What Salus does
Salus provides guardrails that validate an AI agent's actions before they execute. It runs as a policy-aware proxy in front of the tools an agent already calls, intercepting tool calls at runtime and deciding whether to allow, clarify, repair, rewrite, or escalate each one.
Key capabilities
- Runtime interception of agent tool calls with a multi-intervention model rather than simple blocking
- Minimal integration via a URL endpoint change, with no agent rewiring or redeployment
- Broad compatibility with OpenAI, Anthropic, Gemini, LangChain, LangGraph, voice platforms like Retell and Vapi, and REST/SDK tools
- Shadow mode to replay historical calls and preview how Salus would have handled them
- A complete audit trail of intercepted actions
Who it's for
Salus targets teams deploying AI agents in high-consequence domains such as healthcare scheduling, financial transactions, and record management, where actions affect customers, money, or systems of record. It addresses the need for trustworthy governance at the moment an agent commits to a real-world action.