What is Faramesh?

Faramesh is the deterministic governance layer described in the checked-in faramesh-core repository. The README is explicit: it sits between the agent and the tools it calls, blocks actions that violate policy, routes some actions to human approval, and writes tamper-evident evidence for audit and compliance.

The security problem is the Action Authorization Boundary. LLMs are stochastic; infrastructure is not. A system prompt is not a control plane. Faramesh makes every tool call cross one hard decision point before anything touches a real system.

Note

This site is intentionally strict about terminology. Use Action Authorization Boundary, Proposed Action, Deterministic Canonicalization, Audit Ledger, and Human-in-the-Loop exactly as defined here.

Faramesh is not an agent framework and it is not a second model watching the first. It is deterministic policy enforcement. The checked-in runtime supports local OSS use, managed Horizon flows, and Nexus enterprise deployment.

Agent -> Faramesh API Server (Policy Engine) -> Tool Executor

Read the Quickstart next, then move to Why Faramesh Exists for the threat model behind the design.