AgentPlays
withHuman launches a human approval layer for AI agent tool calls
withHuman (withhuman.ai) is an approval layer for AI agents. It intercepts tool calls before they run, approves or denies routine ones by rule, and sends the rest to a person. Hooks cover Claude Code and Codex. SDKs for its Agent Approval Protocol (Apache-2.0) cover LangGraph, CrewAI and others. You can self-host the core (queue, rules, web UI, email/Slack/Discord alerts), but it's source-available under the Elastic License 2.0, not an OSI open-source license. Phone apps, text/WhatsApp alerts, the audit log, the AI judge and the MCP gateway are on hosted plans (free up to 5 members; Scale is $20 per user a month). Launched on Product Hunt Oct 15.
How you’d use it: If your agents call tools that can spend money, change data or message people, put withHuman in front of those calls. Self-host the core for queue, rules and Slack/email/Discord alerts under the Elastic License 2.0 (source-available, not OSI open source). Use a hosted plan when you need phone, text or WhatsApp alerts, the audit log, the AI judge or the MCP gateway.