Salus
Salus provides runtime authorization for AI agents, sitting between agents and the tools that perform side effects to hold unsupported actions before they reach production systems. The platform evaluates actions against policy, evidence, and intent, returning exact missing evidence and next steps so agents can repair and retry. It integrates with existing stacks including OpenAI, Anthropic, LangChain, and MCP, and records every decision as a receipt showing policy, evidence, timing, verdict, and execution result.
- Artificial Intelligence
- Developer Tools
- AI Agents
- Cybersecurity
- Software Only
Funding
Raised to date
$500KRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
Founders
Founder details are not available yet.
Product
Problem
AI agents can hold valid credentials, permissions, and produce perfectly formed tool calls yet still take actions based on stale or incorrect information. Authorized access does not guarantee correct behavior, and wrong actions—such as booking a reservation at an outdated price or executing a consequential write—can reach production systems before anyone notices, creating real-world harm and eroding trust in autonomous systems.
Solution
Salus provides runtime authorization for AI agents by sitting between the agent and the tools that perform side effects. When a protected action is attempted, Salus evaluates it against policy, evidence, and intent, holding the action before it reaches the provider. If the action is unsupported, Salus returns the exact missing evidence and the next step, allowing the agent to repair its approach and retry. Every decision is recorded as a receipt showing the policy, evidence, timing, verdict, and execution result. The service works with existing agent loops, models, frameworks, and backends without requiring changes to the underlying stack.
Target Audience
Primary customers are engineering teams and platform groups deploying AI agents in production environments where consequential actions—such as bookings, refunds, or writes to external systems—need safeguard controls before execution.
Features
- Side-effect protection via a Python SDK that wraps existing tool implementations with `salus.protect()` and a `side_effect=True` flag
- Runtime holds that prevent provider calls for unsupported writes, with execution status confirming `provider_called: false`
- Structured recovery instructions that return missing evidence and next steps to the agent, such as `missing: live_available_itinerary`
- Decision receipts documenting policy, evidence, timing, verdict, and execution result for every protected action
- Shadow and enforcement modes, allowing teams to test policies before activating them
- Bulkhead-style deployment across supported frameworks including OpenAI, Anthropic, LangChain, LangGraph, CrewAI, AutoGen, Vapi, Retell, and MCP