AgentBreeder™ — v2.0

The only agent platform that doesn't pick a winner.

Build with anyone's framework. Deploy to anyone's cloud. Govern automatically. One YAML, one command — Apache 2.0, no vendor lock-in.

PyPI PyPI Downloads npm Python License CI Coverage PRs Welcome

LangGraph OpenAI Agents Claude SDK CrewAI Google ADK MCP

Quick Start · Install · Docs · Contributing

Your company has 47 AI agents. Nobody knows what they cost, who approved them, or which ones are still running. Three teams built the same summarizer. The security team hasn't audited any of them.

AgentBreeder fixes this.

Write one agent.yaml. Run agentbreeder deploy. Your agent is live — with RBAC, cost tracking, audit trail, and org-wide discoverability. Automatic. Not optional.

The Problem

AI coding tools make it easy to build agents. Nobody has made it easy to ship them responsibly.

What happens today What happens with AgentBreeder
Every framework has its own deploy story One YAML, any framework, any cloud
No RBAC — anyone deploys anything RBAC validated before the first container builds
No cost tracking — $40k surprise cloud bills Cost attributed per team, per agent, per model
No audit trail — "who deployed that?" Every deploy logged with who, what, when, where
No discoverability — duplicate agents everywhere Org-wide registry — search before you build
Governance is bolted on after the fact Governance is a structural side effect of deploying

Governance is not configuration. It is a side effect of the pipeline. There is no way to skip it.

How It Works

Eight atomic steps run in sequence:

parse → RBAC check → *(approval gate if required)* → resolve deps → build container → provision infra → deploy → health check → register. If any step fails, the entire deploy rolls back.

Three Ways to Build

All three tiers compile to the same internal format. Same deploy pipeline. Same governance. No lock-in.

Tier Who How Eject to
No Code PMs, analysts, citizen builders Visual drag-and-drop canvas — pick model, tools, prompts from the registry Low Code
Low Code ML engineers, DevOps Write agent.yaml in any IDE Full Code (agentbreeder eject)
Full Code Senior engineers, researchers Python/TS SDK with full programmatic control —

Use with Claude Code

Install the /agent-build architect as a Claude Code plugin:

claude plugin marketplace add agentbreeder/agentbreeder
claude plugin install agent-build@agentbreeder

Then run /agent-build in Claude Code to scaffold an agent (it recommends framework, model, memory, RAG, and deploy target, then generates agent.yaml + code).

Documentation

User docs (guides, references, examples) — agentbreeder.io/docs
How-To guides Install, deploy, orchestrate, evaluate
Quickstart Full local platform in one command
Self-hosting Run the platform on your own Kubernetes via Helm (deploy/helm/agentbreeder)
CLI reference All commands and flags
SDK reference Python + TypeScript full-code SDK

For contributors — internal engineering references in this repo:
Contributing How to contribute — setup, standards, PR process
Architecture Platform architecture — deploy pipeline, abstractions, data model
Design Feature design docs — RBAC, LiteLLM gateway, polyglot agents

Changelog · Roadmap · Issues · Discussions · Discord · Apache 2.0 · Trademark · Code of conduct · CLA · Security · Governance