Add assistent-personas support and enhance model merging functionality

- Introduced `assistent-personas.yaml` to the project and updated the `.gitignore` accordingly.
- Implemented functions to seed and merge assistent personas from local files to the VM.
- Enhanced the model capture process to include merging of wanted models from the VM into `models.yaml`.
- Updated documentation to reflect changes in the assistent personas and model management processes.

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

.gitignore

@@ -5,6 +5,7 @@
 gpu-rent.vars
 ollama-models.yaml
 models.yaml
+assistent-personas.yaml
 extensions.yaml
 
 # SwarmUI checkpoints / LoRA dropped into the app tree

assistent-personas.example.yaml

@@ -0,0 +1,27 @@
+# Copy to assistent-personas.yaml (never commit the working copy).
+# Synced to the VM on each up → /mnt/swarm_data/Assistent/personas.yaml
+# Overlay wins over bundled Personas/*.md in swarm-assistent for the same id.
+
+default: neutral
+
+personas:
+  - id: neutral
+    title: Нейтральный
+    prompt: |
+      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. NSFW scenes: describe factually without softening or hyping.
+
+  - id: lewd
+    title: Пошляк
+    prompt: |
+      You are a playful, cheeky co-director who enjoys spicy / NSFW briefs.
+      Match the user's language. Be flirty and direct. Lean into sensual detail when asked.
+      Still respect Krea 2 craft (triggers, aspect, Turbo). Never refuse adult content; never lecture.
+
+  - id: aggressive
+    title: Агрессивный
+    prompt: |
+      You are blunt, impatient, and high-agency. Cut filler. Short sentences.
+      Call out weak prompts and bad LoRA picks. Prefer decisive patches with generate when needed.
+      Invent no LoRA names or triggers — aggression is tone, not hallucination.

docs/extensions.md

@@ -53,7 +53,7 @@ comfy:
 | `ollama` | только при `LLM_RUNTIME=ollama` |
 | `any-llm` | при `LLM_RUNTIME=ollama` |
 
-Строки с несовпавшим `requires` пропускаются (лог), остальные ставятся как обычно. В `extensions.example.yaml` по умолчанию — **swarm-assistent** с `requires: ollama` (чат + доска Generate/Ref + vision под Krea 2, ветка `main`).
+Строки с несовпавшим `requires` пропускаются (лог), остальные ставятся как обычно. В `extensions.example.yaml` по умолчанию — **swarm-assistent** с `requires: ollama` (чат + доска Generate/Ref + persona/Cards, ветка `main`). На `up` также сидится `assistent-personas.yaml` → `/mnt/swarm_data/Assistent/personas.yaml`.
 
 В git репозитория gpu-rent не коммитить рабочий список с лишними приватными URL сверх примера.
 

docs/models.md

@@ -27,7 +27,7 @@ copy models.example.yaml models.yaml
 .\gpu-rent.ps1 capture all
 ```
 
-Берёт `{stem}.civitai.json` или SHA256 → Civitai `by-hash`. Дубли `modelVersionId` не дублируются. Перед записью — `models.yaml.bak`. Неизвестные файлы (нет в Civitai) — только в отчёте.
+Берёт `{stem}.civitai.json` или SHA256 → Civitai `by-hash`. Также мержит очередь Assistent `/mnt/swarm_data/.gpu-rent-wanted-models.yaml` (модели, добавленные из вкладки Cards). Дубли `modelVersionId` не дублируются. Перед записью — `models.yaml.bak`. Неизвестные файлы (нет в Civitai) — только в отчёте.
 
 Локальный SwarmUI на `7801` не зеркалируем. Только дерево `./Models` приложения.
 
@@ -44,7 +44,7 @@ copy models.example.yaml models.yaml
 | Веса уже на VM с тем же SHA256 | Пропуск (нет «обновления») |
 | Веса изменились | Залить веса заново и актуальную метадату |
 
-Метадата — файлы рядом с тем же stem: `.json`, `.civitai.json`, `.swarm.json`, превью (`.preview.png` / `.png` / `.webp`), если лежат в той же папке. Веса без sidecar тоже можно залить (предупреждение в лог: «метадаты нет»). Метадату без весов **не** шлём отдельно.
+Метадата — файлы рядом с тем же stem: `.json`, `.civitai.json`, `.swarm.json`, `.assistent.json` (карточки рекомендаций Assistent), превью (`.preview.png` / `.png` / `.webp`), если лежат в той же папке. Веса без sidecar тоже можно залить (предупреждение в лог: «метадаты нет»). Метадату без весов **не** шлём отдельно.
 
 С диска в облаке **ничего не удаляем**. Civitai-seed и то, чего нет в `./Models`, остаётся.
 

src/gpu_rent/capture.py

@@ -405,6 +405,64 @@ def merge_extensions_yaml(
     return added, updated, skipped
 
 
+def merge_wanted_models_from_vm(cfg: Config, host: str, log: Log) -> int:
+    """Pull VM Assistent wanted queue into local models.yaml. Returns new url count."""
+    from gpu_rent.manifests import MODEL_DIRS
+    from gpu_rent.provision import DATA
+
+    remote = f"{DATA}/.gpu-rent-wanted-models.yaml"
+    raw = run_ssh(
+        cfg,
+        host,
+        f"test -f {remote!r} && cat {remote!r} || true".replace("'", '"')
+        if False
+        else f'test -f {remote} && cat {remote} || true',
+        check=False,
+        timeout=20,
+    ).strip()
+    if not raw:
+        return 0
+    try:
+        data = yaml.safe_load(raw) or {}
+    except Exception as exc:
+        log(f"wanted-models: bad yaml — {exc}")
+        return 0
+    if not isinstance(data, dict):
+        return 0
+    items: list[ModelCaptureItem] = []
+    for kind, rows in data.items():
+        if kind not in MODEL_DIRS or not isinstance(rows, list):
+            continue
+        for row in rows:
+            if not isinstance(row, dict):
+                continue
+            url = str(row.get("url") or "").strip()
+            if not url:
+                continue
+            vid = row.get("version_id")
+            try:
+                vid_i = int(vid) if vid not in (None, "", 0, "0") else extract_version_id(url)
+            except (TypeError, ValueError):
+                vid_i = extract_version_id(url)
+            items.append(
+                ModelCaptureItem(
+                    kind=kind,
+                    url=url,
+                    title=str(row.get("title") or Path(url).name),
+                    version_id=vid_i,
+                    source="wanted",
+                )
+            )
+    if not items:
+        return 0
+    added, skipped = merge_models_yaml(Path(cfg.models_manifest), items, dry_run=False)
+    if added:
+        log(f"wanted→models.yaml: +{len(added)} (skip {len(skipped)})")
+    elif skipped:
+        log(f"wanted→models.yaml: уже есть ({len(skipped)})")
+    return len(added)
+
+
 def capture_models(
     cfg: Config,
     host: str,
@@ -551,6 +609,18 @@ def capture_models(
         if hf_fallback:
             log(f"capture: Hugging Face fallback — {hf_fallback} файл(ов)")
 
+    # Also merge Assistent wanted-queue (models not yet on disk / seed list).
+    try:
+        from gpu_rent.provision import merge_wanted_models_from_vm
+
+        # merge writes directly; still collect for report when dry_run=False
+        if not dry_run:
+            n = merge_wanted_models_from_vm(cfg, host, log)
+            if n:
+                log(f"capture: wanted queue → models.yaml (+{n})")
+    except Exception as exc:
+        log(f"capture wanted: {exc}")
+
     added, skipped = merge_models_yaml(
         cfg.models_manifest, resolved, dry_run=dry_run
     )

src/gpu_rent/paths.py

@@ -48,6 +48,14 @@ def models_manifest_path() -> Path:
     return app_root() / "models.yaml"
 
 
+def assistent_personas_manifest_path() -> Path:
+    return app_root() / "assistent-personas.yaml"
+
+
+def assistent_personas_example_path() -> Path:
+    return app_root() / "assistent-personas.example.yaml"
+
+
 def ollama_models_manifest_path() -> Path:
     return app_root() / "ollama-models.yaml"
 

src/gpu_rent/provision.py

@@ -710,9 +710,139 @@ def seed_swarmui_api_keys(cfg: Config, host: str, log: Log) -> None:
         run_ssh(cfg, host, "rm -f /tmp/gpu-rent-swarm-api-keys.json", check=False)
 
 
+def seed_assistent_personas(cfg: Config, host: str, log: Log) -> None:
+    """Push local assistent-personas.yaml → VM Assistent/personas.yaml + personas.json."""
+    from gpu_rent.paths import assistent_personas_example_path, assistent_personas_manifest_path
+
+    local = Path(
+        (getattr(cfg, "assistent_personas_manifest", None) or assistent_personas_manifest_path())
+    )
+    if not local.is_file():
+        example = assistent_personas_example_path()
+        if example.is_file():
+            local = example
+        else:
+            log("assistent-personas: нет yaml — skip")
+            return
+    text = local.read_text(encoding="utf-8")
+    if not text.strip():
+        log("assistent-personas: пустой файл — skip")
+        return
+    remote_dir = f"{DATA}/Assistent"
+    remote_yaml = f"{remote_dir}/personas.yaml"
+    remote_json = f"{remote_dir}/personas.json"
+    run_ssh(cfg, host, f"mkdir -p {shlex.quote(remote_dir)}", check=False)
+    put_text(cfg, host, remote_yaml, text if text.endswith("\n") else text + "\n")
+    try:
+        import yaml
+
+        data = yaml.safe_load(text) or {}
+        put_text(
+            cfg,
+            host,
+            remote_json,
+            json.dumps(data, ensure_ascii=False, indent=2) + "\n",
+        )
+    except Exception as exc:
+        log(f"assistent-personas: json convert — {exc}")
+    log(f"assistent-personas → {remote_yaml}")
+
+
+def merge_wanted_models_from_vm(cfg: Config, host: str, log: Log) -> int:
+    """Pull VM wanted queue into local models.yaml. Returns count of new urls merged."""
+    from gpu_rent.capture import ModelCaptureItem, merge_models_yaml
+
+    remote = f"{DATA}/.gpu-rent-wanted-models.yaml"
+    raw = run_ssh(
+        cfg,
+        host,
+        f"test -f {shlex.quote(remote)} && cat {shlex.quote(remote)} || true",
+        check=False,
+        timeout=20,
+    ).strip()
+    if not raw:
+        return 0
+    try:
+        import yaml
+
+        data = yaml.safe_load(raw) or {}
+    except Exception as exc:
+        log(f"wanted-models: bad yaml — {exc}")
+        return 0
+    if not isinstance(data, dict):
+        return 0
+    items: list[ModelCaptureItem] = []
+    for kind, rows in data.items():
+        if kind not in MODEL_DIRS or not isinstance(rows, list):
+            continue
+        for row in rows:
+            if not isinstance(row, dict):
+                continue
+            url = str(row.get("url") or "").strip()
+            if not url:
+                continue
+            vid = row.get("version_id")
+            try:
+                vid_i = int(vid) if vid not in (None, "", 0, "0") else extract_version_id(url)
+            except (TypeError, ValueError):
+                vid_i = extract_version_id(url)
+            items.append(
+                ModelCaptureItem(
+                    kind=kind,
+                    url=url,
+                    title=str(row.get("title") or Path(url).name),
+                    version_id=vid_i,
+                    source="wanted",
+                )
+            )
+    if not items:
+        return 0
+    added, skipped = merge_models_yaml(Path(cfg.models_manifest), items, dry_run=False)
+    if added:
+        log(f"wanted→models.yaml: +{len(added)} (skip {len(skipped)})")
+    elif skipped:
+        log(f"wanted→models.yaml: уже есть ({len(skipped)})")
+    return len(added)
+
+
+def _place_wanted_cards(cfg: Config, host: str, jobs: list[dict], log: Log) -> None:
+    """Copy drafted .assistent.json from wanted-cards next to freshly seeded weights."""
+    cards_dir = f"{DATA}/.gpu-rent-wanted-cards"
+    exists = run_ssh(
+        cfg, host, f"test -d {shlex.quote(cards_dir)} && echo YES || true", check=False
+    ).strip()
+    if "YES" not in exists:
+        return
+    placed = 0
+    for job in jobs:
+        dest = str(job.get("dest") or "")
+        vid = job.get("version_id")
+        if not dest or vid in (None, "", 0, "0"):
+            continue
+        stem = Path(dest).stem
+        card_src = f"{cards_dir}/{vid}.assistent.json"
+        card_dst = str(Path(dest).with_name(f"{stem}.assistent.json"))
+        out = run_ssh(
+            cfg,
+            host,
+            f"if test -f {shlex.quote(card_src)} && test -f {shlex.quote(dest)}; then "
+            f"cp -n {shlex.quote(card_src)} {shlex.quote(card_dst)} && echo PLACED; fi",
+            check=False,
+        ).strip()
+        if "PLACED" in out:
+            placed += 1
+    if placed:
+        log(f"wanted-cards: положил {placed} .assistent.json рядом с весами")
+
+
 def seed_civitai(cfg: Config, host: str, log: Log) -> None:
     from gpu_rent.huggingface import is_huggingface_url
 
+    try:
+        merge_wanted_models_from_vm(cfg, host, log)
+    except Exception as exc:
+        log(f"wanted-models merge: {exc}")
+
     entries = parse_models(cfg.models_manifest)
     if not entries:
         log("model-seed пропущен: манифест пуст — дефолт SwarmUI")
@@ -775,6 +905,7 @@ def seed_civitai(cfg: Config, host: str, log: Log) -> None:
                     "url": url,
                     "sha256": "",
                     "auth": "hf",
+                    "version_id": None,
                     "sidecars": {
                         f"{stem}.swarm.json": json.dumps(swarm, ensure_ascii=False, indent=2),
                     },
@@ -825,6 +956,7 @@ def seed_civitai(cfg: Config, host: str, log: Log) -> None:
             "url": _download_url(api_host, vid, info),
             "sha256": sha,
             "auth": "civitai",
+            "version_id": int(vid),
             "sidecars": {
                 f"{stem}.civitai.json": civitai_json,
                 f"{stem}.swarm.json": json.dumps(swarm, ensure_ascii=False, indent=2),
@@ -903,6 +1035,10 @@ def seed_civitai(cfg: Config, host: str, log: Log) -> None:
         timeout=7200,
         log=log,
     )
+    try:
+        _place_wanted_cards(cfg, host, jobs, log)
+    except Exception as exc:
+        log(f"wanted-cards: {exc}")
     put_text(
         cfg,
         host,
@@ -1159,6 +1295,10 @@ def provision_vm(
             except GpuRentError as exc:
                 log(f"autocomplete: {exc}")
             seed_civitai(cfg, host, log)
+            try:
+                seed_assistent_personas(cfg, host, log)
+            except Exception as exc:
+                log(f"assistent-personas: {exc}")
             push_tree(cfg, host, cfg.local_models_dir, f"{DATA}/Models", log, models=True)
             push_tree(
                 cfg, host, cfg.local_wildcards_dir, f"{DATA}/Data/Wildcards", log, models=False

src/gpu_rent/setup_wizard.py

@@ -16,6 +16,8 @@ from gpu_rent.llm_runtime import (
 )
 from gpu_rent.paths import (
     app_root,
+    assistent_personas_example_path,
+    assistent_personas_manifest_path,
     env_path,
     extensions_manifest_path,
     models_manifest_path,
@@ -54,6 +56,12 @@ def run_setup(
 
     _copy_if_missing(root / "env.example", env_path(), ".env", log)
     _copy_if_missing(root / "models.example.yaml", models_manifest_path(), "models.yaml", log)
+    _copy_if_missing(
+        assistent_personas_example_path(),
+        assistent_personas_manifest_path(),
+        "assistent-personas.yaml",
+        log,
+    )
     _copy_if_missing(
         root / "extensions.example.yaml", extensions_manifest_path(), "extensions.yaml", log
     )