Ship Assistent disk persist, park LLM, and memory UI cleanup.

Split the extension into partials, persist chats on the data volume, park/warm the chat model around Generate, and drop dual raw/persona dump paths.

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

AssistentMemoryApi.cs

@@ -173,6 +173,75 @@ public partial class SwarmAssistentExtension
         }
     }
 
+    public async Task<JObject> AssistentSearchMemory(Session session, string query, string kind = null, int limit = 10, string persona = null, string baseUrl = null, string embed_model = null)
+    {
+        if (Memory is null)
+        {
+            return new JObject { ["error"] = "memory not ready" };
+        }
+        query = (query ?? "").Trim();
+        if (query.Length < 2)
+        {
+            return new JObject { ["error"] = "query required" };
+        }
+        string pid = AssistentConfig.SafeId(persona) ?? Config.DefaultPersonaId();
+        try
+        {
+            JArray rows = await Memory.SearchAsync(MemoryBaseUrl(baseUrl), query, kind, limit, MemoryEmbedModel(embed_model), Config.PersonaExtendsChain(pid));
+            return new JObject { ["success"] = true, ["query"] = query, ["kind"] = kind ?? "", ["memories"] = rows };
+        }
+        catch (Exception ex)
+        {
+            return new JObject { ["error"] = $"memory search: {ex.Message}" };
+        }
+    }
+
+    public async Task<JObject> AssistentGetMemory(Session session, string kind, string key, string persona = null)
+    {
+        await Task.CompletedTask;
+        if (Memory is null)
+        {
+            return new JObject { ["error"] = "memory not ready" };
+        }
+        string pid = AssistentConfig.SafeId(persona) ?? Config.DefaultPersonaId();
+        JObject row = Memory.Get(kind, key, Config.PersonaExtendsChain(pid));
+        if (row is null)
+        {
+            return new JObject { ["success"] = true, ["missing"] = true, ["kind"] = kind, ["key"] = key };
+        }
+        return new JObject { ["success"] = true, ["memory"] = row };
+    }
+
+    public async Task<JObject> AssistentLookupTags(Session session, string query, int limit = 20)
+    {
+        await Task.CompletedTask;
+        if (Memory is null)
+        {
+            return new JObject { ["error"] = "memory not ready" };
+        }
+        query = (query ?? "").Trim();
+        if (query.Length < 1)
+        {
+            return new JObject { ["error"] = "query required" };
+        }
+        try
+        {
+            JArray tags = Memory.LookupTags(query, limit);
+            return new JObject
+            {
+                ["success"] = true,
+                ["query"] = query,
+                ["tags"] = tags,
+                ["indexed"] = Memory.TagCount(),
+                ["csv"] = Memory.FindAutocompleteCsv() ?? "",
+            };
+        }
+        catch (Exception ex)
+        {
+            return new JObject { ["error"] = $"tag lookup: {ex.Message}" };
+        }
+    }
+
     /// <summary>The gpu-rent wanted queue (models pending the next <c>up</c>) — count + entries.</summary>
     public async Task<JObject> AssistentListWanted(Session session)
     {