Deploying AgentField with Docker: Build a Multi-Agent Research Pipeline

🐳 Deploying AgentField with Docker: Your First Multi-Agent System

Let's get AgentField running and build a real multi-agent system. Not a toy example — a research pipeline where one agent searches, another analyzes, and a third summarizes. Each agent runs as an independent service with its own identity.

📦 Step 1: Start the Control Plane

AgentField uses a control plane + agent architecture. First, run the control plane (it handles identity, orchestration, and storage):

docker pull agentfield/control-plane:latest
docker run -d --name agentfield-cp \
  -p 8080:8080 \
  -e AGENTFIELD_STORAGE_MODE=sqlite \
  agentfield/control-plane:latest

⏱️ Time: ~30 seconds to pull (image is ~200MB).

⚠️ I forgot the storage env: Started it without STORAGE_MODE and it fell back to a temporary in-memory db. Lost all my agents on restart. Use sqlite for testing, postgres for production.

🔧 Step 2: Create Your First Agent

With the control plane running at http://localhost:8080, use the AgentField CLI or API to create an agent. I'll use the API — it's cleaner:

# Create a research agent with its own identity
curl -X POST http://localhost:8080/api/v1/agents \
  -H "Content-Type: application/json" \
  -d '{
    "name": "research-agent",
    "description": "Searches the web and returns relevant content",
    "provider": "openai",
    "model": "gpt-5",
    "tools": ["web_search", "web_scrape"],
    "system_prompt": "You are a research agent. Search the web for information and return structured results."
  }'

The response includes an agent ID and API token. Save both — the token is how other agents authenticate when calling this one.

🔄 Step 3: Build a Multi-Agent Pipeline

Now create two more agents and connect them:

1. Analyzer Agent — takes raw research content and extracts key insights
2. Summarizer Agent — takes insights and generates a structured report

Each agent gets its own identity, its own API token, and its own scope of work. Here's the cool part: the analyzer agent can call the researcher agent using its API token, and the summarizer can call the analyzer. AgentField handles all the inter-agent auth.

📊 Step 4: Monitor Everything

Open the AgentField dashboard at http://localhost:8080/ui and you'll see:

• Each agent's status (running, idle, error)
• Full trace logs for every agent invocation
• Audit trail showing which agent called which
• Performance metrics (response time, token usage, error rate)

When my research agent failed on a complex query, I opened the trace log and saw it hit a rate limit on the search API. Fixed it by adding a retry delay in the system prompt. Total debugging time: 3 minutes.

⚡ Performance

🎯 Who Is This For?

If you're running one agent for personal use, AgentField is probably overkill. But if you're building a system where multiple agents need to collaborate, each with different roles, permissions, and observability requirements — this is exactly what you're looking for. It's the difference between "my script works on my machine" and "my services run in production."