@massa-ai/opencode-plugin 1.58.0 → 1.60.0

package/dist/config-cli.js

@@ -400,6 +400,11 @@ function configDir(app) {
 var init_xdg = () => {};
 
 // ../../packages/shared/dist/config/embedding-dimensions.js
+function knownEmbeddingDimensions(model) {
+  if (!model)
+    return;
+  return KNOWN_EMBEDDING_DIMENSIONS[model.trim()];
+}
 var KNOWN_EMBEDDING_DIMENSIONS;
 var init_embedding_dimensions = __esm(() => {
   KNOWN_EMBEDDING_DIMENSIONS = {
@@ -427,10 +432,33 @@ function parseLmStudioModelList(body) {
     return null;
   return data.map((entry) => entry.id).filter((id) => typeof id === "string");
 }
-var LOCAL_INFERENCE_IDS, INFERENCE_PROVIDERS;
+function deriveInferenceBaseUrls(providerId, explicitBaseUrl) {
+  const spec = INFERENCE_PROVIDERS[providerId];
+  if (!explicitBaseUrl) {
+    return {
+      embeddingBaseUrl: spec.defaultEmbeddingBaseUrl,
+      llmBaseUrl: spec.defaultLlmBaseUrl
+    };
+  }
+  let end = explicitBaseUrl.length;
+  while (end > 0 && explicitBaseUrl.charCodeAt(end - 1) === SLASH)
+    end--;
+  const base = explicitBaseUrl.slice(0, end);
+  const suffix = spec.defaultLlmBaseUrl.startsWith(spec.defaultEmbeddingBaseUrl) ? spec.defaultLlmBaseUrl.slice(spec.defaultEmbeddingBaseUrl.length) : "";
+  return {
+    embeddingBaseUrl: base,
+    llmBaseUrl: `${base}${suffix}`
+  };
+}
+var LOCAL_INFERENCE_IDS, INFERENCE_ROLE_DEFAULTS, INFERENCE_PROVIDERS, SLASH = 47;
 var init_inference_providers = __esm(() => {
   init_embedding_dimensions();
   LOCAL_INFERENCE_IDS = ["ollama", "lmstudio"];
+  INFERENCE_ROLE_DEFAULTS = {
+    embedding: { contextWindow: 8192 },
+    instruct: { contextWindow: 16384, temperature: 0.2 },
+    coding: { contextWindow: 32768, temperature: 0 }
+  };
   INFERENCE_PROVIDERS = {
     ollama: {
       id: "ollama",
@@ -442,6 +470,13 @@ var init_inference_providers = __esm(() => {
         dimensions: "OLLAMA_EMBEDDING_DIMENSIONS"
       },
       knownDimensions: KNOWN_EMBEDDING_DIMENSIONS,
+      defaultModels: {
+        embedding: "qwen3-embedding:0.6b",
+        instruct: "qwen3-vl:8b",
+        coding: "qwen2.5-coder:7b"
+      },
+      appliesContextPerRequest: true,
+      embedBatchSize: 64,
       supportsOllamaVersionProbe: true,
       injectsDisableThink: true,
       requiresChatCompletionsApi: false,
@@ -457,8 +492,36 @@ var init_inference_providers = __esm(() => {
         dimensions: "LMSTUDIO_EMBEDDING_DIMENSIONS"
       },
       knownDimensions: {
-        "text-embedding-nomic-embed-text-v1.5": 768
+        "text-embedding-nomic-embed-text-v1.5": 768,
+        "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,
       injectsDisableThink: false,
       requiresChatCompletionsApi: true,
@@ -473,6 +536,7 @@ var API_PROVIDER_IDS, EMBEDDING_PROVIDER_IDS, SCHEDULER_JOB_KINDS, MAX_MATCH_WOR
 var init_massa_ai_config = __esm(() => {
   init_xdg();
   init_inference_providers();
+  init_embedding_dimensions();
   API_PROVIDER_IDS = ["mistral", "openai", "google", "cohere"];
   EMBEDDING_PROVIDER_IDS = [
     ...LOCAL_INFERENCE_IDS,
@@ -539,9 +603,9 @@ var init_massa_ai_config = __esm(() => {
     },
     embedding: {
       provider: "ollama",
-      model: "qwen3-embedding:4b",
+      model: INFERENCE_PROVIDERS.ollama.defaultModels.embedding,
       baseURL: "http://localhost:11434",
-      dimensions: 2560
+      dimensions: knownEmbeddingDimensions(INFERENCE_PROVIDERS.ollama.defaultModels.embedding) ?? 768
     },
     compression: {
       defaultStrategy: "code_structure",
@@ -580,12 +644,15 @@ var init_massa_ai_config = __esm(() => {
       enabled: false,
       baseUrl: "http://localhost:11434/v1",
       apiKey: "ollama",
-      model: "qwen2.5:7b-instruct",
-      codeModel: "qwen2.5-coder:7b",
+      model: INFERENCE_PROVIDERS.ollama.defaultModels.instruct,
+      codeModel: INFERENCE_PROVIDERS.ollama.defaultModels.coding,
       temperature: 0.2,
       maxOutputTokens: 8000,
       timeoutMs: 90000,
-      disableThink: true
+      disableThink: true,
+      contextWindow: INFERENCE_ROLE_DEFAULTS.instruct.contextWindow,
+      codeContextWindow: INFERENCE_ROLE_DEFAULTS.coding.contextWindow,
+      codeTemperature: INFERENCE_ROLE_DEFAULTS.coding.temperature
     },
     memory: {
       decay: {
@@ -986,6 +1053,7 @@ try {
 // ../../packages/shared/dist/config/index.js
 init_config_loader();
 init_massa_ai_config();
+init_inference_providers();
 init_massa_ai_config();
 init_massa_ai_config();
 init_config_loader();
@@ -1003,8 +1071,6 @@ init_config_loader();
 
 // ../../packages/shared/dist/config/index.js
 init_embedding_dimensions();
-var DEFAULT_LLM_MODEL = "qwen2.5:7b-instruct";
-var DEFAULT_LLM_CODE_MODEL = "qwen2.5-coder:7b";
 function envNum(key, fallback) {
   const s = process.env[key];
   if (s === undefined || s === "")
@@ -1177,6 +1243,15 @@ var DEFAULT_ALLOWED_EXTENSIONS = [
   ".hs"
 ];
 var fileConfig = loadConfigSafe();
+function activeInferenceProviderId() {
+  const providerId = fileConfig.embedding?.provider;
+  if (providerId && LOCAL_INFERENCE_IDS.includes(providerId)) {
+    return providerId;
+  }
+  return "ollama";
+}
+var DEFAULT_LLM_MODEL = INFERENCE_PROVIDERS[activeInferenceProviderId()].defaultModels.instruct;
+var DEFAULT_LLM_CODE_MODEL = INFERENCE_PROVIDERS[activeInferenceProviderId()].defaultModels.coding;
 var fileCacheL1Bytes = fileConfig.cache?.l1MaxSizeMB ? fileConfig.cache.l1MaxSizeMB * 1024 * 1024 : undefined;
 var fileCacheL2Bytes = fileConfig.cache?.l2MaxSizeMB ? fileConfig.cache.l2MaxSizeMB * 1024 * 1024 : undefined;
 var resolvedDataDir = getGlobalDataDir();
@@ -1219,7 +1294,10 @@ var defaultConfig = {
     temperature: envNum("MASSA_AI_LLM_TEMPERATURE", fileConfig.llm?.temperature ?? 0.2),
     maxOutputTokens: envNum("MASSA_AI_LLM_MAX_OUTPUT_TOKENS", fileConfig.llm?.maxOutputTokens ?? 8000),
     timeoutMs: envNum("MASSA_AI_LLM_TIMEOUT_MS", fileConfig.llm?.timeoutMs ?? 90000),
-    disableThink: envBool("MASSA_AI_LLM_DISABLE_THINK", fileConfig.llm?.disableThink ?? true)
+    disableThink: envBool("MASSA_AI_LLM_DISABLE_THINK", fileConfig.llm?.disableThink ?? true),
+    contextWindow: fileConfig.llm?.contextWindow ?? INFERENCE_ROLE_DEFAULTS.instruct.contextWindow,
+    codeContextWindow: fileConfig.llm?.codeContextWindow ?? INFERENCE_ROLE_DEFAULTS.coding.contextWindow,
+    codeTemperature: envNum("MASSA_AI_LLM_CODE_TEMPERATURE", fileConfig.llm?.codeTemperature ?? INFERENCE_ROLE_DEFAULTS.coding.temperature)
   },
   memory: {
     decay: {
@@ -3420,7 +3498,7 @@ Examples:
   massa-ai-config init
   massa-ai-config init --lmstudio
   massa-ai-config init --mistral your-api-key
-  massa-ai-config use ollama --model qwen3-embedding:4b
+  massa-ai-config use ollama --model qwen3-embedding:0.6b
   massa-ai-config use lmstudio
   massa-ai-config use mistral --api-key your-key
   massa-ai-config set embedding.dimensions 1024
@@ -3502,7 +3580,7 @@ async function runCli(argv) {
         console.log("\u2713 Configured for OpenAI embeddings");
       } else if (options.lmstudio) {
         const config2 = loadConfig();
-        const model = "text-embedding-nomic-embed-text-v1.5";
+        const model = INFERENCE_PROVIDERS.lmstudio.defaultModels.embedding;
         config2.embedding = {
           provider: "lmstudio",
           model,
@@ -3510,6 +3588,8 @@ async function runCli(argv) {
           dimensions: INFERENCE_PROVIDERS.lmstudio.knownDimensions[model] ?? 768
         };
         config2.llm.baseUrl = INFERENCE_PROVIDERS.lmstudio.defaultLlmBaseUrl;
+        config2.llm.model = INFERENCE_PROVIDERS.lmstudio.defaultModels.instruct;
+        config2.llm.codeModel = INFERENCE_PROVIDERS.lmstudio.defaultModels.coding;
         saveConfig(config2);
         console.log("\u2713 Configured for LM Studio (local) embeddings");
       } else {
@@ -3562,21 +3642,29 @@ Using defaults:`);
       }
       const config2 = loadConfig();
       if (provider === "ollama") {
+        const model = options.model || INFERENCE_PROVIDERS.ollama.defaultModels.embedding;
+        const urls = deriveInferenceBaseUrls("ollama", options["base-url"]);
         config2.embedding = {
           provider: "ollama",
-          model: options.model || "qwen3-embedding:4b",
-          baseURL: options["base-url"] || "http://localhost:11434",
-          dimensions: 2560
+          model,
+          baseURL: urls.embeddingBaseUrl,
+          dimensions: knownEmbeddingDimensions(model) ?? 768
         };
+        config2.llm.baseUrl = urls.llmBaseUrl;
+        config2.llm.model = INFERENCE_PROVIDERS.ollama.defaultModels.instruct;
+        config2.llm.codeModel = INFERENCE_PROVIDERS.ollama.defaultModels.coding;
       } else if (provider === "lmstudio") {
-        const model = options.model || "text-embedding-nomic-embed-text-v1.5";
+        const model = options.model || INFERENCE_PROVIDERS.lmstudio.defaultModels.embedding;
+        const urls = deriveInferenceBaseUrls("lmstudio", options["base-url"]);
         config2.embedding = {
           provider: "lmstudio",
           model,
-          baseURL: options["base-url"] || INFERENCE_PROVIDERS.lmstudio.defaultEmbeddingBaseUrl,
+          baseURL: urls.embeddingBaseUrl,
           dimensions: INFERENCE_PROVIDERS.lmstudio.knownDimensions[model] ?? 768
         };
-        config2.llm.baseUrl = options["base-url"] || INFERENCE_PROVIDERS.lmstudio.defaultLlmBaseUrl;
+        config2.llm.baseUrl = urls.llmBaseUrl;
+        config2.llm.model = INFERENCE_PROVIDERS.lmstudio.defaultModels.instruct;
+        config2.llm.codeModel = INFERENCE_PROVIDERS.lmstudio.defaultModels.coding;
       } else if (provider === "mistral") {
         if (!options["api-key"]) {
           console.error("Error: --api-key required for Mistral");

package/dist/index.js

@@ -399,6 +399,11 @@ function configDir(app) {
 var init_xdg = () => {};
 
 // ../../packages/shared/dist/config/embedding-dimensions.js
+function knownEmbeddingDimensions(model) {
+  if (!model)
+    return;
+  return KNOWN_EMBEDDING_DIMENSIONS[model.trim()];
+}
 var KNOWN_EMBEDDING_DIMENSIONS;
 var init_embedding_dimensions = __esm(() => {
   KNOWN_EMBEDDING_DIMENSIONS = {
@@ -426,10 +431,33 @@ function parseLmStudioModelList(body) {
     return null;
   return data.map((entry) => entry.id).filter((id) => typeof id === "string");
 }
-var LOCAL_INFERENCE_IDS, INFERENCE_PROVIDERS;
+function deriveInferenceBaseUrls(providerId, explicitBaseUrl) {
+  const spec = INFERENCE_PROVIDERS[providerId];
+  if (!explicitBaseUrl) {
+    return {
+      embeddingBaseUrl: spec.defaultEmbeddingBaseUrl,
+      llmBaseUrl: spec.defaultLlmBaseUrl
+    };
+  }
+  let end = explicitBaseUrl.length;
+  while (end > 0 && explicitBaseUrl.charCodeAt(end - 1) === SLASH)
+    end--;
+  const base = explicitBaseUrl.slice(0, end);
+  const suffix = spec.defaultLlmBaseUrl.startsWith(spec.defaultEmbeddingBaseUrl) ? spec.defaultLlmBaseUrl.slice(spec.defaultEmbeddingBaseUrl.length) : "";
+  return {
+    embeddingBaseUrl: base,
+    llmBaseUrl: `${base}${suffix}`
+  };
+}
+var LOCAL_INFERENCE_IDS, INFERENCE_ROLE_DEFAULTS, INFERENCE_PROVIDERS, SLASH = 47;
 var init_inference_providers = __esm(() => {
   init_embedding_dimensions();
   LOCAL_INFERENCE_IDS = ["ollama", "lmstudio"];
+  INFERENCE_ROLE_DEFAULTS = {
+    embedding: { contextWindow: 8192 },
+    instruct: { contextWindow: 16384, temperature: 0.2 },
+    coding: { contextWindow: 32768, temperature: 0 }
+  };
   INFERENCE_PROVIDERS = {
     ollama: {
       id: "ollama",
@@ -441,6 +469,13 @@ var init_inference_providers = __esm(() => {
         dimensions: "OLLAMA_EMBEDDING_DIMENSIONS"
       },
       knownDimensions: KNOWN_EMBEDDING_DIMENSIONS,
+      defaultModels: {
+        embedding: "qwen3-embedding:0.6b",
+        instruct: "qwen3-vl:8b",
+        coding: "qwen2.5-coder:7b"
+      },
+      appliesContextPerRequest: true,
+      embedBatchSize: 64,
       supportsOllamaVersionProbe: true,
       injectsDisableThink: true,
       requiresChatCompletionsApi: false,
@@ -456,8 +491,36 @@ var init_inference_providers = __esm(() => {
         dimensions: "LMSTUDIO_EMBEDDING_DIMENSIONS"
       },
       knownDimensions: {
-        "text-embedding-nomic-embed-text-v1.5": 768
+        "text-embedding-nomic-embed-text-v1.5": 768,
+        "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,
       injectsDisableThink: false,
       requiresChatCompletionsApi: true,
@@ -472,6 +535,7 @@ var API_PROVIDER_IDS, EMBEDDING_PROVIDER_IDS, SCHEDULER_JOB_KINDS, MAX_MATCH_WOR
 var init_massa_ai_config = __esm(() => {
   init_xdg();
   init_inference_providers();
+  init_embedding_dimensions();
   API_PROVIDER_IDS = ["mistral", "openai", "google", "cohere"];
   EMBEDDING_PROVIDER_IDS = [
     ...LOCAL_INFERENCE_IDS,
@@ -538,9 +602,9 @@ var init_massa_ai_config = __esm(() => {
     },
     embedding: {
       provider: "ollama",
-      model: "qwen3-embedding:4b",
+      model: INFERENCE_PROVIDERS.ollama.defaultModels.embedding,
       baseURL: "http://localhost:11434",
-      dimensions: 2560
+      dimensions: knownEmbeddingDimensions(INFERENCE_PROVIDERS.ollama.defaultModels.embedding) ?? 768
     },
     compression: {
       defaultStrategy: "code_structure",
@@ -579,12 +643,15 @@ var init_massa_ai_config = __esm(() => {
       enabled: false,
       baseUrl: "http://localhost:11434/v1",
       apiKey: "ollama",
-      model: "qwen2.5:7b-instruct",
-      codeModel: "qwen2.5-coder:7b",
+      model: INFERENCE_PROVIDERS.ollama.defaultModels.instruct,
+      codeModel: INFERENCE_PROVIDERS.ollama.defaultModels.coding,
       temperature: 0.2,
       maxOutputTokens: 8000,
       timeoutMs: 90000,
-      disableThink: true
+      disableThink: true,
+      contextWindow: INFERENCE_ROLE_DEFAULTS.instruct.contextWindow,
+      codeContextWindow: INFERENCE_ROLE_DEFAULTS.coding.contextWindow,
+      codeTemperature: INFERENCE_ROLE_DEFAULTS.coding.temperature
     },
     memory: {
       decay: {
@@ -985,6 +1052,7 @@ try {
 // ../../packages/shared/dist/config/index.js
 init_config_loader();
 init_massa_ai_config();
+init_inference_providers();
 init_massa_ai_config();
 init_massa_ai_config();
 init_config_loader();
@@ -1002,8 +1070,6 @@ init_config_loader();
 
 // ../../packages/shared/dist/config/index.js
 init_embedding_dimensions();
-var DEFAULT_LLM_MODEL = "qwen2.5:7b-instruct";
-var DEFAULT_LLM_CODE_MODEL = "qwen2.5-coder:7b";
 function envNum(key, fallback) {
   const s = process.env[key];
   if (s === undefined || s === "")
@@ -1176,6 +1242,15 @@ var DEFAULT_ALLOWED_EXTENSIONS = [
   ".hs"
 ];
 var fileConfig = loadConfigSafe();
+function activeInferenceProviderId() {
+  const providerId = fileConfig.embedding?.provider;
+  if (providerId && LOCAL_INFERENCE_IDS.includes(providerId)) {
+    return providerId;
+  }
+  return "ollama";
+}
+var DEFAULT_LLM_MODEL = INFERENCE_PROVIDERS[activeInferenceProviderId()].defaultModels.instruct;
+var DEFAULT_LLM_CODE_MODEL = INFERENCE_PROVIDERS[activeInferenceProviderId()].defaultModels.coding;
 var fileCacheL1Bytes = fileConfig.cache?.l1MaxSizeMB ? fileConfig.cache.l1MaxSizeMB * 1024 * 1024 : undefined;
 var fileCacheL2Bytes = fileConfig.cache?.l2MaxSizeMB ? fileConfig.cache.l2MaxSizeMB * 1024 * 1024 : undefined;
 var resolvedDataDir = getGlobalDataDir();
@@ -1218,7 +1293,10 @@ var defaultConfig = {
     temperature: envNum("MASSA_AI_LLM_TEMPERATURE", fileConfig.llm?.temperature ?? 0.2),
     maxOutputTokens: envNum("MASSA_AI_LLM_MAX_OUTPUT_TOKENS", fileConfig.llm?.maxOutputTokens ?? 8000),
     timeoutMs: envNum("MASSA_AI_LLM_TIMEOUT_MS", fileConfig.llm?.timeoutMs ?? 90000),
-    disableThink: envBool("MASSA_AI_LLM_DISABLE_THINK", fileConfig.llm?.disableThink ?? true)
+    disableThink: envBool("MASSA_AI_LLM_DISABLE_THINK", fileConfig.llm?.disableThink ?? true),
+    contextWindow: fileConfig.llm?.contextWindow ?? INFERENCE_ROLE_DEFAULTS.instruct.contextWindow,
+    codeContextWindow: fileConfig.llm?.codeContextWindow ?? INFERENCE_ROLE_DEFAULTS.coding.contextWindow,
+    codeTemperature: envNum("MASSA_AI_LLM_CODE_TEMPERATURE", fileConfig.llm?.codeTemperature ?? INFERENCE_ROLE_DEFAULTS.coding.temperature)
   },
   memory: {
     decay: {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@massa-ai/opencode-plugin",
-  "version": "1.58.0",
+  "version": "1.60.0",
   "description": "massa-ai plugin for OpenCode - Semantic code search, memory, and context compression",
   "type": "module",
   "main": "dist/index.js",
@@ -24,8 +24,8 @@
   "dependencies": {
     "@opencode-ai/plugin": "^1.2.15",
     "@opencode-ai/sdk": "^1.2.15",
-    "@massa-ai/core": "^1.58.0",
-    "@massa-ai/shared": "^1.58.0"
+    "@massa-ai/core": "^1.60.0",
+    "@massa-ai/shared": "^1.60.0"
   },
   "devDependencies": {
     "@types/node": "^22.10.5",