Habenula

Your agents on your terms.

Habenula is the private, open and responsible way to control your personal AI agents. It's an app for everyday people, families and small teams that makes advanced AI control easy, trustworthy and secure, so you can hand off bigger tasks and still have the final say.

Try the 10-second loop How it works What is an agent?
Ask your AI about us

The stakes

The industry shipped the power and skipped the controls.

Agents now send, spend and delete on your behalf, governed by a dialog box. The vulnerabilities were known and shipped anyway. Underneath it all, the brakes were dismantled to win the race.

Read the manifesto (14 min) →

You hold the reins

More capable machines should mean more control for you, not less.

Our values → Our vision →

Our unique approach

Personal controls you can verify.

Rules you can read decide every action, not the model, and every decision lands on a record you can check yourself. You allow a call once or for a time you choose. Nothing is allowed forever.

The state of agent governance →

Try the 10-second loop

Answer the request the way you would in the app. The strip under the card shows what happens next: checked, logged, then run.

Try for real →
on the record

A guided walkthrough. Pick a choice to advance.

What it's built around

The model does the thinking. Habenula governs the doing.

Six promises Habenula keeps for you. They're built in, not settings you have to remember.

  • 01

    Every action gets checked first

    Before your agent does anything, Habenula checks it against your rules. Anything it doesn't recognize waits for you.

  • 02

    The AI never holds your keys

    The keys to your accounts stay locked away. They're used only at the moment an action runs, and never handed to the AI.

  • 03

    A record no one can quietly change

    Every action is written down before it happens. You can check the record yourself, on your own computer.

  • 04

    Precise permissions

    You allow an action once, or for a time you pick. Nothing is allowed forever.

  • 05

    Spending limits it can't raise

    You decide how much it can spend. Anything over your limit waits for your OK.

  • 06

    An off switch that really works

    One command takes back every permission at once. Your accounts stay connected, so starting again is easy.

See how our product comes together →

Open source

Control you can't verify isn't control.

Everything that holds your credentials, decides what your agent may do, and records what it did is open — AGPL v3, with the audit-log verifier under MIT so anyone can check our records independently. Read it, run it, hold us to it.

Read the source License

Ready to govern your agents?

Habenula is early and built in public. Tell us what you would want it to govern, or run it on your own machine today.

Contact us Run it