burr

Why we included this project

Most chatbot and agent projects start simple, then get messy the moment the thing has to remember a conversation and survive a restart. Burr keeps that under control by letting you build the application as plain Python actions connected by transitions, so the flow stays visible in the code instead of hiding inside long prompts. Its state handling is the part that stands out: pluggable persisters can save a run's state to your own database, pause mid-flight, and pick up later from the same spot. The built-in web UI then gives you live tracing of those runs on your own infrastructure, which helps a lot when you need to figure out why a production agent made the choice it did. It also works alongside whatever LLM framework you already use, so you get stateful agent tooling without being locked into one vendor's stack.