Meet OpenGuardrail#
An open-source action guardrail for AI agents: define the checks, collect the evidence, and verify the decision before a tool acts.
Why use it?#
Bring the policy, check results and evidence for an operation into one traceable assessment. Bind policies to resources so new Runs select the appropriate revision. Your tool can verify the record before acting; your team can inspect it afterward.
How it works#
- Define the requirements. Publish a policy with checks and evidence requirements.
- Submit an assessment. Your agent or integration performs the checks and submits its observations.
- Verify, then act. OpenGuardrail records
ALLOWorDENY. The tool verifies the evidence, validity and business permissions before execution.
Try it#
Run the first-batch check with your coding agent. Follow the failed write to a captured order still marked pending, and inspect the evidence behind the decision to hold the next batch.
For file cleanup without a model, use the native download. To try the Agent plugin, use pi or DeepSeek Harness.
Ready to connect your tool? Read the integration guide.