Skip to content

DOCKER.md

Latest commit

 

History

History
384 lines (295 loc) · 8.53 KB

DOCKER.md

File metadata and controls

384 lines (295 loc) · 8.53 KB

Docker Deployment Guide

This guide explains how to run the Statistics Agent Team using Docker.

Note: For local development without Docker, see the main README.

Overview

The Docker setup runs all four agents in a single container:

  • Research Agent (port 8001): Web search for source URLs
  • Synthesis Agent (port 8004): Extract statistics from URLs using LLM ⭐ NEW
  • Verification Agent (port 8002): Verifies statistics from sources
  • Eino Orchestration Agent (port 8003): Coordinates the 4-agent workflow

Quick Start

Using Docker Compose (Recommended)

  1. Set up environment variables

Create a .env file in the project root:

# LLM Provider (gemini, claude, openai, ollama)
LLM_PROVIDER=gemini

# LLM API Keys (provide at least one)
GEMINI_API_KEY=your_gemini_api_key_here
CLAUDE_API_KEY=your_claude_api_key_here
OPENAI_API_KEY=your_openai_api_key_here

# Search API Keys (required for real web search)
SEARCH_PROVIDER=serper
SERPER_API_KEY=your_serper_api_key_here
# OR use SerpAPI
# SEARCH_PROVIDER=serpapi
# SERPAPI_API_KEY=your_serpapi_key_here

# For Ollama (local models)
OLLAMA_URL=http://host.docker.internal:11434
OLLAMA_MODEL=llama2

# Optional: Override default models
# GEMINI_MODEL=gemini-2.0-flash-exp
# CLAUDE_MODEL=claude-3-5-sonnet-20241022
# OPENAI_MODEL=gpt-4
  1. Start the agents
docker-compose up -d
  1. Check status
docker-compose ps
docker-compose logs -f
  1. Test the orchestration endpoint
curl -X POST http://localhost:8003/orchestrate \
  -H "Content-Type: application/json" \
  -d '{
    "topic": "climate change",
    "min_verified_stats": 5,
    "max_candidates": 20,
    "reputable_only": true
  }'
  1. Stop the agents
docker-compose down

Using Docker Directly

  1. Build the image
docker build -t stats-agent-team .
  1. Run the container
docker run -d \
  --name stats-agent-eino \
  -p 8001:8001 \
  -p 8002:8002 \
  -p 8003:8003 \
  -p 8004:8004 \
  -e LLM_PROVIDER=gemini \
  -e GEMINI_API_KEY=your_api_key_here \
  -e SEARCH_PROVIDER=serper \
  -e SERPER_API_KEY=your_serper_key_here \
  stats-agent-team
  1. View logs
docker logs -f stats-agent-eino
  1. Stop the container
docker stop stats-agent-eino
docker rm stats-agent-eino

Service Endpoints

Once running, the following endpoints are available:

Research Agent (Port 8001)

  • POST http://localhost:8001/research - Web search for source URLs
  • GET http://localhost:8001/health - Health check

Synthesis Agent (Port 8004) ⭐ NEW

  • POST http://localhost:8004/synthesize - Extract statistics from URLs
  • GET http://localhost:8004/health - Health check

Verification Agent (Port 8002)

  • POST http://localhost:8002/verify - Verify statistics
  • GET http://localhost:8002/health - Health check

Orchestration Agent (Port 8003)

  • POST http://localhost:8003/orchestrate - Full 4-agent workflow
  • GET http://localhost:8003/health - Health check

Note: While all agents are exposed for testing and troubleshooting, typical usage only requires calling the orchestration endpoint.

Configuration

Environment Variables

Variable Description Default Required
LLM_PROVIDER LLM provider to use gemini Yes
GEMINI_API_KEY Google Gemini API key - If using Gemini
CLAUDE_API_KEY Anthropic Claude API key - If using Claude
OPENAI_API_KEY OpenAI API key - If using OpenAI
OLLAMA_URL Ollama server URL http://host.docker.internal:11434 If using Ollama
GEMINI_MODEL Gemini model name gemini-2.0-flash-exp No
CLAUDE_MODEL Claude model name claude-3-5-sonnet-20241022 No
OPENAI_MODEL OpenAI model name gpt-4 No
OLLAMA_MODEL Ollama model name llama2 No
SEARCH_PROVIDER Search provider serper No
SERPER_API_KEY Serper API key - For real search
SERPAPI_API_KEY SerpAPI key - Alternative search

LLM Provider Options

Gemini (Default)

LLM_PROVIDER=gemini
GEMINI_API_KEY=your_key_here

Claude

LLM_PROVIDER=claude
CLAUDE_API_KEY=your_key_here

OpenAI

LLM_PROVIDER=openai
OPENAI_API_KEY=your_key_here

Ollama (Local)

LLM_PROVIDER=ollama
OLLAMA_URL=http://host.docker.internal:11434
OLLAMA_MODEL=llama2

API Examples

Orchestrate Full Workflow

curl -X POST http://localhost:8003/orchestrate \
  -H "Content-Type: application/json" \
  -d '{
    "topic": "renewable energy adoption",
    "min_verified_stats": 10,
    "max_candidates": 30,
    "reputable_only": true
  }'

Response:

{
  "topic": "renewable energy adoption",
  "statistics": [
    {
      "name": "Global renewable capacity growth",
      "value": 83,
      "unit": "%",
      "source": "International Energy Agency",
      "source_url": "https://www.iea.org/...",
      "excerpt": "Renewable capacity grew by 83% in 2023...",
      "verified": true,
      "date_found": "2025-12-13T10:30:00Z"
    }
  ],
  "total_candidates": 25,
  "verified_count": 12,
  "failed_count": 13,
  "timestamp": "2025-12-13T10:30:15Z"
}

Research Only

curl -X POST http://localhost:8001/research \
  -H "Content-Type: application/json" \
  -d '{
    "topic": "electric vehicles",
    "min_statistics": 5,
    "max_statistics": 10,
    "reputable_only": true
  }'

Verification Only

curl -X POST http://localhost:8002/verify \
  -H "Content-Type: application/json" \
  -d '{
    "candidates": [
      {
        "name": "EV market share",
        "value": 18,
        "unit": "%",
        "source": "Bloomberg NEF",
        "source_url": "https://about.bnef.com/...",
        "excerpt": "Electric vehicles reached 18% market share..."
      }
    ]
  }'

Troubleshooting

Container won't start

Check logs:

docker-compose logs -f stats-agent-eino

Common issues:

  • Missing API keys: Ensure you've set the appropriate *_API_KEY variable
  • Port conflicts: Another service using ports 8001-8003
  • Invalid LLM provider: Use gemini, claude, openai, or ollama

Health check failing

Test individual agents:

curl http://localhost:8001/health  # Research
curl http://localhost:8002/health  # Verification
curl http://localhost:8003/health  # Orchestration

Ollama connection issues

If using Ollama on the host:

  • Use http://host.docker.internal:11434 (Mac/Windows)
  • Use http://172.17.0.1:11434 (Linux)
  • Ensure Ollama is running: ollama serve

View agent communication

Enable debug logging:

docker-compose logs -f stats-agent-eino | grep "Orchestration\|Research\|Verification"

Building for Production

Multi-platform builds

docker buildx build \
  --platform linux/amd64,linux/arm64 \
  -t stats-agent-team:latest \
  .

Optimize image size

The Dockerfile uses multi-stage builds and Alpine Linux for minimal size:

  • Builder stage: ~1.2GB (includes Go toolchain)
  • Runtime stage: ~50MB (only binaries + Alpine)

Security scanning

docker scan stats-agent-team:latest

Performance Tuning

Resource limits (docker-compose.yml)

services:
  stats-agent-eino:
    deploy:
      resources:
        limits:
          cpus: '2.0'
          memory: 2G
        reservations:
          cpus: '1.0'
          memory: 1G

Timeout configuration

Agents have built-in HTTP timeouts:

  • Research: 30s read/write, 60s idle
  • Verification: 45s read/write, 90s idle
  • Orchestration: 60s read/write, 120s idle

Integration with Other Services

Behind a reverse proxy (nginx)

upstream stats_orchestration {
    server localhost:8003;
}

server {
    listen 80;
    server_name stats.example.com;

    location / {
        proxy_pass http://stats_orchestration;
        proxy_set_header Host $host;
        proxy_set_header X-Real-IP $remote_addr;
    }
}

Kubernetes deployment

See KUBERNETES.md for Kubernetes manifests (if available).

Monitoring

Health checks

Docker Compose includes automatic health checks every 30 seconds.

Metrics

To add Prometheus metrics, mount a volume for logs:

volumes:
  - ./logs:/var/log/stats-agent

References