oh-my-opencode-kikokikok 2.15.2 → 2.15.4

package/dist/cli/index.js

@@ -2253,7 +2253,7 @@ var require_picocolors = __commonJS((exports, module) => {
 var require_package = __commonJS((exports, module) => {
   module.exports = {
     name: "oh-my-opencode-kikokikok",
-    version: "2.15.2",
+    version: "2.15.4",
     description: "OpenCode plugin - custom agents (oracle, librarian) and enhanced features",
     main: "dist/index.js",
     types: "dist/index.d.ts",

package/dist/features/letta-memory/adapter.d.ts

@@ -3,7 +3,10 @@ export declare class LettaAdapter {
     private config;
     private endpoint;
     private agentCache;
+    private detectedEmbeddingModel;
+    private modelDetectionPromise;
     constructor(config: LettaConfig);
+    private detectEmbeddingModel;
     add(input: AddMemoryInput): Promise<Memory>;
     search(input: SearchMemoryInput): Promise<MemorySearchResult[]>;
     get(id: string): Promise<Memory | null>;
@@ -13,6 +16,8 @@ export declare class LettaAdapter {
     getStats(): Promise<MemoryStats>;
     isAvailable(): Promise<boolean>;
     private getOrCreateAgent;
+    private agentNeedsEmbeddingUpdate;
+    private recreateAgentWithCorrectEmbedding;
     private getAgent;
     private listAgents;
     private request;

package/dist/features/letta-memory/types.d.ts

@@ -26,8 +26,25 @@ export interface LettaConfig {
     agentPrefix?: string;
     /** LLM model for agent (e.g., "openai/gpt-4.1") */
     llmModel?: string;
-    /** Embedding model for semantic search */
+    /**
+     * Embedding model for semantic search.
+     * Use format "provider/model" e.g.:
+     * - "openai/text-embedding-3-large" (best quality, 3072 dimensions)
+     * - "openai/text-embedding-3-small" (good balance, 1536 dimensions)
+     * - "openai/text-embedding-ada-002" (legacy)
+     * - "letta/letta-free" (default, requires Letta cloud auth)
+     *
+     * If not set, auto-detects local proxy models with "openai" provider.
+     */
     embeddingModel?: string;
+    /**
+     * Preferred embedding model for auto-detection.
+     * When embeddingModel is not set and multiple proxy models are available,
+     * this determines which one to prefer. Partial match on model name.
+     * Default: "text-embedding-3-small"
+     * Set to "text-embedding-3-large" for better quality.
+     */
+    preferredEmbeddingModel?: string;
     /** Auto-rehydrate memories on session start */
     autoRehydrate?: boolean;
     /** Layers to rehydrate */
@@ -43,6 +60,12 @@ export interface LettaAgent {
     metadata?: Record<string, unknown>;
     memory_blocks?: LettaBlock[];
     tools?: string[];
+    embedding?: string;
+    embedding_config?: {
+        handle?: string;
+        embedding_model?: string;
+        embedding_endpoint?: string;
+    };
 }
 /** Letta Memory Block (Core Memory - always in-context) */
 export interface LettaBlock {
@@ -131,9 +154,16 @@ export interface LettaApiResponse<T> {
     error?: string;
     message?: string;
 }
+/** Letta search result item (different from passage) */
+export interface LettaSearchResultItem {
+    timestamp: string;
+    content: string;
+    tags?: string[];
+}
+/** Letta search response */
 export interface LettaSearchResponse {
-    passages: LettaPassage[];
-    total?: number;
+    results: LettaSearchResultItem[];
+    count: number;
 }
 export interface LettaAgentListResponse {
     agents: LettaAgent[];

package/dist/index.js

@@ -43788,10 +43788,46 @@ class LettaAdapter {
   config;
   endpoint;
   agentCache = new Map;
+  detectedEmbeddingModel = null;
+  modelDetectionPromise = null;
   constructor(config3) {
     this.config = config3;
     this.endpoint = config3.endpoint ?? DEFAULT_ENDPOINT2;
   }
+  async detectEmbeddingModel() {
+    if (this.config.embeddingModel) {
+      return this.config.embeddingModel;
+    }
+    if (this.detectedEmbeddingModel) {
+      return this.detectedEmbeddingModel;
+    }
+    if (this.modelDetectionPromise) {
+      await this.modelDetectionPromise;
+      return this.detectedEmbeddingModel ?? DEFAULT_EMBEDDING_MODEL;
+    }
+    this.modelDetectionPromise = (async () => {
+      try {
+        const response2 = await fetch(`${this.endpoint}/v1/models`, {
+          method: "GET",
+          redirect: "follow",
+          signal: AbortSignal.timeout(1e4)
+        });
+        if (!response2.ok) {
+          return;
+        }
+        const models = await response2.json();
+        const proxyEmbeddingModels = models.filter((m) => m.name.includes("embedding") && m.model_endpoint.includes("host.docker.internal") && m.provider_name === "openai");
+        if (proxyEmbeddingModels.length > 0) {
+          const preferredName = this.config.preferredEmbeddingModel ?? "text-embedding-3-small";
+          const preferred = proxyEmbeddingModels.find((m) => m.name.includes(preferredName));
+          const model = preferred ?? proxyEmbeddingModels[0];
+          this.detectedEmbeddingModel = `openai/${model.name}`;
+        }
+      } catch {}
+    })();
+    await this.modelDetectionPromise;
+    return this.detectedEmbeddingModel ?? DEFAULT_EMBEDDING_MODEL;
+  }
   async add(input) {
     if (!this.config.enabled) {
       throw new Error("Letta is not enabled");
@@ -43809,7 +43845,8 @@ class LettaAdapter {
         tags
       })
     });
-    const passage = await response2.json();
+    const data = await response2.json();
+    const passage = Array.isArray(data) ? data[0] : data;
     return this.passageToMemory(passage, input.layer, agent.id);
   }
   async search(input) {
@@ -43833,8 +43870,17 @@ class LettaAdapter {
         }
         const response2 = await this.request(`/v1/agents/${agent.id}/archival-memory/search?${params.toString()}`, { method: "GET" });
         const data = await response2.json();
-        const layerResults = data.map((passage, index) => ({
-          memory: this.passageToMemory(passage, layer, agent.id),
+        const searchResults = data.results ?? [];
+        const layerResults = searchResults.map((item, index) => ({
+          memory: {
+            id: `search-${agent.id}-${item.timestamp}-${index}`,
+            content: item.content,
+            layer,
+            metadata: item.tags ? { tags: item.tags } : undefined,
+            createdAt: item.timestamp,
+            source: "passage",
+            agentId: agent.id
+          },
           score: 1 - index * 0.05
         }));
         results.push(...layerResults);
@@ -43956,15 +44002,60 @@ class LettaAdapter {
   }
   async getOrCreateAgent(layer) {
     const existing = await this.getAgent(layer);
-    if (existing)
+    if (existing) {
+      const needsUpdate = await this.agentNeedsEmbeddingUpdate(existing);
+      if (needsUpdate) {
+        return this.recreateAgentWithCorrectEmbedding(existing, layer);
+      }
       return existing;
+    }
+    const embeddingModel = await this.detectEmbeddingModel();
+    const agentName = this.getAgentName(layer);
+    const response2 = await this.request("/v1/agents", {
+      method: "POST",
+      body: JSON.stringify({
+        name: agentName,
+        model: this.config.llmModel ?? DEFAULT_LLM_MODEL,
+        embedding: embeddingModel,
+        memory_blocks: [
+          { label: "persona", value: `OpenCode memory agent for ${layer} layer` },
+          { label: "human", value: this.getUserId(layer) }
+        ],
+        metadata: {
+          layer,
+          user_id: this.getUserId(layer),
+          created_by: "oh-my-opencode"
+        }
+      })
+    });
+    const agent = await response2.json();
+    this.agentCache.set(layer, agent);
+    return agent;
+  }
+  async agentNeedsEmbeddingUpdate(agent) {
+    if (this.config.embeddingModel) {
+      return false;
+    }
+    const embeddingHandle = agent.embedding_config?.handle ?? agent.embedding;
+    if (!embeddingHandle)
+      return false;
+    if (embeddingHandle === "letta/letta-free") {
+      const detected = await this.detectEmbeddingModel();
+      return detected !== "letta/letta-free";
+    }
+    return false;
+  }
+  async recreateAgentWithCorrectEmbedding(existingAgent, layer) {
+    await this.request(`/v1/agents/${existingAgent.id}`, { method: "DELETE" }).catch(() => {});
+    this.agentCache.delete(layer);
+    const embeddingModel = await this.detectEmbeddingModel();
     const agentName = this.getAgentName(layer);
     const response2 = await this.request("/v1/agents", {
       method: "POST",
       body: JSON.stringify({
         name: agentName,
         model: this.config.llmModel ?? DEFAULT_LLM_MODEL,
-        embedding: this.config.embeddingModel ?? DEFAULT_EMBEDDING_MODEL,
+        embedding: embeddingModel,
         memory_blocks: [
           { label: "persona", value: `OpenCode memory agent for ${layer} layer` },
           { label: "human", value: this.getUserId(layer) }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "oh-my-opencode-kikokikok",
-  "version": "2.15.2",
+  "version": "2.15.4",
   "description": "OpenCode plugin - custom agents (oracle, librarian) and enhanced features",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",