Phinq

Phinq is an open-source runtime governance layer for AI agents. It sits between an agent and the tools it calls, intercepting every tool call before execution. Each action is classified by risk: safe, reversible actions pass through automatically; risky or irreversible ones — deleting records, reading credentials, sending payments — are held for human approval. If no one responds in time, the action is blocked. It fails closed.

The enforcement is structural, not advisory. A prompt can be ignored; an API-boundary intercept can't. You can run Phinq as a proxy, an in-process SDK, or a plain HTTP gate, so it works with agents that speak OpenAI, Anthropic, or Gemini APIs, including Claude Code, Codex, Gemini CLI, LangChain, CrewAI, Mastra, and MCP. Every decision — passed, held, approved, denied — goes into a hash-chained, append-only audit log. Each entry includes the hash of the one before it, so tampering is detectable. The project is MIT licensed and available on GitHub.

Who is this for? Teams running AI agents that touch sensitive or destructive actions — internal automation, customer-facing products, anything where an agent could act while no one is watching. Phinq gives them a real decision point before damage and a verifiable record afterward. Hosted policy controls and team workflows are future features; the open-source layer is available today.