Memnox

Your agents can already do the work.
Memnox makes it safe to let them do it without you.

The control plane for autonomous work: one policy and execution layer above
Claude Code, Codex, Cursor and the rest, with a face on every screen you carry.

              

Website · Docs · Runtime · Desktop · Mobile

_{One run, one question. The agent works on its own, the screen and the phone ask once, and a signed yes releases it.}

Most people never let an agent run unattended, and the reason is not that it cannot do the work. It is that nobody can say in advance what it will touch. Memnox answers that before every tool call, so the question stops being "what did my agent do?" and becomes "what can I safely let it do next?"

1. Every tool call is ruled on. The runtime sits inside the session your agent already works in and answers each call with one of three verdicts: allow, ask or deny.
2. What needs a person finds you. It reaches you in the agent's own prompt, in the island at the top of your Mac, Windows or Linux screen, or on your iPhone or Android phone.
3. Every yes is signed. Allow goes through Touch ID, Windows Hello, Face ID or a fingerprint, and is signed by a key your device guards. Deny needs nothing.
4. Everything is on the record. What each agent did is kept, so a session that went wrong can be explained and rewound.

No model decides anything, and your code and your secrets never leave your machine.

macOS. The island grows out of the notch to hear you, answer out loud, show every working session and ask about a held call, with Touch ID on every allow.	iPhone. The same call waits in the Dynamic Island and on the lock screen, and goes through Face ID.
Windows and Linux. The same island, the same three answers and the same voice, built screen for screen from the Mac and signed after Windows Hello or a polkit check.	Android. Every session on your computer, what needs you and the voice, with the fingerprint on every remote yes.

A held call is one question wherever it is asked. The terminal, the island and the phone all answer the same call, and whichever answers first releases the agent.

The apps are previews. Windows and Linux are built and tested in CI, and neither phone app is in a store yet, so each repository's README says how to install or build it today.

flowchart LR
    subgraph mac["Your computer"]
        agents["Claude Code · Codex · Cursor<br/>Gemini CLI · Windsurf"]
        runtime["memnox<br/>the runtime and its daemon"]
        desk["Memnox desktop<br/>macOS · Windows · Linux"]
        agents -- "every tool call" --> runtime
        runtime -- "allow · ask · deny" --> agents
        desk <-- "local socket" --> runtime
    end
    phone["Memnox mobile<br/>iPhone · Android"]
    cloud["Memnox Cloud<br/>the team around the agents"]
    phone <-- "end to end encrypted Noise tunnel" --> runtime
    runtime -. "optional" .-> cloud

The runtime is the authority. The apps are a face, a voice and a signing key over it, and none of them holds a rule of its own. Memnox Cloud is added on top once there is a second person, and the runtime keeps enforcing when it cannot be reached.

Sign in to Slack, Gmail, GitHub, Linear, Google Docs or Notion once, and the apps can work in them for you. Say "tell #eng the deploy is out and thank Bo", or let an agent propose a reply, and it arrives as a draft: which account it goes out as, who it goes to, and the exact words in a box you can change. Nothing goes out until you press Send, and Send is signed. Decline needs nothing.

npm install -g memnox
memnox setup

setup finds the agents on your machine and goes through them one at a time: what each can already reach, what to call it, and whether to put it under Memnox. Then open your agent the way you always do. There is nothing to launch and nothing to remember.

Memnox on your machine governs the agent. Memnox Cloud governs the team around it: one fleet across every laptop, server and pipeline, approvals your team can answer from Slack or Discord, the decisions your team has settled put in front of every agent, and one record of what every agent did. It is hosted, it needs no change to how anybody works, and it starts free at memnox.com.

Pull requests are welcome in all three open source repositories. Each README says how to build and test the code there, and CI runs those tests on every pull request. If you are unsure where something belongs, open an issue and ask, because that is a good first contribution too.