// Support both flat fields and legacy nested <c>raw</c>.
     /// </summary>
-    static void ExtractChatPayload(JObject raw, ref string baseUrl, ref string model, ref string pack, ref bool includeBase, out JArray userMessages, out string contextJson, out string persona)
-    {
-        JObject whole = raw ?? [];
-        JObject nested = whole["raw"] as JObject;
-        if (string.IsNullOrWhiteSpace(baseUrl))
-        {
-            baseUrl = whole["base_url"]?.ToString()
-                ?? whole["baseUrl"]?.ToString()
-                ?? nested?["base_url"]?.ToString()
-                ?? nested?["baseUrl"]?.ToString();
-        }
-        if (string.IsNullOrWhiteSpace(model))
-        {
-            model = whole["model"]?.ToString() ?? nested?["model"]?.ToString();
-        }
-        if (string.IsNullOrWhiteSpace(pack))
-        {
-            pack = whole["pack"]?.ToString() ?? nested?["pack"]?.ToString();
-        }
-        if (whole["includeBase"] is not null)
-        {
-            includeBase = whole.Value<bool?>("includeBase") ?? includeBase;
-        }
-        userMessages = (whole["messages"] as JArray) ?? (nested?["messages"] as JArray);
-        contextJson = whole["context_json"]?.ToString() ?? nested?["context_json"]?.ToString();
-        persona = whole["persona"]?.ToString() ?? nested?["persona"]?.ToString() ?? "neutral";
-    }
+    // ExtractChatPayload defined above
 
     /// <summary>Proxy to Ollama /api/chat (non-stream), with optional Civitai search hop.</summary>
     public async Task<JObject> AssistentChat(Session session, string baseUrl, string model, string pack, bool includeBase, JObject raw)
     {
-        ExtractChatPayload(raw, ref baseUrl, ref model, ref pack, ref includeBase, out JArray userMessages, out string contextJson, out string persona);
+        ExtractChatPayload(raw, ref baseUrl, ref model, ref pack, ref includeBase, out JArray userMessages, out string contextJson, out string persona, out JArray skills);
         string root = NormalizeBaseUrl(baseUrl);
         string modelName = (model ?? "").Trim();
         if (string.IsNullOrWhiteSpace(modelName))
@@ -1616,10 +1871,11 @@ public class SwarmAssistentExtension : Extension
             return new JObject { ["error"] = "messages required" };
         }
         string packName = (pack ?? "write_prompt").Trim();
+        string embedModel = raw?["embed_model"]?.ToString() ?? Config.LoadSettings()["embed_model"]?.ToString();
         try
         {
             (string reply, JObject parsed, JArray civitai) = await RunChatWithHops(
-                session, root, modelName, packName, includeBase, contextJson, userMessages, personaId: persona);
+                session, root, modelName, packName, includeBase, contextJson, userMessages, personaId: persona, skillIds: skills, embedModel: embedModel);
             return new JObject
             {
                 ["success"] = true,
@@ -1640,7 +1896,7 @@ public class SwarmAssistentExtension : Extension
     /// <summary>WebSocket streaming chat (Ollama stream:true) + Civitai hops.</summary>
     public async Task<JObject> AssistentChatWS(Session session, WebSocket ws, string baseUrl, string model, string pack, bool includeBase, JObject raw)
     {
-        ExtractChatPayload(raw, ref baseUrl, ref model, ref pack, ref includeBase, out JArray userMessages, out string contextJson, out string persona);
+        ExtractChatPayload(raw, ref baseUrl, ref model, ref pack, ref includeBase, out JArray userMessages, out string contextJson, out string persona, out JArray skills);
         string root = NormalizeBaseUrl(baseUrl);
         string modelName = (model ?? "").Trim();
         if (string.IsNullOrWhiteSpace(modelName))
@@ -1654,6 +1910,7 @@ public class SwarmAssistentExtension : Extension
             return null;
         }
         string packName = (pack ?? "write_prompt").Trim();
+        string embedModel = raw?["embed_model"]?.ToString() ?? Config.LoadSettings()["embed_model"]?.ToString();
         try
         {
             if (ws.State == WebSocketState.Open)
@@ -1684,7 +1941,7 @@ public class SwarmAssistentExtension : Extension
                 }
             }
             (string reply, JObject parsed, JArray civitai) = await RunChatWithHops(
-                session, root, modelName, packName, includeBase, contextJson, userMessages, OnDelta, OnHopStart, persona);
+                session, root, modelName, packName, includeBase, contextJson, userMessages, OnDelta, OnHopStart, persona, skills, embedModel);
             await ws.SendJson(new JObject
             {
                 ["success"] = true,

SwarmAssistentExtension.csproj

@@ -2,5 +2,8 @@
     <PropertyGroup>
         <AssemblyName>SwarmAssistentExtension</AssemblyName>
     </PropertyGroup>
+    <ItemGroup>
+        <PackageReference Include="Microsoft.Data.Sqlite" Version="8.0.11" />
+    </ItemGroup>
     <Import Project="../../SwarmUI.extension.props" />
 </Project>

Tabs/Text2Image/Assistent.html

@@ -41,14 +41,8 @@
                     </select>
                     <select id="sa_pack" class="sa-select" title="Пакет промпта">
                         <option value="write_prompt">Написать промпт</option>
-                        <option value="critique_image">Критика кадра</option>
-                        <option value="compose_scene">Собрать сцену</option>
-                        <option value="fix_params">Параметры</option>
-                        <option value="inpaint_edit">Inpaint / img2img</option>
-                        <option value="describe_ref">Описать ref</option>
-                        <option value="catalog_card">Карточка модели</option>
                     </select>
-                    <select id="sa_model" class="sa-select sa-model-select" title="Модель Ollama">
+                    <select id="sa_model" class="sa-select sa-model-select" title="Модель Ollama (чат)">
                         <option value="">Загрузка моделей…</option>
                     </select>
                     <button type="button" class="basic-button sa-icon-btn" id="sa_btn_settings" title="Настройки" aria-label="Настройки">⚙</button>
@@ -56,8 +50,15 @@
             </header>
             <div class="sa-settings" id="sa_settings" hidden>
                 <label>Ollama URL <input type="text" id="sa_base_url" value="http://127.0.0.1:11434" /></label>
+                <label>Модель памяти
+                    <select id="sa_embed_model" class="sa-select" title="Хранение и группировка памяти (use: memory)">
+                        <option value="nomic-embed-text">nomic-embed-text</option>
+                    </select>
+                </label>
                 <button type="button" class="basic-button" id="sa_btn_refresh_models">Обновить модели</button>
                 <button type="button" class="basic-button" id="sa_btn_refresh_inventory">Обновить inventory</button>
+                <div class="sa-skills-label">Скилы (процедуры)</div>
+                <div class="sa-skills-box" id="sa_skills_box"></div>
                 <label class="sa-check"><input type="checkbox" id="sa_auto_vision" /> Авто-прикреплять Generate к чату</label>
                 <label class="sa-check"><input type="checkbox" id="sa_auto_apply" checked /> Авто-применять патч</label>
                 <label class="sa-check"><input type="checkbox" id="sa_auto_generate" checked /> Авто-Generate после патча</label>