agenttic

Agenttic

Give your AI agent a signed safety grade — an independent agent-safety certification you can verify.

Agenttic traces any AI agent, tests its safety-relevant behavior, and issues a signed, offline-verifiable dossier with an honest Tier grade (A/B/C). It measures what was actually tested — untested behavior is reported as NOT ASSESSED, not quietly passed.

Install

pip install agenttic

Optional framework adapters (lazy — the base install pulls no framework SDK):

pip install "agenttic[langgraph]"   # LangGraph / LangChain
pip install "agenttic[openai]"      # OpenAI Agents SDK
pip install "agenttic[all]"         # all adapters + OpenTelemetry

Requires Python 3.12+.

Quickstart

Get a signed grade in under a minute — no API key — by certifying the built-in reference agent:

agenttic certify --mock --out dossier.json   # → a signed Tier A/B/C dossier
agenttic dossier verify dossier.json         # → offline signature + evidence check

Or wrap your own agent with one line and trace its behavior:

from agenttic import trace

agent = trace(my_agent)   # auto-detects LangGraph / OpenAI Agents / plain callables

What you get

• A signed, offline-verifiable dossier — Ed25519-signed, checkable without calling home, backed by hash-chained evidence of exactly what was tested.
• Honest coverage — results state what was measured; anything untested is marked NOT ASSESSED rather than assumed safe.
• A single Tier grade (A/B/C) you can publish and third parties can verify.

Agenttic reports the measured behavior of an agent under the tests that were run. It is a measurement of tested behavior, not a guarantee of safety in every situation.

Links

• Homepage & docs: https://agenttic.io
• Source: https://github.com/omarharoun/Agenttic