Salus

Salus is a runtime guardrail platform for AI agents that intercepts and validates every tool call before it executes. Backed by Y Combinator, Salus helps teams prevent costly agent mistakes by enforcing policies, requiring evidence grounding, and enabling agents to self-correct through structured feedback. It integrates with popular AI frameworks using a single decorator per tool call, making it easy to add safety layers to existing agent workflows.

• Runtime Guardrails: Intercept and validate every tool call at execution time — block actions that violate policy or aren't grounded in evidence before they execute.
• Self-Repair: When an action is blocked, Salus returns structured feedback explaining exactly what failed, enabling the agent to self-correct and retry automatically.
• Full Visibility: Stream every agent interaction in real-time with full traces, token usage, and latency breakdowns for complete observability.
• Evals: Generate thousands of adversarial and realistic scenarios to measure tool-call correctness before deployment.
• Simple Integration: Add protection with a single @session.protect decorator per tool call — no major refactoring required.
• Framework Compatibility: Works out of the box with OpenAI, Anthropic, LangChain, LangGraph, and CrewAI.
• Evidence-Based Execution: Enforce dependency chains between tool calls, requiring evidence-producing steps to complete before commit actions run.
• pip Install: Get started quickly with pip install salus-ai and wrap your tools with the Salus session decorator.