Interbolt
An open-source Python library for provenance-based security enforcement in AI agent systems.
Interbolt is an open-source Python library for provenance-based security enforcement in AI agent systems.
You mark untrusted data where it enters an agent. Interbolt propagates that mark through your code and evaluates a YAML and CEL policy at each guarded tool call. The policy returns allow, block, or require approval based on where the call's arguments came from. Decisions are deterministic and run in process, with no model and no network call involved.
Results for Interbolt as a prompt injection defense on the AgentDojo benchmark, including the utility cost, are in Prompt Injection Defense by Provenance: AgentDojo Benchmark Results.
Interbolt is published under the Apache 2.0 license. It is no longer in active product development.