withHuman
withHumanより
A bouncer for your agents: it holds risky tool calls until the right person says yes, and waves routine ones through.
- 向いている用途
- Teams running several agents (coding, support, internal) that want one set of approval rules.
- 料金の目安
- Open source, plus a hosted version.
withHuman intercepts agents' tool calls before they run. Rules approve or deny routine calls automatically and route risky ones to the right person. It uses an open spec, the Agent Approval Protocol, with hooks for Claude Code and Codex and SDKs for Python and TypeScript.
合うところ
- One rulebook across many agents and frameworks.
- Open protocol rather than a closed integration.
合わないところ
- Starts with MCP tool calls; CLI and raw HTTP calls are harder to gate.
- If the service is unreachable, gated calls fail by design.
タグ: エージェント · オープンソース · 自動化
メモ更新 2026年10月16日
このメモはまだ日本語ではないため、英語で表示しています。
エージェントの他のツール
- ChatGPT — The default general assistant. Broad enough for everyday writing and light tool use, and less predictable than Claude when a long edit has to stay careful.
- Claude — The best default for long documents and careful agent-style work. Easier to steer than most chats, and less of a do-everything platform than ChatGPT.
- Devin Oncall — An on-call agent that starts the moment an alert fires, digs through telemetry and code, and hands you a fix as a pull request.
- Gemini — Worth opening when the work already lives in Google, or when you want image generation in the same chat. Not the first pick for code or a long edit.