🚧 Project in development · not open to the public yet.
Security and control

Agents that act need limits you control.

BeeAgents is built so that powerful assistants can still be supervised: encrypted keys, permissions per bee, approvals and isolation.

How it works

Keys stay sealed

Provider keys are stored encrypted and only travel to the provider they belong to.

You set the permissions

Tool groups and authorized folders per bee; imported bees arrive without keys or sensitive permissions.

Risky actions wait

Actions with side effects ask for your yes, in the app or in Telegram. Plan mode changes nothing.

You can undo and recover

A 30-day trash bin, restorable file copies and backups.

What it can do

🔐

Encrypted keys

Sealed on your server; the browser never gets them back.

🧩

Permissions per bee

Groups of tools and folders for each bee.

✋

Approvals

In the app and in Telegram. Unattended tasks deny actions that need permission.

📦

Isolation

The Device runs in Docker; cloud accounts run in separate instances.

🧯

Limits

Daily token limits per bee and global, alerts at 80%, and spend tracking.

🛡️

Account protection

Two-factor sign-in, login throttling, and blocked private-network access for browsing and MCP.

Where your data goes

Your chats, notes and keysYour own server or computer (self-hosting), or your isolated cloud account.
What you send to an AIThe AI provider you chose, under its own terms.
Anything to BeeAgentsOnly account and billing data, if you use the cloud service.

Honest status

  • Tested with simulated models and a private installation.
  • An independent security audit has not been done yet.
  • Third-party skills, MCP servers and imported bees are text or code from others: review them before trusting them.

Keep exploring

Back to the roadmap