Faramesh offers a governance‑as‑code runtime that sits alongside AI agents in production, intercepting every tool call and evaluating it against a single declarative policy file.
Funding
Funding not disclosed
Founders
Product
Problem
AI agents in production can be compromised by prompt injections, misbehaving model steps, or stolen API credentials, leading to unauthorized tool usage and potentially harmful actions. These failures often go undetected until after the agent has executed privileged operations, exposing organizations to security and compliance risks.
Solution
Faramesh provides a runtime governance-as-code layer that intercepts every tool call made by an AI agent and evaluates it against a single, declarative policy file. The lightweight daemon runs alongside the agent, enforcing identity verification, credential checks, and custom policy decisions before any action is performed. Each decision and tool invocation is recorded in a tamper‑evident, hash‑chained audit log that can be verified offline for compliance and forensic analysis. The solution integrates directly with popular frameworks such as LangChain, LangGraph, CrewAI, OpenAI, Claude, and others, requiring only one runtime component and one governance file on the user’s own infrastructure.
Target Audience
Primary customers are developers and engineering teams building production AI agents using LangChain, LangGraph, CrewAI, or similar frameworks, as well as enterprises that need to enforce security and compliance for AI‑driven automation.
Features
- Single governance.fms file defines all identity, credential, and policy rules for the agent
- Runtime daemon intercepts and decides on every tool call (permit, defer, deny) before execution
- Hash‑chained, tamper‑evident audit trail for each decision, enabling offline verification and compliance reporting
- Framework‑agnostic integration with LangChain, LangGraph, CrewAI, OpenAI Agents, Claude Code, and other major AI toolkits
- Conditional rule engine supports checks on request amount, host, method, and any agent‑provided arguments
- Quickstart workflow generates a starter policy file and configures the daemon with three commands