1 / 1
【Overview】 Fact0 is a self-hosted tool for keeping a record of what your agents actually did. It captures execution spans, tool results, supported model content and audit events from Python agents through its SDK, and it captures supported prompts, assistant text, tool calls and errors from Claude Code sessions through a plugin and a collector. Everything arrives in one local workspace that you run on your own infrastructure. The gap it fills is the one between a run finishing and being able to explain it. Rather than reconstructing a failure from scattered logs, you open the recorded steps with their inputs and outputs and move between a graph, a timing waterfall and the individual spans. From there you can replay the saved execution events at your own pace and walk through the run as it happened. It is aimed at developers already running Python agents who want a durable record of what those agents did, at Claude Code users who want their coding sessions on file, and at operators who need captured agent data and signing keys to stay inside their own environment. The project is MIT licensed and built around one owner and one workspace, so the emphasis is on an inspectable local record rather than a shared team service. 【Key features】 • Python agent capture: Records execution spans, tool results, supported model content and audit events from Python agents via the Python SDK. • Claude Code session capture: Captures supported prompts, assistant text, tool calls and errors from Claude Code sessions through a plugin and collector. • Execution history inspection: Shows recorded steps with inputs and outputs across a graph, timing waterfall and recorded spans. • Hash-chained audit records: Lets you check stored events for consistency against their hashes and preceding links. • Execution replay: Replays saved execution events at your own pace to walk through a recorded run. • Local-first deployment: Starts the API, dashboard and PostgreSQL with Docker Compose so data and signing keys stay with your instance. 【Use cases】 1. Inspect captured prompts, tool calls and results. 2. See where an agent run failed. 3. Replay recorded execution traces. 【Pros and cons】 Pros: - Suits developers who want to answer "what did my agent actually do" months later, because the record is inspectable rather than a stream of log lines. - A good fit for operators with data-residency constraints, since the dashboard, database and signing keys live on infrastructure they control. - The hash chaining and signed exports give reviewers something concrete to check rather than a claim of correctness. Cons: - Python capture depends on your own instrumentation, so coverage is only as good as what you wire up. - Claude Code capture relies on supported hooks and transcript formats; missing or unsupported content is marked partial or unavailable. - Hash chains do not prove complete capture, truthful source data, or protection from an administrator replacing the whole history. - The ZIP archive in a signed export is not itself signed. - Replay reconstructs history; it does not rerun the agent. - The current supported release is for one owner and one workspace. 【FAQ】 Q: Does replaying a trace re-execute my agent? A: No. Replay reconstructs the recorded history from saved execution events so you can step through it at your own pace; the agent itself is not run again. Q: Where do the captured data and signing keys live? A: On your own infrastructure. The API, dashboard and PostgreSQL are started with Docker Compose, so both the captured data and the signing keys stay with your instance. Q: Can the hash-chained records prove that every agent action was captured? A: They let you check stored events for consistency against their hashes and preceding links, but they do not prove complete capture, truthful source data, or protection from an administrator replacing the whole history.
free
WPS AI is an AI assistant built into WPS Office that helps users write documents, generate presentation outlines and…
Baidu Comate is an AI coding assistant built by Baidu on its ERNIE model. It goes beyond simple autocomplete: it…
UniConverter is an AI media conversion and enhancement tool for video, audio, and images. It combines everyday…
Added
11/17/2026
Views
23
Clicks
2
6 similar tools to consider instead of Fact0