Alpha preview: pin 3.0.0a3 when you install. A bare pip install agentbreeder still installs the unmaintained 2.x: known issues.
agentbreeder

Legacy 2.x

AgentBreeder 2.x is what a bare pip install gives you today. It is unmaintained, superseded by the focused rebuild, and has known functional and security issues.

Unmaintained: do not use beyond local experiments

AgentBreeder 2.x (PyPI agentbreeder up to 2.7.3, and the matching Docker images) is no longer maintained and will not receive fixes. Its self-hosted API server has known security issues: do not expose it to untrusted users or networks, and do not use it with production data or credentials. Report anything you find through GitHub private vulnerability reporting.

What happened to 2.x

2.x set out to cover agent building, registries, multi-cloud deploys and governance in one platform. A full end-to-end test of 2.7.3 on 2026-11-07 found that much of what its docs described did not work as written. Rather than keep documenting it, AgentBreeder was rebuilt around one job it can do reliably today: deploying an existing agent container behind a deploy-time identity gate. The rest of these docs cover that rebuild. It shares no code with the 2.x API server or dashboard.

Known issues in 2.7.3

These were reproduced on a fresh install. They are not exhaustive.

  • Local deploy reports failure. agentbreeder deploy with a local target reports FAILED with an empty message even when the container came up healthy, and registers nothing.
  • Scaffolded agents. Of the templates agentbreeder init generates, only LangGraph produced an agent that answered end to end.
  • Studio dashboard. The 2.7.2 and 2.7.3 dashboard images render a blank page. Deploying from Studio does not build or run a container.
  • Evaluations. agentbreeder eval run does not call the agent; its scores are simulated.
  • SDK. The documented Python SDK examples do not match the published package, and Agent.deploy() does not deploy.
  • Governance. Access control, audit logging and cost tracking in the API server do not enforce or persist as the 2.x docs described.
  • Cloud targets. AWS, GCP, Azure and Kubernetes deploys were not re-verified and should be treated as unverified.

If you are using 2.x

  • Keep it to local, single-user experiments, and keep the API server off shared networks.
  • Pin the version you have; there will be no 2.7.4.
  • A bare pip install agentbreeder installs 2.7.3. To stay on 2.x explicitly, pin agentbreeder<3.
  • To try the rebuild, install the 3.0.0a3 preview with its exact version, in a fresh virtual environment. It is not an upgrade path: there is no migration from 2.x data or agent.yaml files.

2.x documentation

The 2.x docs are preserved in the repository at the v2.7.3 tag. They describe intended behaviour, which the issues above contradict in places.

On this page