Welcome to AgentTrust
Runtime governance for AI agents: every agent action is checked before it runs and recorded after.
Welcome to AgentTrust
AgentTrust sits between your AI agent and what it can do. Before an agent runs a tool or returns an answer, AgentTrust checks it against your policy, scores the risk, and decides what happens next. Every decision is recorded.
What you get
6
Decision outcomes
approve, block, escalate, retry, request evidence, pending
7
Framework adapters
LangChain, LangGraph, CrewAI and more
2
SDKs
Python and TypeScript
3
Deployment modes
OSS, Embedded, Full Edge
- A decision on every action: approve, block, escalate to a person, or ask for more evidence.
- Confidence and risk scoring for each agent output.
- Policy as code: your rules live in one place and apply to every agent.
- Human review: risky calls go to a review queue.
- An audit trail of what each agent did and why, aligned with ISO/IEC 42001.
- Trust chain for multi-agent pipelines, so you can see where an output came from.
What you need
- Python 3.10 or newer, or Node.js 18 or newer, to use an SDK.
- An agent or app to govern. Any framework works; there are adapters for the common ones.
- For the full production gateway: PostgreSQL 15+ and Redis 7+ (provided by Docker Compose). You do not need either to try it locally.
Get started
Pick one way in. You can use both.
Use the web console when you want to register agents and watch decisions without writing code first.
- Sign in at console.agent-trust.tech.
- Pick or add a workspace.
- Create your first agent from the Agents page.
Use the SDK when you want to govern an agent in your own code. Nothing else needs to run: an embedded gateway starts inside your process.
# Python
pip install "agentrust-sdk[embedded,retry]"
# TypeScript / Node.js
npm install agentrust-sdkThen follow the Quickstart to wrap your first agent function.
Ways to connect an agent
| Method | How | What you get |
|---|---|---|
@harness decorator | Add one line above each tool function | Every call is checked before it runs |
| Framework adapter | Use the adapter for your framework | Governance fits your existing chains, graphs and crews |
| Zero-code setup | Set environment variables | Governance with no source changes, for third-party or legacy code |
| Embedded gateway | Start it in your process | A real gateway for dev, demos and CI with no infrastructure |
| REST API | Call the validate endpoint | Full control from any language |
Framework adapters
Next steps
- Quickstart: run your first governed agent.
- How It Works: the path from agent action to decision.
- Introduction: the full platform overview.