Open Source: Agent Framework Starter Kit

Building a cyborgenic organization should not require six months of infrastructure work before your first agent does anything useful. The gap between "I want AI agents working as a team" and "I have AI agents working as a team" is too wide. Today we are closing it.

agent-framework-starter is a minimal Python starter kit for building AI agent teams. Multi-agent orchestration, NATS messaging, and task management — everything you need to go from zero to a functioning agent team in five minutes.

What It Is

The starter kit is an opinionated, minimal framework for standing up a team of AI agents that can:

• Communicate via NATS JetStream with durable inboxes and reliable delivery
• Delegate tasks between agents with tracking, acknowledgment, and completion verification
• Orchestrate work through a manager agent that assigns tasks based on agent capabilities
• Manage state with persistent task queues that survive agent restarts

It is not a toy. It is a stripped-down version of the same architecture that powers GenBrain AI's cyborgenic organization — the company where AI agents fill real roles and ship real work every day. We removed the GenBrain-specific business logic and kept the structural foundation.

Why We Built This

We talk to teams building multi-agent systems every week. The same conversation happens repeatedly:

"We have agents that work individually. We cannot get them to work together."

The problem is never the individual agent's capability. GPT-4, Claude, Gemini — they are all capable enough. The problem is coordination. How does Agent A tell Agent B to do something? How does Agent B report back? What happens when Agent C fails halfway through a task? Who tracks what is done and what is not?

These are solved problems in distributed systems engineering. But most teams building AI agents are not distributed systems engineers. They are ML engineers, product builders, and startup founders who want agents that collaborate — not a graduate course in message queue design.

The starter kit packages the coordination layer so you can focus on what your agents actually do.

Five-Minute Quickstart

# Clone the repo
git clone https://github.com/GenBrainAI/agent-framework-starter
cd agent-framework-starter

# Start NATS and the example agent team
docker compose up -d

# Your agent team is running:
#   - manager agent (orchestrates work)
#   - worker-alpha (general tasks)
#   - worker-beta (general tasks)

# Assign a task
python cli.py assign "Review the README for typos"

That is it. Three agents running, communicating over NATS, with task assignment and completion tracking. From git clone to working agent team in under five minutes.

What Is Inside

The repository is structured for clarity, not cleverness:

agent-framework-starter/
  agents/
    base.py          # Base agent class with NATS connectivity
    manager.py       # Orchestrator that delegates tasks
    worker.py        # Worker agent template
  messaging/
    inbox.py         # Per-agent durable inbox
    tasks.py         # Task queue with ack/nack/retry
    events.py        # Event broadcasting
  config/
    agents.yaml      # Agent definitions and capabilities
    nats.yaml        # NATS connection and stream config
  docker-compose.yml # One-command local setup
  cli.py             # Command-line task assignment

Base Agent — Every agent extends BaseAgent, which handles NATS connection, inbox subscription, heartbeat, and graceful shutdown. You write the handle_task() method. The framework handles everything else.

Manager Agent — Routes tasks to available workers based on declared capabilities. Tracks task state. Handles timeouts and retries. Escalates failures. This is the pattern described in our multi-agent architecture documentation, implemented and ready to run.

Task Management — Tasks flow through a defined lifecycle: assigned, accepted, in-progress, completed, or failed. Each transition is recorded. Nothing gets lost.

Designed for Extension

The starter kit is deliberately minimal. It gives you the coordination skeleton. You add the intelligence.

Common extensions we expect teams to build:

• Specialized workers — Replace the generic worker with agents that have domain expertise: code review, data analysis, content generation, customer support
• LLM integration — The base agent template includes hooks for connecting any LLM provider. Bring your own model, API key, and prompt strategy
• Custom routing — The manager's task routing is pluggable. Route by skill, by load, by priority, or by whatever logic your organization needs
• Persistence — Swap the default in-memory state for PostgreSQL, Redis, or whatever your stack uses
• Observability — Add OpenTelemetry traces, Prometheus metrics, structured logging

The goal is to get you to "agents collaborating" fast, then get out of your way as you build something production-grade.

From Starter Kit to Cyborgenic Organization

The starter kit is the first step. The path from here to a full cyborgenic organization looks like this:

1. Start with the kit — Get familiar with agent communication and task delegation patterns
2. Define real roles — Replace generic workers with agents that own specific functions in your organization
3. Add NATS patterns — Integrate the nats-agent-patterns library for production-grade messaging
4. Deploy to production — Move from Docker Compose to your actual infrastructure
5. Scale to agent.ceo — When you need enterprise orchestration, agent.ceo provides the full platform: scheduling, verification, cross-agent knowledge sharing, and more

We built agent.ceo to run GenBrain AI. We are open-sourcing the foundation so you can build your first agent team without starting from scratch.

Get Started

GitHub: github.com/GenBrainAI/agent-framework-starter

Clone it. Run the quickstart. Build your first agent team today. If you are new to AI agents, start with our guide on what AI agents actually are and then come back here to build one.

Star the repo. File issues. Contribute patterns. The more teams building cyborgenic organizations, the faster this future arrives.

GenBrain AI is the company behind agent.ceo, building the next generation of autonomous agent orchestration.

Try agent.ceo

SaaS — Get started with 1 free agent-week at agent.ceo.

Enterprise — For private installation on your own infrastructure, contact enterprise@agent.ceo.

agent.ceo is built by GenBrain AI — a GenAI-first autonomous agent orchestration platform. General inquiries: hello@agent.ceo | Security: security@agent.ceo