@massa-ai/tools-api 1.59.0 → 1.60.0

package/dist/index.js

@@ -487,13 +487,33 @@ var init_inference_providers = __esm(() => {
       },
       knownDimensions: {
         "text-embedding-nomic-embed-text-v1.5": 768,
-        "text-embedding-qwen3-embedding-0.6b": 1024
+        "text-embedding-qwen3-embedding-0.6b": 1024,
+        "qwen3-embedding-0.6b-dwq": 1024
       },
       defaultModels: {
         embedding: "text-embedding-qwen3-embedding-0.6b",
         instruct: "qwen3-vl-8b-instruct",
         coding: "qwen2.5-coder-7b-instruct"
       },
+      mlxModels: {
+        embedding: {
+          repo: "mlx-community/Qwen3-Embedding-0.6B-4bit-DWQ",
+          model: "qwen3-embedding-0.6b-dwq"
+        },
+        instruct: {
+          repo: "mlx-community/Qwen3-VL-8B-Instruct-4bit",
+          model: "qwen3-vl-8b-instruct"
+        },
+        coding: {
+          repo: "mlx-community/Qwen2.5-Coder-7B-Instruct-4bit",
+          model: "qwen2.5-coder-7b-instruct"
+        }
+      },
+      ggufRepos: {
+        embedding: "Qwen/Qwen3-Embedding-0.6B-GGUF",
+        instruct: "lmstudio-community/Qwen3-VL-8B-Instruct-GGUF",
+        coding: "lmstudio-community/Qwen2.5-Coder-7B-Instruct-GGUF"
+      },
       appliesContextPerRequest: false,
       embedBatchSize: 64,
       supportsOllamaVersionProbe: false,
@@ -128211,6 +128231,7 @@ class PgObservationStore {
   mirror = new Map;
   hydrated = false;
   hydrating = null;
+  inflight = new Map;
   hydrateFailedAt = 0;
   static HYDRATE_RETRY_MS = 30000;
   getClient() {
@@ -128262,46 +128283,53 @@ class PgObservationStore {
     const cachedCanonical = getProjectIdentityAliasResolver().resolveCached(obs.projectId);
     this.mirror.set(obs.id, cachedCanonical && cachedCanonical !== obs.projectId ? { ...obs, projectId: cachedCanonical } : obs);
     this.ensureHydrated();
-    (async () => {
-      try {
-        const prisma2 = this.getClient();
-        const canonicalProjectId = await getProjectIdentityAliasResolver().resolve(obs.projectId);
-        if (canonicalProjectId !== obs.projectId) {
-          this.mirror.set(obs.id, { ...obs, projectId: canonicalProjectId });
-        }
-        await prisma2.$executeRaw`
-          INSERT INTO observations (
-            id, project_id, session_id, source, category, payload_json, importance, created_at, agent_id, attribution_source
-          ) VALUES (
-            ${obs.id},
-            ${canonicalProjectId},
-            ${obs.sessionId},
-            ${obs.source},
-            ${obs.category ?? null},
-            ${obs.payloadJson},
-            ${obs.importance},
-            ${obs.createdAt}::bigint,
-            ${obs.agentId ?? null},
-            ${obs.attributionSource ?? null}
-          )
-          ON CONFLICT (id) DO UPDATE SET
-            project_id       = EXCLUDED.project_id,
-            session_id       = EXCLUDED.session_id,
-            source           = EXCLUDED.source,
-            category         = EXCLUDED.category,
-            payload_json     = EXCLUDED.payload_json,
-            importance       = EXCLUDED.importance,
-            created_at       = EXCLUDED.created_at,
-            agent_id         = EXCLUDED.agent_id,
-            attribution_source = EXCLUDED.attribution_source
-        `;
-      } catch (e) {
-        logger.warn("PgObservationStore.insert failed (best-effort)", {
-          id: obs.id,
-          error: e.message
-        });
+    this.chainWrite(obs.id, async () => {
+      const prisma2 = this.getClient();
+      const canonicalProjectId = await getProjectIdentityAliasResolver().resolve(obs.projectId);
+      if (canonicalProjectId !== obs.projectId) {
+        this.mirror.set(obs.id, { ...obs, projectId: canonicalProjectId });
       }
-    })();
+      await prisma2.$executeRaw`
+        INSERT INTO observations (
+          id, project_id, session_id, source, category, payload_json, importance, created_at, agent_id, attribution_source
+        ) VALUES (
+          ${obs.id},
+          ${canonicalProjectId},
+          ${obs.sessionId},
+          ${obs.source},
+          ${obs.category ?? null},
+          ${obs.payloadJson},
+          ${obs.importance},
+          ${obs.createdAt}::bigint,
+          ${obs.agentId ?? null},
+          ${obs.attributionSource ?? null}
+        )
+        ON CONFLICT (id) DO UPDATE SET
+          project_id       = EXCLUDED.project_id,
+          session_id       = EXCLUDED.session_id,
+          source           = EXCLUDED.source,
+          category         = EXCLUDED.category,
+          payload_json     = EXCLUDED.payload_json,
+          importance       = EXCLUDED.importance,
+          created_at       = EXCLUDED.created_at,
+          agent_id         = EXCLUDED.agent_id,
+          attribution_source = EXCLUDED.attribution_source
+      `;
+    });
+  }
+  chainWrite(key, fn) {
+    const prev = this.inflight.get(key) ?? Promise.resolve();
+    const next = prev.then(fn).catch((e) => {
+      logger.warn("PgObservationStore.insert failed (best-effort)", {
+        id: key,
+        error: e.message
+      });
+    });
+    this.inflight.set(key, next);
+    next.then(() => {
+      if (this.inflight.get(key) === next)
+        this.inflight.delete(key);
+    });
   }
   listRecent(projectId, limit) {
     this.ensureHydrated();
@@ -128328,6 +128356,9 @@ class PgObservationStore {
     await this.ensureHydrated();
   }
   async __drain() {
+    const pending = Array.from(this.inflight.values());
+    if (pending.length > 0)
+      await Promise.allSettled(pending);
     await new Promise((r2) => setTimeout(r2, 10));
   }
 }
@@ -183870,9 +183901,14 @@ var profileRoutes = new Elysia({ prefix: "/api/v1/profiles" }).get("/", ({ query
 
 // src/routes/config.ts
 init_dist();
+init_inference_providers();
 var CONFIG_DETAIL = {
   tags: ["config"]
 };
+function defaultEmbedBatchSize(provider) {
+  const spec = typeof provider === "string" && LOCAL_INFERENCE_IDS.includes(provider) ? INFERENCE_PROVIDERS[provider] : INFERENCE_PROVIDERS.ollama;
+  return spec.embedBatchSize;
+}
 var SENSITIVE_FIELDS = {
   database: ["url"],
   embedding: ["apiKey"],
@@ -183897,7 +183933,15 @@ var configRoutes = new Elysia({ prefix: "/api/v1/config" }).get("/", ({ set: set
   const config3 = loadConfig();
   const masked = maskSensitive(config3);
   const restart = restartNeededSections(config3);
-  const defaults2 = maskSensitive(defaultMassaAiConfig);
+  const shipped = maskSensitive(defaultMassaAiConfig);
+  const defaults2 = {
+    ...shipped,
+    embedding: {
+      ...shipped.embedding,
+      contextWindow: INFERENCE_ROLE_DEFAULTS.embedding.contextWindow,
+      batchSize: defaultEmbedBatchSize(config3.embedding?.provider)
+    }
+  };
   set3.status = 200;
   return {
     success: true,
@@ -183907,7 +183951,7 @@ var configRoutes = new Elysia({ prefix: "/api/v1/config" }).get("/", ({ set: set
   detail: {
     ...CONFIG_DETAIL,
     summary: "Get current config with sensitive fields masked",
-    description: "Returns the current config.json with security.apiKey, llm.apiKey, embedding.apiKey, and database.url masked to '***'. Includes restartNeededSections \u2014 the subset of [database, embedding, llm, security] present in the config \u2014 and defaults, the shipped default config (also masked) the Config tab falls back to for any field the persisted file omits."
+    description: "Returns the current config.json with security.apiKey, llm.apiKey, embedding.apiKey, and database.url masked to '***'. Includes restartNeededSections \u2014 the subset of [database, embedding, llm, security] present in the config \u2014 and defaults, the shipped default config (also masked) the Config tab falls back to for any field the persisted file omits. defaults.embedding.contextWindow and defaults.embedding.batchSize are derived rather than shipped: they come from the role table and the resolved provider's seam entry, because defaultMassaAiConfig deliberately leaves both unset (PDM-12)."
   }
 }).get("/reveal", ({ query, set: set3 }) => {
   const section = query.section;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@massa-ai/tools-api",
-  "version": "1.59.0",
+  "version": "1.60.0",
   "author": "luizgmassa",
   "description": "massa-ai REST API server - Semantic code search, memory, and context compression",
   "type": "module",
@@ -21,8 +21,8 @@
     "test": "bun scripts/run-tests-isolated.ts"
   },
   "dependencies": {
-    "@massa-ai/core": "^1.59.0",
-    "@massa-ai/shared": "^1.59.0",
+    "@massa-ai/core": "^1.60.0",
+    "@massa-ai/shared": "^1.60.0",
     "elysia": "^1.2.25",
     "@elysiajs/swagger": "^1.2.0",
     "@elysiajs/cors": "^1.2.0",