What does it do?
Fairway certifies AI agents before they ship, enforces policy on every action at runtime, and keeps an audit trail a compliance reviewer can read.
What problem does it solve?
AI agents decide, call tools and act without a deterministic, auditable path. Traditional security and monitoring tools do not cover that. Teams cannot show an agent was safe before launch or that each action was governed afterward. Fairway gates each action and records the evidence.
Who is it for?
Teams deploying autonomous or tool-using AI agents into real workflows, especially regulated ones. Two audiences: enterprise buyers (CISOs, risk officers, CTOs, heads of AI) and AI builders shipping on LangGraph, CrewAI, MCP or custom stacks. Personas: AI/platform engineers, security/CISO, risk/compliance, QA/eval.
How is it different from other agent tools?
It is not an agent framework. It is a trust layer inside the harness of an existing agent: certify before launch, enforce policy inline at runtime, audit after. The decision path is deterministic-first (four model-free gates run before any LLM sees the payload), and it can run self-hosted so data stays in the customer network.
How do I get started?
pip install "agentrust-py[embedded,retry]", then wrap an agent with @harness or call auto_instrument(). Or run the Governance Baseline: full pipeline, enforcement switched off, about 20 minutes to install.
Integrations
LangGraph, CrewAI, MCP, plus OpenAI and LangChain via auto_instrument(). TypeScript and Python SDKs.
