Guardrails for AI agents — budget limits, kill switches, and human approval gates.

SidClaw governance integration for LangChain.js — policy evaluation, human approval, and audit trails

Agent Firewall for AI and MCP tool calls: deterministic allow, ask, deny permissions with human approval and audit hooks.

ntfy tells you things. gate asks you things. Human approval as a curl.

Spending policy for autonomous agents: per-call caps, rolling budgets, allowlists, and human-approval escalation. The single choke point in front of every signer.

Pure, tested risk classifiers that decide which AI-agent actions need a human approval on an AgentBee hardware key.

User-approval seam (ctx.approval) for the DeepSeek Harness: one-shot permission decisions dispatched to composed answerers over the approval/request waterfall, fail-closed by default

Agent-to-agent relay with human approval gates. E2E-encrypted rooms over a Cloudflare Durable Object relay; every inbound request and outbound response is approved by a human.

Policy-as-code authorization for AI SDK tool calls, powered by [Open Policy Agent](https://www.openpolicyagent.org/).

TypeScript SDK for governing AI agent tool calls with ActionProxy policy, human approval, one-time grants, and audit

Approval composer takeover over the scoped Remote Event waterfall

Returns from a pool of 15m human-readable IDs

Append-only project-local learning ledger for Pi with two separate trust gates: manual human approval, and a bounded auto-safe machine gate that is on by default in Pi-trusted projects. Selective fork of Matt Devy's pi-continuous-learning.

MCP server that lets AI agents request human approval and block until a person decides in Enkreesee.

Habena — keep your AI agent on a short rein. MCP middleware proxy: policy guardrails, spend caps, and human approval for AI agents.

Protocol-neutral catalog validation, authorization, and execution core for OpenGeni tools.

AI workflow orchestration with human approval gates. Self-hosted, open source.

Backend SDK that cryptographically enforces human approval on protected operations. Verify a HumanAuth receipt JWS, check it against the action and plan your service is about to execute, and atomically reserve the receipt's replay slot — all in one call.

YAML mandate (policy) loader and evaluator for the Cord protocol — defines what an agent may do without human approval.

Model Context Protocol (stdio) server for HumanStep — let agents create human-approval decisions and close the loop (wait / continuations).