Open-source Grok Bot alternative for named AI teammates. Local-first desktop, web, and mobile clients with per-bot tools, instructions, and memory. Independent fork of T3 Code.
Permission modes
A permission mode controls which actions a bot can take before it must ask you.
The mode is set per chat from the mode control in the message composer. Changing it in one chat does not change any other chat. A chat created from inside another chat keeps that chat's mode. Otherwise, new chats use Settings > Sandbox > Local execution and use Auto review by default.
The modes
Supervised asks before commands and file changes. Work outside the workspace stays restricted.
Auto-accept edits allows file edits and asks before commands and other actions.
Auto allows routine actions and asks before risky ones. Each provider maps this mode to its own permission system. A provider without an equivalent can fall back to supervised behavior.
Full access allows routine commands and edits without local prompts. The bot still asks its own questions and still stops at protected actions.
Protected actions
Questions never need approval. A question with choices shows those choices inline. A free-text question uses the message composer for the answer.
Akeru's runtime always asks before an action that sends data, pays, deletes, changes production, publishes, exposes secrets, signs, refunds, or changes an account. Unknown mutating actions also ask. Each approval applies only to the pending action.
Installed MCP tools ask unless the server marks the tool read-only. Provider adapters can add their own approval rules.
Choose a mode
Use Supervised for unfamiliar bot work or a repository where an unwanted command is expensive.
Use Auto-accept edits when you want the refactor but still want to inspect shell commands.
Use Full access in a sandbox or a checkout that you can restore. Check the workspace with Git after the bot finishes.
Web, desktop, and mobile use the same four labels.