AgentTrust Docs

Overview

Frequently asked questions about AgentTrust, organized by category.

AgentTrust is a runtime governance platform for AI agents. It validates agent outputs, scores confidence and risk, enforces policy decisions, and maintains an audit trail — before outputs reach users or downstream systems.

How is AgentTrust different from observability tools?

Observability tools (LangSmith, Datadog LLM) record what happened. AgentTrust intervenes — it can block, escalate, or require evidence before outputs proceed. It also enforces policy-as-code and human review queues.

Do I need a gateway to use AgentTrust?

No. OSS mode works without any gateway (schema validation only). Embedded mode runs a gateway in-process (SQLite). Full edge requires PostgreSQL + Redis.

What agent frameworks are supported?

LangChain, LangGraph, CrewAI, AutoGen, Claude Agents, OpenAI Agents, MCP, FastAPI, Django, n8n, Express/Node, and any custom Python/HTTP agent via @harness or direct client.

Installation & Setup

What Python version is required?

Python ≥ 3.10. Node.js ≥ 18 for TypeScript SDK.

How long does setup take?

Embedded quickstart: ~30 seconds
Docker Compose full stack: ~5 minutes
Kubernetes production: ~30–60 minutes

Can I add governance without changing my code?

Yes. Use auto_instrument(), autoload (PYTHONSTARTUP), or zero-code env var setup. See Zero-Code Setup.

Configuration

What's the difference between `AGENTRUST_KEY` and `AGENTRUST_API_KEY`?

Python SDK uses AGENTRUST_KEY. TypeScript SDK uses AGENTRUST_API_KEY. Same purpose, different env var names.

How do I instantly disable governance?

export AGENTRUST_ENABLED=false

No code change or image rebuild required.

What failure mode should I use?

Staging: closed — surfaces gateway issues
Production: open — agent availability during gateway maintenance
Air-gap: queue — buffer and replay later

Tiers & Billing

What can I do on the free/OSS tier?

OSS: schema validation, no API key. Free: evidence validation, tool trust, auto-decision, local audit. See Tiers and Capabilities.

Which adapters require paid tiers?

LangGraph, CrewAI, AutoGen, Claude Agents, OpenAI Agents require Team ($149/mo). MCP requires Developer ($29/mo).

How is tier enforced?

SDK reads tier from API key JWT. Gateway enforces authoritatively when AUTH_ENABLED=true.

Security

Is my data encrypted?

Audit payloads can be encrypted at rest with AES-256-GCM. TLS in transit is operator responsibility. See Security.

Where is data stored?

Embedded: local SQLite. Full edge: PostgreSQL (hot store, 90-day payloads) + optional S3/GCS archive (7 years).

AgentTrust provides GDPR tooling (PII detection, erasure endpoint, retention policies). Operators must complete their own DPIA and configure policies appropriately. See Data Governance.