Introduction
Quantify and mitigate AI risk in your agents.
Harbor is an AI risk platform for agents. It finds vulnerabilities before you ship with white-box assessment and adversarial testing, quantifies risk in production by scoring every trace, and caps what your agents' tools will accept with guardrails.
Quickstarts
Send your first trace
Instrument your agent with the Harbor SDK and see a scored trace in minutes.
Route through the Gateway
Point your LLM endpoint at Harbor for complete trajectory capture and inline enforcement.
Add your first guardrail
Cap what your agents' tools will accept, enforced inline at runtime.
Scan your repositories
Find vulnerabilities in the AI systems in your code, on every pull request.
How Harbor works
Harbor runs a simple loop — test before release, score in production, guard at runtime. Read the Harbor overview to see how the pieces fit together.