Viktor.comAn AI coworker that actually does the work
Promoted
Impri
Human-in-the-loop approval inbox for AI agents
10 followers
Human-in-the-loop approval inbox for AI agents
10 followers
Your AI agent proposes an action (send an email, post a comment, run a command). You approve, reject, or edit it from a web, mobile, or Slack/Discord/Telegram inbox, and it only runs on yes. The gate is a data dependency, not a prompt the model can skip. Open source (MIT), self-hostable, MCP + REST.
interesting idea, but claude code when set to /remote-control or done via web already pings me with questions. I maybe could be convinced, what are the benefits of using this vs. just Claude?
@mjohnson42 yeah for an interactive claude code session youre right, the permission prompts and remote control already keep you in the loop, and for that case you probably dont need this.
where it earns its place is the unattended stuff. scheduled scripts, cron jobs, a few agents running while youre asleep, or agents that arent claude at all. theres no live session there to stop and ask you, so it pushes the proposed action to an inbox and blocks until you approve from your phone whenever you get to it. its plain rest/mcp so it works with langchain, crewai or a bash script too, and every decision goes to an audit log. so less an alternative to claude code, more the piece for agents running without you watching.