phinn

every agent action, accounted for

What is it?

Runtime gateway for AI agents: phinn sits between your agents and tools, requiring human approvals for risky actions and generating cryptographically verifiable receipts for every call.

Who is it for?

Built for Security, Platform, and Compliance teams shipping AI into production.

Value

  • Default-deny for AI agents– phinn sits in front of HTTP tools and (later) MCP, enforcing least-privilege capabilities per call.
  • Human approvals where work happens– Slack-based approvals for risky actions like "DELETE customer" or "wire funds".
  • Tamper-evident receipts– hash-chained, signed receipts for every action, verifiable offline with phinnctl verify.
  • SIEM-ready telemetry– exports to Splunk/Datadog with OTel semantics so SecOps can actually search what the agent did.

How it works

Agent
↓
phinn Gateway
↓
Human Approval
↓
HTTP Tool

Default-deny: All agent actions are blocked by default. Only explicitly approved, capability-scoped tokens allow execution.

Human-in-the-loop: Define which actions require approval via policy configuration. Every decision is cryptographically signed and recorded.

Tamper-evident audit trail: Hash-chained receipts for every action, verifiable offline. SIEM-ready telemetry streams to Splunk/Datadog.

Contact Us