Salus GTM brain

Salus sells runtime authorization for AI agents to teams deploying agents that take consequential actions.

category

AI agent runtime authorization

market profile

  • text: Agents can have valid permissions and still execute incorrect actions based on stale or missing evidence.
  • text: Organizations deploying agents need enforcement and decision evidence at the point of tool execution.

personas

  • title: AI engineering lead
    pains:
    • An agent may issue a refund or change a booking without checking current facts.
    • Evals and logs identify failures but do not necessarily prevent an incorrect write.
  • title: Security or AI governance leader
    pains:
    • Permissions alone do not show whether a particular action is justified.
    • Auditors and incident responders need evidence of policy decisions and execution outcomes.
  • title: Agent platform product owner
    pains:
    • Unsafe autonomous actions can block rollout to production.
    • Manual review of every action slows completion and raises operating costs.

icp

  • text: Software companies and enterprise teams deploying AI agents with tools that move money, change records or call external systems.
  • text: Teams using agent frameworks or custom Python tool executors that need a runtime control layer.

message pillars

  • text: Stop unsupported writes before execution.
  • text: Give agents actionable feedback to repair a proposed action.
  • text: Preserve evidence of decisions and outcomes.

objections

  • objection: We already have agent evaluations and observability.
    response: Salus targets the pre-execution boundary; compare prevented actions and latency against the existing stack in a pilot.
  • objection: Will enforcement disrupt live workflows?
    response: Pricing documentation says each route can start in shadow mode, allowing decisions to be reviewed before enforcement.
  • objection: Will we need to rebuild our agent?
    response: The site describes wrapping the tool executor and keeping the model, framework and backend unchanged; test the actual integration.

competitors

  • name: Straiker
    why: Offers runtime guardrails covering tool manipulation and other AI security risks.
  • name: Guardrails AI
    why: Provides validation guardrails for AI applications, with overlap in reliability budgets.
  • name: F5 AI Guardrails
    why: Provides controls for AI inputs, outputs and agent tool use.

keywords

  • AI agent runtime guardrails
  • agent action authorization
  • AI agent tool call validation
  • agent policy enforcement
  • prevent AI agent errors
  • AI agent security

voice

  • Technical and direct
  • Risk-aware and action-oriented

signals

  • text: An organization is moving an agent with refund, booking or record-write permissions into production.
  • text: A team is adding audit requirements or shadow testing to agent tool execution.

stack

  • Python
  • OpenAI
  • Anthropic
  • LangChain
  • LangGraph
  • CrewAI
  • AutoGen
  • Vapi
  • Retell
  • MCP

hiring

  • Founding SWE / MLE

pricing

model: flat
starting price: $500/month

motion

hybrid

channels

  • YC launch
  • LinkedIn
  • developer documentation
  • direct sales

confidence

high

Founders

  • name: Vedant Singh
    title: Founder/CTO
    role: CTO
    bio: Building to validate your AI agent's actions before they execute AI researcher and formerly @Stanford CS

Back to YC companies