Treehouse is an external observability and interpretability tool for behavior-tree-driven agents
Project description
🌳 Treehouse
Treehouse is an external observability and interpretability tool for behavior-tree–driven agents.
It lets you observe how behavior unfolds in real time — which branches were evaluated, which paths were taken, and where execution stalled or failed — without interfering with execution.
Treehouse is where you go to watch behavior trees think.
What Treehouse Does
- Consumes semantic execution events
- Reconstructs behavior tree execution over time
- Surfaces decisions in structural, human-readable form
Treehouse treats behavior trees — not logs or metrics — as the unit of meaning.
Event Boundary
Treehouse relies on the shared event boundary defined here:
../docs/event-boundary.md
This boundary is Treehouse’s primary input API. Any runtime that emits compatible events can be observed — Vivarium is a natural fit, not a requirement.
What Treehouse Is Not
- Not a runtime
- Not a controller
- Not coupled to a specific agent framework
Treehouse observes. It does not direct.
Status
Treehouse is early-stage and exploratory.
The focus is on correctness, clarity, and faithful reconstruction of behavior — polish comes later.
Run the behavior. Observe the decisions. Understand the agent.
