Phinq is an open source runtime governance layer for AI agents that classifies tool calls by risk, pauses irreversible actions for human approval, and records decisions in a tamper-evident audit log.
Phinq is an open source runtime governance layer for AI agents. It sits between an agent’s proposed tool call and the execution boundary, classifies the action by risk, and decides whether it should pass, pause for approval, or be blocked. Safe actions continue automatically, while irreversible or higher-risk actions are held until a human decision is made.
The product is designed for agent workflows where a prompt alone is not enough to control behavior. Phinq enforces governance structurally at the API or execution layer, records each decision in an append-only audit log, and can run in shadow mode during setup so teams can observe decisions before turning on enforcement.
Phinq intercepts each tool call before execution and evaluates it from the tool name, arguments, rules, and relevant session context. That lets governance happen at the execution boundary instead of relying on prompts or agent instructions.
The classifier assigns a risk-aware outcome such as ALLOW or HOLD. The documentation says it uses the action’s intent and session velocity as part of the decision path.
Safe actions pass through automatically, while risky actions pause for review. In enforce mode, a HOLD becomes a pending approval rather than being forwarded to the tool.
Approvals can be resolved through the configured Phinq channel, with CLI, Telegram, and Slack named in the docs. Approved actions are released, while denied or timed-out actions are blocked.
Every decision is written to an append-only, hash-chained audit log. The log records what the agent attempted, how it was classified, and whether it passed, waited, or was denied.
Phinq can run as a proxy, an in-process SDK, or a plain HTTP gate, and the docs note compatibility with OpenAI, Anthropic, Gemini, Claude Code, Codex, Gemini CLI, LangChain, CrewAI, Mastra, and MCP.
Use Phinq when an agent can take actions that should not be left to prompt discipline alone, such as deletes, credential reads, payments, or external messages. The system holds risky actions until a human approves them.
Teams can start in shadow mode to observe how the classifier behaves before enforcing holds. That makes it easier to inspect decision patterns and add governance without immediately blocking work.
Phinq is suited to teams that want a decision record for later review. Each action is logged with its classification and outcome, creating evidence for debugging or post-incident review.
Because Phinq can run as a proxy, SDK, or HTTP gate, it can fit different application architectures and client libraries without requiring a single framework.
The docs mention resolution through CLI, Telegram, or Slack integration, which makes it useful for teams that want humans to approve risky actions in a familiar channel.
Phinq is an open source runtime governance layer for AI agents. It sits between an agent and the tool execution boundary so each tool call can be classified before execution.
Phinq intercepts the tool call, classifies it by risk, lets safe actions pass, and holds or denies higher-risk actions. If an action is held, it can be resolved through the configured approval channel; an unanswered hold fails closed.
The docs say Phinq can resolve holds through CLI, Telegram, or Slack integration. The documentation also says the proxy observes OpenAI-style, Anthropic Messages, and Gemini traffic, and the TypeScript SDK gates calls in process.
Yes. The audit log is append-only JSONL, canonicalized with JCS and chained with SHA-256. Each entry records intent, classification, decisions, hold transitions, and usage metadata.
Phinq starts in shadow mode by default, where it classifies and logs without enforcing HOLD decisions until PHINQ_ENFORCE=1 is set.
CreateOS Sandbox is an isolated compute environment for running code and agent workloads in Firecracker micro-VMs with private networking and SDK, CLI, or MCP control.
AakarDev AI helps teams manage AI provider access, project setup, logs, and analytics in one dashboard. BYOK support included.
Trigger.dev chat agent is a durable AI chat backend for stateful conversations that survive refreshes, crashes, and long-running turns with AI SDK useChat.
ByteAsk is a terminal-first AI coding agent for C and C++ that edits repos and verifies changes with compilers, debuggers, sanitizers, and tests.
Codex Plugins bundle reusable skills, app integrations, and MCP servers into workflows you can install in the Codex app or use from Codex CLI.
hob is an independent workspace for coding agents, with local control over sessions, terminals, history, routing, and follow-up work.
