Launching today
HarnessWeaver
One spec for the controls every AI app needs
2 followers
One spec for the controls every AI app needs
2 followers
Open-source AI Harness framework for the controls every AI app needs: budgets, guardrails, tool permissions, approvals and tracing, set in one harness.yaml. Your AI framework executes; your gateway connects; your observability platform monitors; HarnessWeaver governs.
Free
Launch Team
Render Sandboxes: a computer for every agentRun untrusted code safely. Idle sandboxes cost nothing
Promoted
Hi Product Hunters,
I’m Muthu, and I built HarnessWeaver.
Every AI app needs the same controls around the model: PII redaction, spend caps, validated output, tool permissions, approvals and tracing. Teams rewrite them inside each app, and the platform team can’t see or require them.
HarnessWeaver is a framework each team uses to build its own harness. You keep your executor (raw SDK calls, Pydantic AI, the Vercel AI SDK), describe the controls in one harness.yaml, and it enforces them at 13 hook points.
What I think is different:
A platform team’s org baseline locks controls. An app that tries to switch one off fails to load.
Bounds are errors, not clamps.
New guards start in shadow mode.
Every example and test runs offline, with no API keys.
It’s Apache-2.0, for Python and TypeScript, and any other language through a sidecar. The quickstart takes about fifteen minutes: https://github.com/Wise-DreamerZ/HarnessWeaver