Burr

Burr is an open-source framework by DAGWorks (the team behind Hamilton) for building stateful AI applications as state machines, released around 2024. You express your application as actions (nodes that read and update an immutable state dictionary) connected by transitions with conditions, giving you explicit control over complex decisions, human-in-the-loop steps, and idempotent, self-persisting workflows.

The core is a dependency-free low-abstraction Python library with async support, built-in state persistence (including PostgreSQL and SQLite persisters for checkpointing), a telemetry UI for introspection and debugging of execution, and integrations for state storage and tracking. Because state transitions are explicit, Burr applications are easy to reason about, test, and debug — a deliberate contrast to magic-heavy agent frameworks.

Limitation: Burr is a lower-level building block, not a batteries-included agent platform — you write the agent logic yourself, and there is no hosted product or managed deployment. Teams wanting out-of-the-box agents should look at higher-level frameworks instead.

pip install burr, then define actions and transitions with the ApplicationBuilder API.