RunWitness Alpha
Alpha launch

Autonomous agents with receipts.

RunWitness is shipping as a local-first witness layer for agent work: run the task, observe every important action, approve risky steps, and export a receipt humans can inspect.

CI Version v0.1.0 License Apache-2.0 Node 22.19+
Try the demo Clone on GitHub Project board
Animated RunWitness receipt flow showing task intake, policy review, approval, command capture, and receipt export.
From task to policy decision to receipt, RunWitness keeps the evidence visible.

Shipping now

What the alpha includes

v0.1.0
  • SQLite run ledger with append-only events and timeline APIs.
  • Policy checks, approval records, sandbox preflight, and rollback evidence.
  • Receipt JSON and Markdown exports with file, command, and test evidence.
  • RunWitness OpenClaw plugin, Hermes skill pack example, and stdio MCP server.

Try it locally

First command

$ git clone https://github.com/karurikwao/runwitness.git
$ cd runwitness
$ npm ci
$ npm run verify

Ecosystem status

Compatibility wording

RunWitness integrations are official from the RunWitness repo side. OpenClaw, Hermes, OpenAI, Anthropic, and other hosts become upstream-official only if those ecosystems accept, list, or ship them.

Alpha limits

Not pretending

This is not a hosted SaaS control plane, universal OS sandbox, complete nested-agent tracer, or upstream-endorsed package across every host.

01 Witness

Run commands and collect observable events.

02 Govern

Apply policies and preserve approvals.

03 Integrate

Bridge OpenClaw, Hermes-style skills, MCP hosts, and local CLIs.

04 Prove

Export receipts, rollback bundles, and launch evidence.