Salus is a real-time agent safety API that wraps around AI agents and inspects their intended actions before they are executed, blocking incorrect or unsafe actions and providing immediate feedback that guides the agent to retry with a corrected approach. As AI agents take consequential actions in production systems — writing to databases, sending communications, modifying files — the cost of errors grows significantly; Salus addresses this by introducing a safety inspection layer between the agent’s decision and its action, catching problems before they cause downstream harm. The platform is designed to integrate with existing agent implementations with minimal friction, functioning as an add-on safety layer rather than requiring agents to be rebuilt. It targets SMB and mid-market development teams building agents with real-world action capabilities and differentiates through its focus on pre-execution inspection rather than post-hoc monitoring.
Salus
Real-time agent action inspection that blocks mistakes before they happen.
Compliance
SOC 2
Key Features
- Pre-execution action inspection: Evaluates each action an agent intends to take before it is executed, checking it against defined safety rules and correctness criteria rather than waiting for failures to occur.
- Automatic action blocking: Blocks actions that fail inspection — incorrectly scoped writes, out-of-bounds API calls, unsafe operations — preventing them from reaching production systems.
- Immediate retry feedback: When an action is blocked, provides structured feedback to the agent explaining why, enabling the agent to reformulate and retry rather than failing silently or requiring human intervention.
- Lightweight API integration: Designed to wrap around existing agent implementations via API, allowing teams to add the safety layer to agents they have already built without significant refactoring.
- Configurable safety rules: Allows teams to define the inspection criteria and safety rules appropriate for their specific agent’s action space and risk tolerance.
Use Cases
- For teams deploying agents with write access to production systems: A team running an agent with the ability to modify databases or send customer communications deploys Salus as a safety layer, ensuring that incorrectly formed actions are caught and corrected before reaching production rather than discovering errors after the fact.
- For developers reducing the blast radius of agent mistakes: A developer building an autonomous agent for a business workflow uses Salus to constrain the agent’s action space at runtime, reducing the risk that an unexpected edge case causes a costly mistake in a live environment.
Pricing MODELS
Usage-based
Pricing Summary
Salus uses a usage-based API pricing model with custom enterprise pricing for larger deployments. Specific pricing tiers are not publicly listed — teams review current documentation or contact the company for details.
Company Size Fit
Mid-market SMB
Technical Snapshot
API Available
Yes
LLM Provider
N/A
Open Source
No
Deployment Options
Cloud Saas
Notable Customers
Undisclosed