The best way to get work done, instead of chatting about it and doing it yourself. Open-source, secure agents, in Slack and on the web. Self-hosted agent harness for teams: sandboxed, policy-controlled, and never touching a real credential.
Hey Product Hunt 👋
I'm Jonathan, co-founder of OneCLI together with Guy.
AI agents have a problem inside companies. To be useful, an agent needs access to real systems: GitHub, Gmail, your CRM, your cloud. But giving an agent real passwords and keys is risky. Keys can leak.
OneCLI fixes this. It is an open-source agent harness for teams, hosted on your own servers, or you can use our managed service. Every employee gets their own agent in a sandbox. It connects to GitHub, Gmail, Notion, Dropbox, or your CRM from the web or Slack.
The part I like most: agents never hold real secrets. The agent only sees a placeholder. Our gateway adds the real secret at the network layer, for each request, after the request is approved. You cannot steal what is not there.
For sensitive actions, like sending an email or deleting a ticket, the agent asks a human first. We have spent our careers in security. I built zero trust network access (ZTNA) at Axis Security, acquired by HPE. Guy was the first employee at Argon, acquired by Aqua Security. ZTNA never trusts the client. Agents need the same. So we built it.
We’re YC-backed and fully open source, with 350K+ downloads. Have a free tier with no card required, you can self-host in minutes: https://github.com/onecli/onecli
Would genuinely love feedback from the PH community, especially from anyone who tried OpenClaw // Hermes and had issues with setting it or getting into his own company.
Happy to answer questions all day!
Report
Can we also somehow manage access for agent? (so, „The agent asks human“ becomes a bit easier!
@ankita_singh27 Yes, you can manage access for each agent based on who owns it. If you're referring to the approval mechanism, that's a separate thing. Human-in-the-loop is set for risky actions you want to prevent the agent from performing without approval.
OneCLI
OneCLI
@ankita_singh27 Yes, you can manage access for each agent based on who owns it. If you're referring to the approval mechanism, that's a separate thing. Human-in-the-loop is set for risky actions you want to prevent the agent from performing without approval.