@massa-ai/opencode-plugin 1.57.0 → 1.59.0

package/dist/config-cli.js

@@ -399,11 +399,129 @@ 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 = {
+    "qwen3-embedding:8b": 4096,
+    "qwen3-embedding:4b": 2560,
+    "qwen3-embedding:0.6b": 1024,
+    "bge-m3": 1024
+  };
+});
+
+// ../../packages/shared/dist/config/inference-providers.js
+function parseOllamaModelList(body) {
+  if (typeof body !== "object" || body === null)
+    return null;
+  const models = body.models;
+  if (!Array.isArray(models))
+    return null;
+  return models.map((entry) => entry.name).filter((name) => typeof name === "string");
+}
+function parseLmStudioModelList(body) {
+  if (typeof body !== "object" || body === null)
+    return null;
+  const data = body.data;
+  if (!Array.isArray(data))
+    return null;
+  return data.map((entry) => entry.id).filter((id) => typeof id === "string");
+}
+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",
+      defaultEmbeddingBaseUrl: "http://localhost:11434",
+      defaultLlmBaseUrl: "http://localhost:11434/v1",
+      envNames: {
+        model: "OLLAMA_EMBEDDING_MODEL",
+        baseUrl: "OLLAMA_BASE_URL",
+        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,
+      parseModelList: parseOllamaModelList
+    },
+    lmstudio: {
+      id: "lmstudio",
+      defaultEmbeddingBaseUrl: "http://localhost:1234/v1",
+      defaultLlmBaseUrl: "http://localhost:1234/v1",
+      envNames: {
+        model: "LMSTUDIO_EMBEDDING_MODEL",
+        baseUrl: "LMSTUDIO_BASE_URL",
+        dimensions: "LMSTUDIO_EMBEDDING_DIMENSIONS"
+      },
+      knownDimensions: {
+        "text-embedding-nomic-embed-text-v1.5": 768,
+        "text-embedding-qwen3-embedding-0.6b": 1024
+      },
+      defaultModels: {
+        embedding: "text-embedding-qwen3-embedding-0.6b",
+        instruct: "qwen3-vl-8b-instruct",
+        coding: "qwen2.5-coder-7b-instruct"
+      },
+      appliesContextPerRequest: false,
+      embedBatchSize: 64,
+      supportsOllamaVersionProbe: false,
+      injectsDisableThink: false,
+      requiresChatCompletionsApi: true,
+      parseModelList: parseLmStudioModelList
+    }
+  };
+});
+
 // ../../packages/shared/dist/config/massa-ai-config.js
 import path2 from "path";
-var SCHEDULER_JOB_KINDS, MAX_MATCH_WORK = 1e5, MAX_IGNORE_PATTERNS = 1024, DEFAULT_CAPTURE_POLICY, DEFAULT_SCHEDULER_CONFIG, defaultMassaAiConfig;
+var API_PROVIDER_IDS, EMBEDDING_PROVIDER_IDS, SCHEDULER_JOB_KINDS, MAX_MATCH_WORK = 1e5, MAX_IGNORE_PATTERNS = 1024, DEFAULT_CAPTURE_POLICY, DEFAULT_SCHEDULER_CONFIG, defaultMassaAiConfig;
 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,
+    ...API_PROVIDER_IDS
+  ];
   SCHEDULER_JOB_KINDS = [
     "memory-consolidation",
     "decay-sweep",
@@ -465,9 +583,9 @@ 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",
@@ -506,12 +624,15 @@ 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: {
@@ -672,11 +793,12 @@ function loadConfig() {
   try {
     const content = fs.readFileSync(CONFIG_FILE, "utf-8");
     const userConfig = JSON.parse(content);
+    const embeddingBase = userConfig.embedding?.provider && userConfig.embedding.provider !== defaultMassaAiConfig.embedding.provider ? {} : defaultMassaAiConfig.embedding;
     return {
       ...defaultMassaAiConfig,
       ...userConfig,
       database: { ...defaultMassaAiConfig.database, ...userConfig.database },
-      embedding: { ...defaultMassaAiConfig.embedding, ...userConfig.embedding },
+      embedding: { ...embeddingBase, ...userConfig.embedding },
       compression: { ...defaultMassaAiConfig.compression, ...userConfig.compression },
       cache: { ...defaultMassaAiConfig.cache, ...userConfig.cache },
       logging: { ...defaultMassaAiConfig.logging, ...userConfig.logging },
@@ -821,18 +943,27 @@ function initConfig() {
 function getConfigForEnv() {
   const config = loadConfig();
   const env = {};
-  if (config.embedding.provider === "ollama") {
+  const provider = config.embedding.provider;
+  if (provider === "ollama") {
     env.OLLAMA_EMBEDDING_MODEL = config.embedding.model;
     env.OLLAMA_BASE_URL = config.embedding.baseURL || "http://localhost:11434";
     if (config.embedding.dimensions) {
       env.OLLAMA_EMBEDDING_DIMENSIONS = String(config.embedding.dimensions);
     }
-  } else if (config.embedding.provider === "mistral") {
+  } else if (provider === "lmstudio") {
+    env.LMSTUDIO_EMBEDDING_MODEL = config.embedding.model;
+    env.LMSTUDIO_BASE_URL = config.embedding.baseURL || INFERENCE_PROVIDERS.lmstudio.defaultEmbeddingBaseUrl;
+    if (config.embedding.dimensions) {
+      env.LMSTUDIO_EMBEDDING_DIMENSIONS = String(config.embedding.dimensions);
+    }
+  } else if (provider === "mistral") {
     env.MISTRAL_API_KEY = config.embedding.apiKey || "";
     env.MISTRAL_TEXT_EMBEDDING_MODEL = config.embedding.model;
-  } else if (config.embedding.provider === "openai") {
+  } else if (provider === "openai") {
     env.OPENAI_API_KEY = config.embedding.apiKey || "";
     env.OPENAI_EMBEDDING_MODEL = config.embedding.model;
+  } else {
+    console.error(`[getConfigForEnv] embedding.provider "${provider}" has no env-projection branch \u2014 no embedding env vars were set`);
   }
   env.LOG_LEVEL = config.logging.level;
   env.ENABLE_METRICS = String(config.logging.enableMetrics);
@@ -842,6 +973,7 @@ var CONFIG_DIR, CONFIG_FILE, ConfigParseError, ConfigWriteConflictError, migrati
 var init_config_loader = __esm(() => {
   init_massa_ai_config();
   init_xdg();
+  init_inference_providers();
   CONFIG_DIR = configDir("massa-ai");
   CONFIG_FILE = path3.join(CONFIG_DIR, "config.json");
   ConfigParseError = class ConfigParseError extends Error {
@@ -901,6 +1033,7 @@ 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();
@@ -917,8 +1050,7 @@ init_xdg();
 init_config_loader();
 
 // ../../packages/shared/dist/config/index.js
-var DEFAULT_LLM_MODEL = "qwen2.5:7b-instruct";
-var DEFAULT_LLM_CODE_MODEL = "qwen2.5-coder:7b";
+init_embedding_dimensions();
 function envNum(key, fallback) {
   const s = process.env[key];
   if (s === undefined || s === "")
@@ -1091,6 +1223,15 @@ 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();
@@ -1133,7 +1274,10 @@ 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: {
@@ -3275,11 +3419,13 @@ function formatBootstrapReport(report) {
 // ../../packages/shared/dist/bootstrap/index.js
 init_config_loader();
 // src/config-cli.ts
+init_inference_providers();
 import { promises as fs11 } from "fs";
 import path15 from "path";
 import os8 from "os";
 import { fileURLToPath } from "url";
 var __dirname2 = path15.dirname(fileURLToPath(import.meta.url));
+var WRITABLE_PROVIDERS = ["ollama", "lmstudio", "mistral", "openai", "google", "cohere"];
 var GENERATOR_MARKER = "scripts/generate-subagent-artifacts.ts";
 var GENERATOR_MARKER_MAX_LEVELS = 6;
 function formatVariantSync(results) {
@@ -3299,16 +3445,18 @@ Usage:
 Commands:
   init              Initialize massa-ai configuration
     --ollama          Use Ollama (local, default)
+    --lmstudio        Use LM Studio (local)
     --mistral <key>   Use Mistral with API key
     --openai <key>    Use OpenAI with API key
 
   path              Show config file path
   show              Show current configuration
   set <key> <val>   Set a configuration value
-  use <provider>    Switch embedding provider
-    --api-key <key>   API key (required for mistral/openai)
+  use <provider>    Switch embedding provider (ollama, lmstudio, mistral,
+                    openai, google, cohere)
+    --api-key <key>   API key (required for mistral/openai/google/cohere)
     --model <name>    Model name
-    --base-url <url>  Base URL (for ollama)
+    --base-url <url>  Base URL (for ollama/lmstudio)
 
   agents            Manage the 18 subagent specialist definitions
     agents install [--user|--project]   Write 18 agent .md files
@@ -3328,8 +3476,10 @@ Commands:
 
 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
   massa-ai-config agents install --user
@@ -3408,6 +3558,20 @@ async function runCli(argv) {
         };
         saveConfig(config2);
         console.log("\u2713 Configured for OpenAI embeddings");
+      } else if (options.lmstudio) {
+        const config2 = loadConfig();
+        const model = INFERENCE_PROVIDERS.lmstudio.defaultModels.embedding;
+        config2.embedding = {
+          provider: "lmstudio",
+          model,
+          baseURL: INFERENCE_PROVIDERS.lmstudio.defaultEmbeddingBaseUrl,
+          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 {
         console.log("\u2713 Configured for Ollama (local) embeddings");
       }
@@ -3452,18 +3616,35 @@ Using defaults:`);
     }
     case "use": {
       const provider = args[1];
-      if (!provider || !["ollama", "mistral", "openai"].includes(provider)) {
-        console.error("Provider must be: ollama, mistral, or openai");
+      if (!provider || !WRITABLE_PROVIDERS.includes(provider)) {
+        console.error(`Provider must be one of: ${WRITABLE_PROVIDERS.join(", ")}`);
         return 1;
       }
       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 || INFERENCE_PROVIDERS.lmstudio.defaultModels.embedding;
+        const urls = deriveInferenceBaseUrls("lmstudio", options["base-url"]);
+        config2.embedding = {
+          provider: "lmstudio",
+          model,
+          baseURL: urls.embeddingBaseUrl,
+          dimensions: INFERENCE_PROVIDERS.lmstudio.knownDimensions[model] ?? 768
         };
+        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");
@@ -3486,6 +3667,28 @@ Using defaults:`);
           apiKey: options["api-key"],
           dimensions: 1536
         };
+      } else if (provider === "google") {
+        if (!options["api-key"]) {
+          console.error("Error: --api-key required for Google");
+          return 1;
+        }
+        config2.embedding = {
+          provider: "google",
+          model: options.model || "gemini-embedding-001",
+          apiKey: options["api-key"],
+          dimensions: 3072
+        };
+      } else if (provider === "cohere") {
+        if (!options["api-key"]) {
+          console.error("Error: --api-key required for Cohere");
+          return 1;
+        }
+        config2.embedding = {
+          provider: "cohere",
+          model: options.model || "embed-english-v3.0",
+          apiKey: options["api-key"],
+          dimensions: 1024
+        };
       }
       saveConfig(config2);
       console.log(`\u2713 Switched to ${provider} embeddings`);

package/dist/index.js

@@ -398,11 +398,129 @@ 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 = {
+    "qwen3-embedding:8b": 4096,
+    "qwen3-embedding:4b": 2560,
+    "qwen3-embedding:0.6b": 1024,
+    "bge-m3": 1024
+  };
+});
+
+// ../../packages/shared/dist/config/inference-providers.js
+function parseOllamaModelList(body) {
+  if (typeof body !== "object" || body === null)
+    return null;
+  const models = body.models;
+  if (!Array.isArray(models))
+    return null;
+  return models.map((entry) => entry.name).filter((name) => typeof name === "string");
+}
+function parseLmStudioModelList(body) {
+  if (typeof body !== "object" || body === null)
+    return null;
+  const data = body.data;
+  if (!Array.isArray(data))
+    return null;
+  return data.map((entry) => entry.id).filter((id) => typeof id === "string");
+}
+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",
+      defaultEmbeddingBaseUrl: "http://localhost:11434",
+      defaultLlmBaseUrl: "http://localhost:11434/v1",
+      envNames: {
+        model: "OLLAMA_EMBEDDING_MODEL",
+        baseUrl: "OLLAMA_BASE_URL",
+        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,
+      parseModelList: parseOllamaModelList
+    },
+    lmstudio: {
+      id: "lmstudio",
+      defaultEmbeddingBaseUrl: "http://localhost:1234/v1",
+      defaultLlmBaseUrl: "http://localhost:1234/v1",
+      envNames: {
+        model: "LMSTUDIO_EMBEDDING_MODEL",
+        baseUrl: "LMSTUDIO_BASE_URL",
+        dimensions: "LMSTUDIO_EMBEDDING_DIMENSIONS"
+      },
+      knownDimensions: {
+        "text-embedding-nomic-embed-text-v1.5": 768,
+        "text-embedding-qwen3-embedding-0.6b": 1024
+      },
+      defaultModels: {
+        embedding: "text-embedding-qwen3-embedding-0.6b",
+        instruct: "qwen3-vl-8b-instruct",
+        coding: "qwen2.5-coder-7b-instruct"
+      },
+      appliesContextPerRequest: false,
+      embedBatchSize: 64,
+      supportsOllamaVersionProbe: false,
+      injectsDisableThink: false,
+      requiresChatCompletionsApi: true,
+      parseModelList: parseLmStudioModelList
+    }
+  };
+});
+
 // ../../packages/shared/dist/config/massa-ai-config.js
 import path2 from "path";
-var SCHEDULER_JOB_KINDS, MAX_MATCH_WORK = 1e5, MAX_IGNORE_PATTERNS = 1024, DEFAULT_CAPTURE_POLICY, DEFAULT_SCHEDULER_CONFIG, defaultMassaAiConfig;
+var API_PROVIDER_IDS, EMBEDDING_PROVIDER_IDS, SCHEDULER_JOB_KINDS, MAX_MATCH_WORK = 1e5, MAX_IGNORE_PATTERNS = 1024, DEFAULT_CAPTURE_POLICY, DEFAULT_SCHEDULER_CONFIG, defaultMassaAiConfig;
 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,
+    ...API_PROVIDER_IDS
+  ];
   SCHEDULER_JOB_KINDS = [
     "memory-consolidation",
     "decay-sweep",
@@ -464,9 +582,9 @@ 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",
@@ -505,12 +623,15 @@ 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: {
@@ -671,11 +792,12 @@ function loadConfig() {
   try {
     const content = fs.readFileSync(CONFIG_FILE, "utf-8");
     const userConfig = JSON.parse(content);
+    const embeddingBase = userConfig.embedding?.provider && userConfig.embedding.provider !== defaultMassaAiConfig.embedding.provider ? {} : defaultMassaAiConfig.embedding;
     return {
       ...defaultMassaAiConfig,
       ...userConfig,
       database: { ...defaultMassaAiConfig.database, ...userConfig.database },
-      embedding: { ...defaultMassaAiConfig.embedding, ...userConfig.embedding },
+      embedding: { ...embeddingBase, ...userConfig.embedding },
       compression: { ...defaultMassaAiConfig.compression, ...userConfig.compression },
       cache: { ...defaultMassaAiConfig.cache, ...userConfig.cache },
       logging: { ...defaultMassaAiConfig.logging, ...userConfig.logging },
@@ -820,18 +942,27 @@ function initConfig() {
 function getConfigForEnv() {
   const config = loadConfig();
   const env = {};
-  if (config.embedding.provider === "ollama") {
+  const provider = config.embedding.provider;
+  if (provider === "ollama") {
     env.OLLAMA_EMBEDDING_MODEL = config.embedding.model;
     env.OLLAMA_BASE_URL = config.embedding.baseURL || "http://localhost:11434";
     if (config.embedding.dimensions) {
       env.OLLAMA_EMBEDDING_DIMENSIONS = String(config.embedding.dimensions);
     }
-  } else if (config.embedding.provider === "mistral") {
+  } else if (provider === "lmstudio") {
+    env.LMSTUDIO_EMBEDDING_MODEL = config.embedding.model;
+    env.LMSTUDIO_BASE_URL = config.embedding.baseURL || INFERENCE_PROVIDERS.lmstudio.defaultEmbeddingBaseUrl;
+    if (config.embedding.dimensions) {
+      env.LMSTUDIO_EMBEDDING_DIMENSIONS = String(config.embedding.dimensions);
+    }
+  } else if (provider === "mistral") {
     env.MISTRAL_API_KEY = config.embedding.apiKey || "";
     env.MISTRAL_TEXT_EMBEDDING_MODEL = config.embedding.model;
-  } else if (config.embedding.provider === "openai") {
+  } else if (provider === "openai") {
     env.OPENAI_API_KEY = config.embedding.apiKey || "";
     env.OPENAI_EMBEDDING_MODEL = config.embedding.model;
+  } else {
+    console.error(`[getConfigForEnv] embedding.provider "${provider}" has no env-projection branch \u2014 no embedding env vars were set`);
   }
   env.LOG_LEVEL = config.logging.level;
   env.ENABLE_METRICS = String(config.logging.enableMetrics);
@@ -841,6 +972,7 @@ var CONFIG_DIR, CONFIG_FILE, ConfigParseError, ConfigWriteConflictError, migrati
 var init_config_loader = __esm(() => {
   init_massa_ai_config();
   init_xdg();
+  init_inference_providers();
   CONFIG_DIR = configDir("massa-ai");
   CONFIG_FILE = path3.join(CONFIG_DIR, "config.json");
   ConfigParseError = class ConfigParseError extends Error {
@@ -900,6 +1032,7 @@ 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();
@@ -916,8 +1049,7 @@ init_xdg();
 init_config_loader();
 
 // ../../packages/shared/dist/config/index.js
-var DEFAULT_LLM_MODEL = "qwen2.5:7b-instruct";
-var DEFAULT_LLM_CODE_MODEL = "qwen2.5-coder:7b";
+init_embedding_dimensions();
 function envNum(key, fallback) {
   const s = process.env[key];
   if (s === undefined || s === "")
@@ -1090,6 +1222,15 @@ 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();
@@ -1132,7 +1273,10 @@ 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": "1.57.0",
+  "version": "1.59.0",
   "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": "^1.57.0",
-    "@massa-ai/shared": "^1.57.0"
+    "@massa-ai/core": "^1.59.0",
+    "@massa-ai/shared": "^1.59.0"
   },
   "devDependencies": {
     "@types/node": "^22.10.5",