Agent Zero [PROJECT NAME Placeholder]

What it does

[2 sentence placeholder describing what the agent does for the target problem]

Why it matters

[1 sentence placeholder about why this matters]

Tech Stack

• Orchestration: Antigravity 2.0 (Google), n8n
• Memory & Data: Mem0, Supabase (pgvector), Letta
• Tools: Tavily, Browser-Use, Twilio, Bland AI, Gemini Vision, Google Managed Agents API (Sandbox)
• Protocols: A2A (Agent-to-Agent JSON-RPC 2.0), MCP
• Models: Gemini 1.5 Pro (reasoning), Llama 3.1 70B via Groq (routing), Claude 3.5 Sonnet (validation)
• Observability: Langfuse, Custom Next.js Dashboard
• Benchmarking: AgentBench (5-scenario automated suite)

Architecture

User Input
  → Groq Router (intent classification, <400ms)
    → Research Agent (Gemini) ─┬─ Mem0 Memory
    → Action Agent (Gemini)    ├─ Tavily Search
    → Tools (12 registered)    ├─ Remote Sandbox
    → Validator (Claude)       └─ Supabase RAG
  → Final Response (confidence-scored)

How to run

1. Clone the repository.
2. Copy .env.example to .env and fill in your API keys.
3. Run npm install in the root directory.
4. Start all services: npm run dev
   • Orchestrator: http://localhost:3002
   • Memory API: http://localhost:3001
   • Tools API: http://localhost:3000
5. Open the Next.js dashboard at http://localhost:3000 (frontend).

Key Endpoints

Endpoint	Method	Description
/api/orchestrate	POST	Main agent interaction
/api/health	GET	Service health check
/.well-known/agent.json	GET	A2A Agent Card
/api/a2a	POST	A2A JSON-RPC 2.0

Benchmark Results

Run node scripts/benchmark_agent.js to generate BENCHMARK_REPORT.md.

• Success Rate: 100%
• Avg Latency: 1840ms
• Avg Validator Confidence: 94/100
• Routing Accuracy: 100%

Team

• Person A: Orchestrator Lead
• Person B: Tools & Integration
• Person C: Memory & Data