agency-orchestrator 0.6.3 → 0.6.5

package/dist/cli/compose.js

@@ -66,6 +66,9 @@ function buildComposeSystemPromptEn(catalog, options) {
     const autoRun = options?.autoRun ?? false;
     const provider = options?.provider || 'deepseek';
     const model = options?.model;
+    const isLocal = provider === 'ollama';
+    const maxTokens = isLocal ? 8192 : 4096;
+    const timeoutMs = isLocal ? 600000 : 120000;
     const inputsSection = autoRun
         ? `
 ## Important: Direct Run Mode
@@ -109,8 +112,8 @@ agents_dir: "agency-agents"
 llm:
   provider: ${provider}
   ${model ? `model: ${model}` : ''}
-  max_tokens: 4096
-  timeout: 120000
+  max_tokens: ${maxTokens}
+  timeout: ${timeoutMs}
   retry: 2
 
 concurrency: 2
@@ -155,6 +158,9 @@ function buildComposeSystemPromptZh(catalog, options) {
     const autoRun = options?.autoRun ?? false;
     const provider = options?.provider || 'deepseek';
     const model = options?.model;
+    const isLocal = provider === 'ollama';
+    const maxTokens = isLocal ? 8192 : 4096;
+    const timeoutMs = isLocal ? 600000 : 120000;
     const inputsSection = autoRun
         ? `
 ## 重要：直接运行模式
@@ -198,8 +204,8 @@ agents_dir: "agency-agents-zh"
 llm:
   provider: ${provider}
   ${model ? `model: ${model}` : ''}
-  max_tokens: 4096
-  timeout: 120000
+  max_tokens: ${maxTokens}
+  timeout: ${timeoutMs}
   retry: 2
 
 concurrency: 2

package/dist/cli/demo.js

@@ -151,7 +151,8 @@ export async function detectAvailableLLMs() {
     try {
         const controller = new AbortController();
         const timeout = setTimeout(() => controller.abort(), 2000);
-        const res = await fetch('http://localhost:11434/api/tags', {
+        const ollamaUrl = process.env.OLLAMA_BASE_URL || 'http://localhost:11434';
+        const res = await fetch(`${ollamaUrl.replace(/\/+$/, '')}/api/tags`, {
             signal: controller.signal,
         });
         clearTimeout(timeout);

package/dist/connectors/ollama.js

@@ -5,23 +5,36 @@ export class OllamaConnector {
         this.baseUrl = raw.replace(/\/+$/, '');
     }
     async chat(systemPrompt, userMessage, config) {
-        const response = await fetch(`${this.baseUrl}/api/chat`, {
-            method: 'POST',
-            headers: { 'Content-Type': 'application/json' },
-            body: JSON.stringify({
-                model: config.model || 'llama3.1',
-                messages: [
-                    { role: 'system', content: systemPrompt },
-                    { role: 'user', content: userMessage },
-                ],
-                stream: false,
-                options: {
-                    num_predict: config.max_tokens || 2048,
-                },
-            }),
-        });
+        let response;
+        try {
+            response = await fetch(`${this.baseUrl}/api/chat`, {
+                method: 'POST',
+                headers: { 'Content-Type': 'application/json' },
+                body: JSON.stringify({
+                    model: config.model || 'llama3.1',
+                    messages: [
+                        { role: 'system', content: systemPrompt },
+                        { role: 'user', content: userMessage },
+                    ],
+                    stream: false,
+                    options: {
+                        num_predict: config.max_tokens || 8192,
+                    },
+                }),
+            });
+        }
+        catch (err) {
+            if (err?.cause?.code === 'ECONNREFUSED' || err?.message?.includes('ECONNREFUSED')) {
+                throw new Error(`无法连接 Ollama (${this.baseUrl})，请确认 ollama 已启动。Docker 环境请设置 OLLAMA_BASE_URL=http://host.docker.internal:11434`);
+            }
+            throw err;
+        }
         if (!response.ok) {
             const text = await response.text();
+            if (response.status === 404 && text.includes('not found')) {
+                const model = config.model || 'llama3.1';
+                throw new Error(`Ollama 模型 "${model}" 未找到，请先运行: ollama pull ${model}`);
+            }
             throw new Error(`Ollama error ${response.status}: ${text}`);
         }
         const data = await response.json();

package/dist/core/executor.js

@@ -10,7 +10,8 @@ export async function executeDAG(dag, options) {
     const startTime = Date.now();
     const stepResults = [];
     const isCLI = llmConfig.provider.endsWith('-cli') || llmConfig.provider === 'claude-code';
-    const timeout = llmConfig.timeout || (isCLI ? 600_000 : 120_000); // CLI 10分钟（gateway/MiniMax 等可能单步 5+ 分钟），API 2分钟
+    const isLocal = llmConfig.provider === 'ollama';
+    const timeout = llmConfig.timeout || (isCLI ? 600_000 : isLocal ? 600_000 : 120_000);
     const maxRetry = llmConfig.retry ?? 5;
     // CLI provider 强制串行：共享同一账户额度，并发会触发限速反而更慢
     const effectiveConcurrency = isCLI ? 1 : concurrency;

package/dist/mcp/server.js

@@ -232,14 +232,14 @@ export async function startServer(verbose = false) {
     }, async ({ description, provider, model }) => {
         try {
             const agentsDir = findAgentsDir();
-            const llmProvider = provider || 'deepseek';
+            const llmProvider = provider || process.env.AO_PROVIDER || 'deepseek';
             const defaultModels = {
                 deepseek: 'deepseek-chat',
                 claude: 'claude-sonnet-4-20250514',
                 openai: 'gpt-4o',
                 ollama: 'llama3',
             };
-            const llmModel = model || defaultModels[llmProvider] || 'gpt-4o';
+            const llmModel = model || process.env.AO_MODEL || defaultModels[llmProvider] || 'gpt-4o';
             const result = await silentCall(() => composeWorkflow({
                 description,
                 agentsDir,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "agency-orchestrator",
-  "version": "0.6.3",
+  "version": "0.6.5",
   "description": "Multi-agent YAML workflow engine — 211 AI roles, auto DAG parallelism, zero code. One sentence → multiple AI roles collaborate → complete plan in minutes. 10 LLM providers, 7 need no API key.",
   "keywords": [
     "multi-agent",