Agentfootprint
A TypeScript agent framework that records every read, write, decision, and tool call an agent makes as connected evidence, so a developer can trace a wrong answer back to what the model was given rather than only what the code did.
Agentfootprint targets a class of agent bug that classical logging cannot explain: the code is correct and the infrastructure is healthy, but the agent still picked the wrong tool, believed a wrong fact, or acted on stale memory, because something in what the model was given led it astray. It structures an agent's run so that every context injection, read, write, decision, and tool call is recorded as connected evidence the moment it happens, turning a debugging question into something a developer can query rather than grep for in logs.
It is used to build agents with skills, steering, retrieval-augmented generation, memory, and control flow, and it works with common model providers including Anthropic, OpenAI, Amazon Bedrock, Google Gemini, and Ollama through a swappable provider import. A companion package renders a recorded run as a replayable, frame-by-frame visualization generated directly from the run's own trace, and the framework can be tried offline with no API key before wiring in a real provider.