feat: add AnthropicVertexClient for claude-* models

- @anthropic-ai/vertex-sdk: proper Anthropic Messages API on Vertex
- AnthropicVertexClient: converts OAI message format ↔ Anthropic format,
  handles tool_use blocks, retries 429/503 with backoff
- createLLM: routes anthropic/* and claude-* models through new client
- Tier B/C default: claude-sonnet-4-6 via us-east5 Vertex endpoint
- /generate endpoint: accepts region param for regional endpoint testing

Made-with: Cursor

package-lock.json

@@ -8,6 +8,8 @@
       "name": "vibn-agent-runner",
       "version": "0.1.0",
       "dependencies": {
+        "@anthropic-ai/sdk": "^0.78.0",
+        "@anthropic-ai/vertex-sdk": "^0.14.4",
         "@google/genai": "^1.0.0",
         "cors": "^2.8.5",
         "express": "^4.19.2",
@@ -24,6 +26,121 @@
         "typescript": "^5.4.5"
       }
     },
+    "node_modules/@anthropic-ai/sdk": {
+      "version": "0.78.0",
+      "resolved": "https://registry.npmjs.org/@anthropic-ai/sdk/-/sdk-0.78.0.tgz",
+      "integrity": "sha512-PzQhR715td/m1UaaN5hHXjYB8Gl2lF9UVhrrGrZeysiF6Rb74Wc9GCB8hzLdzmQtBd1qe89F9OptgB9Za1Ib5w==",
+      "license": "MIT",
+      "dependencies": {
+        "json-schema-to-ts": "^3.1.1"
+      },
+      "bin": {
+        "anthropic-ai-sdk": "bin/cli"
+      },
+      "peerDependencies": {
+        "zod": "^3.25.0 || ^4.0.0"
+      },
+      "peerDependenciesMeta": {
+        "zod": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/@anthropic-ai/vertex-sdk": {
+      "version": "0.14.4",
+      "resolved": "https://registry.npmjs.org/@anthropic-ai/vertex-sdk/-/vertex-sdk-0.14.4.tgz",
+      "integrity": "sha512-BZUPRWghZxfSFtAxU563wH+jfWBPoedAwsVxG35FhmNsjeV8tyfN+lFriWhCpcZApxA4NdT6Soov+PzfnxxD5g==",
+      "license": "MIT",
+      "dependencies": {
+        "@anthropic-ai/sdk": ">=0.50.3 <1",
+        "google-auth-library": "^9.4.2"
+      }
+    },
+    "node_modules/@anthropic-ai/vertex-sdk/node_modules/gaxios": {
+      "version": "6.7.1",
+      "resolved": "https://registry.npmjs.org/gaxios/-/gaxios-6.7.1.tgz",
+      "integrity": "sha512-LDODD4TMYx7XXdpwxAVRAIAuB0bzv0s+ywFonY46k126qzQHT9ygyoa9tncmOiQmmDrik65UYsEkv3lbfqQ3yQ==",
+      "license": "Apache-2.0",
+      "dependencies": {
+        "extend": "^3.0.2",
+        "https-proxy-agent": "^7.0.1",
+        "is-stream": "^2.0.0",
+        "node-fetch": "^2.6.9",
+        "uuid": "^9.0.1"
+      },
+      "engines": {
+        "node": ">=14"
+      }
+    },
+    "node_modules/@anthropic-ai/vertex-sdk/node_modules/gcp-metadata": {
+      "version": "6.1.1",
+      "resolved": "https://registry.npmjs.org/gcp-metadata/-/gcp-metadata-6.1.1.tgz",
+      "integrity": "sha512-a4tiq7E0/5fTjxPAaH4jpjkSv/uCaU2p5KC6HVGrvl0cDjA8iBZv4vv1gyzlmK0ZUKqwpOyQMKzZQe3lTit77A==",
+      "license": "Apache-2.0",
+      "dependencies": {
+        "gaxios": "^6.1.1",
+        "google-logging-utils": "^0.0.2",
+        "json-bigint": "^1.0.0"
+      },
+      "engines": {
+        "node": ">=14"
+      }
+    },
+    "node_modules/@anthropic-ai/vertex-sdk/node_modules/google-auth-library": {
+      "version": "9.15.1",
+      "resolved": "https://registry.npmjs.org/google-auth-library/-/google-auth-library-9.15.1.tgz",
+      "integrity": "sha512-Jb6Z0+nvECVz+2lzSMt9u98UsoakXxA2HGHMCxh+so3n90XgYWkq5dur19JAJV7ONiJY22yBTyJB1TSkvPq9Ng==",
+      "license": "Apache-2.0",
+      "dependencies": {
+        "base64-js": "^1.3.0",
+        "ecdsa-sig-formatter": "^1.0.11",
+        "gaxios": "^6.1.1",
+        "gcp-metadata": "^6.1.0",
+        "gtoken": "^7.0.0",
+        "jws": "^4.0.0"
+      },
+      "engines": {
+        "node": ">=14"
+      }
+    },
+    "node_modules/@anthropic-ai/vertex-sdk/node_modules/google-logging-utils": {
+      "version": "0.0.2",
+      "resolved": "https://registry.npmjs.org/google-logging-utils/-/google-logging-utils-0.0.2.tgz",
+      "integrity": "sha512-NEgUnEcBiP5HrPzufUkBzJOD/Sxsco3rLNo1F1TNf7ieU8ryUzBhqba8r756CjLX7rn3fHl6iLEwPYuqpoKgQQ==",
+      "license": "Apache-2.0",
+      "engines": {
+        "node": ">=14"
+      }
+    },
+    "node_modules/@anthropic-ai/vertex-sdk/node_modules/node-fetch": {
+      "version": "2.7.0",
+      "resolved": "https://registry.npmjs.org/node-fetch/-/node-fetch-2.7.0.tgz",
+      "integrity": "sha512-c4FRfUm/dbcWZ7U+1Wq0AwCyFL+3nt2bEw05wfxSz+DWpWsitgmSgYmy2dQdWyKC1694ELPqMs/YzUSNozLt8A==",
+      "license": "MIT",
+      "dependencies": {
+        "whatwg-url": "^5.0.0"
+      },
+      "engines": {
+        "node": "4.x || >=6.0.0"
+      },
+      "peerDependencies": {
+        "encoding": "^0.1.0"
+      },
+      "peerDependenciesMeta": {
+        "encoding": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/@babel/runtime": {
+      "version": "7.28.6",
+      "resolved": "https://registry.npmjs.org/@babel/runtime/-/runtime-7.28.6.tgz",
+      "integrity": "sha512-05WQkdpL9COIMz4LjTxGpPNCdlpyimKppYNoJ5Di5EUObifl8t4tuLuUBBZEpoLYOmfvIWrsp9fCl0HoPRVTdA==",
+      "license": "MIT",
+      "engines": {
+        "node": ">=6.9.0"
+      }
+    },
     "node_modules/@cspotcode/source-map-support": {
       "version": "0.8.1",
       "resolved": "https://registry.npmjs.org/@cspotcode/source-map-support/-/source-map-support-0.8.1.tgz",
@@ -1051,6 +1168,55 @@
         "url": "https://github.com/sponsors/ljharb"
       }
     },
+    "node_modules/gtoken": {
+      "version": "7.1.0",
+      "resolved": "https://registry.npmjs.org/gtoken/-/gtoken-7.1.0.tgz",
+      "integrity": "sha512-pCcEwRi+TKpMlxAQObHDQ56KawURgyAf6jtIY046fJ5tIv3zDe/LEIubckAO8fj6JnAxLdmWkUfNyulQ2iKdEw==",
+      "license": "MIT",
+      "dependencies": {
+        "gaxios": "^6.0.0",
+        "jws": "^4.0.0"
+      },
+      "engines": {
+        "node": ">=14.0.0"
+      }
+    },
+    "node_modules/gtoken/node_modules/gaxios": {
+      "version": "6.7.1",
+      "resolved": "https://registry.npmjs.org/gaxios/-/gaxios-6.7.1.tgz",
+      "integrity": "sha512-LDODD4TMYx7XXdpwxAVRAIAuB0bzv0s+ywFonY46k126qzQHT9ygyoa9tncmOiQmmDrik65UYsEkv3lbfqQ3yQ==",
+      "license": "Apache-2.0",
+      "dependencies": {
+        "extend": "^3.0.2",
+        "https-proxy-agent": "^7.0.1",
+        "is-stream": "^2.0.0",
+        "node-fetch": "^2.6.9",
+        "uuid": "^9.0.1"
+      },
+      "engines": {
+        "node": ">=14"
+      }
+    },
+    "node_modules/gtoken/node_modules/node-fetch": {
+      "version": "2.7.0",
+      "resolved": "https://registry.npmjs.org/node-fetch/-/node-fetch-2.7.0.tgz",
+      "integrity": "sha512-c4FRfUm/dbcWZ7U+1Wq0AwCyFL+3nt2bEw05wfxSz+DWpWsitgmSgYmy2dQdWyKC1694ELPqMs/YzUSNozLt8A==",
+      "license": "MIT",
+      "dependencies": {
+        "whatwg-url": "^5.0.0"
+      },
+      "engines": {
+        "node": "4.x || >=6.0.0"
+      },
+      "peerDependencies": {
+        "encoding": "^0.1.0"
+      },
+      "peerDependenciesMeta": {
+        "encoding": {
+          "optional": true
+        }
+      }
+    },
     "node_modules/has-symbols": {
       "version": "1.1.0",
       "resolved": "https://registry.npmjs.org/has-symbols/-/has-symbols-1.1.0.tgz",
@@ -1167,6 +1333,18 @@
         "node": ">=8"
       }
     },
+    "node_modules/is-stream": {
+      "version": "2.0.1",
+      "resolved": "https://registry.npmjs.org/is-stream/-/is-stream-2.0.1.tgz",
+      "integrity": "sha512-hFoiJiTl63nn+kstHGBtewWSKnQLpyb155KHheA1l39uvtO9nWIop1p3udqPcUd/xbF1VLMO4n7OI6p7RbngDg==",
+      "license": "MIT",
+      "engines": {
+        "node": ">=8"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
     "node_modules/isexe": {
       "version": "2.0.0",
       "resolved": "https://registry.npmjs.org/isexe/-/isexe-2.0.0.tgz",
@@ -1197,6 +1375,19 @@
         "bignumber.js": "^9.0.0"
       }
     },
+    "node_modules/json-schema-to-ts": {
+      "version": "3.1.1",
+      "resolved": "https://registry.npmjs.org/json-schema-to-ts/-/json-schema-to-ts-3.1.1.tgz",
+      "integrity": "sha512-+DWg8jCJG2TEnpy7kOm/7/AxaYoaRbjVB4LFZLySZlWn8exGs3A4OLJR966cVvU26N7X9TWxl+Jsw7dzAqKT6g==",
+      "license": "MIT",
+      "dependencies": {
+        "@babel/runtime": "^7.18.3",
+        "ts-algebra": "^2.0.0"
+      },
+      "engines": {
+        "node": ">=16"
+      }
+    },
     "node_modules/jwa": {
       "version": "2.0.1",
       "resolved": "https://registry.npmjs.org/jwa/-/jwa-2.0.1.tgz",
@@ -1871,6 +2062,18 @@
         "node": ">=0.6"
       }
     },
+    "node_modules/tr46": {
+      "version": "0.0.3",
+      "resolved": "https://registry.npmjs.org/tr46/-/tr46-0.0.3.tgz",
+      "integrity": "sha512-N3WMsuqV66lT30CrXNbEjx4GEwlow3v6rr4mCcv6prnfwhS01rkgyFdjPNBYd9br7LpXV1+Emh01fHnq2Gdgrw==",
+      "license": "MIT"
+    },
+    "node_modules/ts-algebra": {
+      "version": "2.0.0",
+      "resolved": "https://registry.npmjs.org/ts-algebra/-/ts-algebra-2.0.0.tgz",
+      "integrity": "sha512-FPAhNPFMrkwz76P7cdjdmiShwMynZYN6SgOujD1urY4oNm80Ou9oMdmbR45LotcKOXoy7wSmHkRFE6Mxbrhefw==",
+      "license": "MIT"
+    },
     "node_modules/ts-node": {
       "version": "10.9.2",
       "resolved": "https://registry.npmjs.org/ts-node/-/ts-node-10.9.2.tgz",
@@ -2004,6 +2207,22 @@
         "node": ">= 8"
       }
     },
+    "node_modules/webidl-conversions": {
+      "version": "3.0.1",
+      "resolved": "https://registry.npmjs.org/webidl-conversions/-/webidl-conversions-3.0.1.tgz",
+      "integrity": "sha512-2JAn3z8AR6rjK8Sm8orRC0h/bcl/DqL7tRPdGZ4I1CjdF+EaMLmYxBHyXuKL849eucPFhvBoxMsflfOb8kxaeQ==",
+      "license": "BSD-2-Clause"
+    },
+    "node_modules/whatwg-url": {
+      "version": "5.0.0",
+      "resolved": "https://registry.npmjs.org/whatwg-url/-/whatwg-url-5.0.0.tgz",
+      "integrity": "sha512-saE57nupxk6v3HY35+jzBwYa0rKSy0XR8JSxZPwgLr7ys0IBzhGviA1/TUGJLmSVqs8pb9AnvICXEuOHLprYTw==",
+      "license": "MIT",
+      "dependencies": {
+        "tr46": "~0.0.3",
+        "webidl-conversions": "^3.0.0"
+      }
+    },
     "node_modules/which": {
       "version": "2.0.2",
       "resolved": "https://registry.npmjs.org/which/-/which-2.0.2.tgz",

package.json

@@ -10,6 +10,8 @@
     "test": "ts-node src/test.ts"
   },
   "dependencies": {
+    "@anthropic-ai/sdk": "^0.78.0",
+    "@anthropic-ai/vertex-sdk": "^0.14.4",
     "@google/genai": "^1.0.0",
     "cors": "^2.8.5",
     "express": "^4.19.2",

src/llm.ts

@@ -1,20 +1,20 @@
 import { GoogleAuth } from 'google-auth-library';
 import { GoogleGenAI } from '@google/genai';
+import AnthropicVertex from '@anthropic-ai/vertex-sdk';
 import { v4 as uuidv4 } from 'uuid';
 
 // =============================================================================
 // Unified LLM client — OpenAI-compatible message format throughout
 //
-// Two backends:
+// Three backends:
-//   VertexOpenAIClient  — for GLM-5 and other Vertex MaaS models (openai-compat endpoint)
+//   VertexOpenAIClient    — GLM-5 and other Vertex MaaS models (openai-compat endpoint)
-//   GeminiClient        — for Gemini Flash/Pro via @google/genai SDK (API key)
+//   GeminiClient          — Gemini Flash/Pro via @google/genai SDK (API key)
+//   AnthropicVertexClient — Claude models via Anthropic Messages API on Vertex (us-east5)
 //
 // Model tier defaults (overridable via TIER_A/B/C_MODEL env vars):
-//   Tier A: gemini-2.5-flash    — routing, summaries, log parsing (API key, high quota)
+//   Tier A: gemini-2.5-flash       — routing, summaries (API key, high quota)
-//   Tier B: zai-org/glm-5-maas  — coding, feature work (Vertex MaaS, retry on 429)
+//   Tier B: claude-sonnet-4-6      — coding, feature work (Anthropic Vertex, us-east5)
-//   Tier C: zai-org/glm-5-maas  — complex decisions, escalation
+//   Tier C: claude-sonnet-4-6      — complex decisions
-//
-// Claude models are NOT available in this GCP project — do not use anthropic/* IDs.
 // =============================================================================
 
 // ---------------------------------------------------------------------------
@@ -288,6 +288,128 @@ function toGeminiContents(messages: LLMMessage[]): any[] {
     return contents;
 }
 
+// ---------------------------------------------------------------------------
+// Anthropic Vertex client
+// Used for: claude-* models via Vertex AI (proper Anthropic Messages API)
+// Handles tool_calls by converting to/from Anthropic's tool_use blocks.
+// ---------------------------------------------------------------------------
+
+export class AnthropicVertexClient implements LLMClient {
+    modelId: string;
+    private projectId: string;
+    private region: string;
+
+    constructor(modelId: string, opts?: { projectId?: string; region?: string }) {
+        // Strip the "anthropic/" prefix if present — the SDK uses bare model names
+        this.modelId = modelId.startsWith('anthropic/') ? modelId.slice(10) : modelId;
+        this.projectId = opts?.projectId ?? process.env.GCP_PROJECT_ID ?? 'master-ai-484822';
+        this.region = opts?.region ?? process.env.CLAUDE_REGION ?? 'us-east5';
+    }
+
+    private buildClient(): AnthropicVertex {
+        const b64Key = process.env.GCP_SA_KEY_BASE64;
+        if (b64Key) {
+            try {
+                const jsonStr = Buffer.from(b64Key, 'base64').toString('utf8');
+                const credentials = JSON.parse(jsonStr);
+                return new AnthropicVertex({
+                    projectId: this.projectId,
+                    region: this.region,
+                    googleAuth: new GoogleAuth({ credentials, scopes: ['https://www.googleapis.com/auth/cloud-platform'] }) as any,
+                });
+            } catch {
+                console.warn('[llm] AnthropicVertex: SA key decode failed, falling back to metadata server');
+            }
+        }
+        return new AnthropicVertex({ projectId: this.projectId, region: this.region });
+    }
+
+    async chat(messages: LLMMessage[], tools?: LLMTool[], maxTokens = 8192): Promise<LLMResponse> {
+        const client = this.buildClient();
+
+        const system = messages.find(m => m.role === 'system')?.content ?? undefined;
+        const nonSystem = messages.filter(m => m.role !== 'system');
+
+        // Convert OpenAI message format → Anthropic format
+        const anthropicMessages: any[] = nonSystem.map(m => {
+            if (m.role === 'assistant') {
+                const parts: any[] = [];
+                if (m.content) parts.push({ type: 'text', text: m.content });
+                for (const tc of m.tool_calls ?? []) {
+                    parts.push({
+                        type: 'tool_use',
+                        id: tc.id,
+                        name: tc.function.name,
+                        input: JSON.parse(tc.function.arguments || '{}'),
+                    });
+                }
+                return { role: 'assistant', content: parts.length === 1 && parts[0].type === 'text' ? parts[0].text : parts };
+            }
+            if (m.role === 'tool') {
+                return {
+                    role: 'user',
+                    content: [{ type: 'tool_result', tool_use_id: m.tool_call_id, content: m.content ?? '' }],
+                };
+            }
+            return { role: 'user', content: m.content ?? '' };
+        });
+
+        const anthropicTools = (tools ?? []).map(t => ({
+            name: t.function.name,
+            description: t.function.description,
+            input_schema: t.function.parameters,
+        }));
+
+        const MAX_RETRIES = 4;
+        const RETRY_STATUSES = new Set([429, 503]);
+
+        for (let attempt = 0; attempt <= MAX_RETRIES; attempt++) {
+            try {
+                const response = await (client.messages.create as Function)({
+                    model: this.modelId,
+                    max_tokens: maxTokens,
+                    system: system ?? undefined,
+                    messages: anthropicMessages,
+                    tools: anthropicTools.length > 0 ? anthropicTools : undefined,
+                });
+
+                const textContent = response.content
+                    .filter((b: any) => b.type === 'text')
+                    .map((b: any) => b.text)
+                    .join('') || null;
+
+                const tool_calls: LLMToolCall[] = response.content
+                    .filter((b: any) => b.type === 'tool_use')
+                    .map((b: any) => ({
+                        id: b.id,
+                        type: 'function' as const,
+                        function: { name: b.name, arguments: JSON.stringify(b.input ?? {}) },
+                    }));
+
+                return {
+                    content: textContent,
+                    reasoning: null,
+                    tool_calls,
+                    finish_reason: response.stop_reason === 'tool_use' ? 'tool_calls' : 'stop',
+                    usage: response.usage
+                        ? { prompt_tokens: response.usage.input_tokens, completion_tokens: response.usage.output_tokens, total_tokens: response.usage.input_tokens + response.usage.output_tokens }
+                        : undefined,
+                };
+            } catch (err: any) {
+                const status = err?.status ?? err?.statusCode ?? 0;
+                if (RETRY_STATUSES.has(status) && attempt < MAX_RETRIES) {
+                    const waitMs = Math.min(2 ** attempt * 2000 + Math.random() * 500, 30_000);
+                    console.warn(`[llm] Anthropic Vertex ${status} on attempt ${attempt + 1}/${MAX_RETRIES + 1} — retrying in ${Math.round(waitMs / 1000)}s`);
+                    await new Promise(r => setTimeout(r, waitMs));
+                    continue;
+                }
+                throw new Error(`Anthropic Vertex error: ${err?.message ?? String(err)}`);
+            }
+        }
+        throw new Error('Anthropic Vertex: exceeded max retries');
+    }
+}
+
 // ---------------------------------------------------------------------------
 // Factory — createLLM(modelId | tier)
 // ---------------------------------------------------------------------------
@@ -296,8 +418,8 @@ export type ModelTier = 'A' | 'B' | 'C';
 
 const TIER_MODELS: Record<ModelTier, string> = {
     A: process.env.TIER_A_MODEL ?? 'gemini-2.5-flash',
-    B: process.env.TIER_B_MODEL ?? 'zai-org/glm-5-maas',
+    B: process.env.TIER_B_MODEL ?? 'claude-sonnet-4-6',
-    C: process.env.TIER_C_MODEL ?? 'zai-org/glm-5-maas'
+    C: process.env.TIER_C_MODEL ?? 'claude-sonnet-4-6'
 };
 
 export function createLLM(modelOrTier: string | ModelTier, opts?: { temperature?: number }): LLMClient {
@@ -309,6 +431,10 @@ export function createLLM(modelOrTier: string | ModelTier, opts?: { temperature?
         return new GeminiClient(modelId, opts);
     }
 
+    if (modelId.startsWith('anthropic/') || modelId.startsWith('claude-')) {
+        return new AnthropicVertexClient(modelId);
+    }
+
     return new VertexOpenAIClient(modelId, { temperature: opts?.temperature });
 }
 

src/server.ts

@@ -550,18 +550,24 @@ app.post('/agent/approve', async (req: Request, res: Response) => {
 // ---------------------------------------------------------------------------
 
 app.post('/generate', async (req: Request, res: Response) => {
-    const { prompt, model } = req.body as { prompt?: string; model?: string };
+    const { prompt, model, region } = req.body as { prompt?: string; model?: string; region?: string };
     if (!prompt) { res.status(400).json({ error: '"prompt" is required' }); return; }
 
+    // Allow overriding CLAUDE_REGION per-request for testing
+    const prevRegion = process.env.CLAUDE_REGION;
+    if (region) process.env.CLAUDE_REGION = region;
+
     try {
         const llm = createLLM(model ?? 'A', { temperature: 0.3 });
         const messages: import('./llm').LLMMessage[] = [
             { role: 'user', content: prompt }
         ];
         const response = await llm.chat(messages, [], 8192);
-        res.json({ reply: response.content ?? '' });
+        res.json({ reply: response.content ?? '', model: llm.modelId });
     } catch (err) {
-        res.status(500).json({ error: err instanceof Error ? err.message : String(err) });
+        res.status(500).json({ error: err instanceof Error ? err.message : String(err), model });
+    } finally {
+        if (region) process.env.CLAUDE_REGION = prevRegion ?? '';
     }
 });