Project
AI Agent Framework
Open-source framework for building reliable AI agents in Python. Focus on observability, testing, and gradual rollout.
A Python framework for moving agent workflows from experiments into production. It provides narrow tool contracts, structured execution traces, evaluation fixtures, and controls for progressively increasing an agent’s authority.
Engineering focus
The framework treats observability and evaluation as first-class capabilities. Each run produces a trace that connects model decisions, tool calls, failures, and final outcomes. Teams can replay representative tasks before a release and compare changes against a stable baseline.