FAQ

FAQ

What is a cube, a drone, and a role?

A cube is a project workspace served by one local Borg server; a drone is one agent session working inside it; a role is a named behavior you define, such as a builder or a reviewer. See core concepts for the full model.

Which coding agents can Borg coordinate?

Borg works with Claude Code, Codex, and OpenCode. Launch an agent through borg to connect that session to a cube as a drone; each drone can use a different role.

How do agents hand work to each other?

Through the cube's activity log. Drones post compact signals such as STARTING, REVIEW-READY, and DONE. The cube's routing rules wake the roles that need each signal.

Can agents work in separate git worktrees?

Yes, and Borg does it for you. Worktrees let several agents edit and test on isolated branches while still coordinating through the same cube. Give the repository a Git remote first — agents hand work to each other by pushing branches to the remote and fetching them. Step 4 of Get started covers adding each drone: every drone after the first gets its own worktree and branch automatically.

How do I start Borg locally? Can I connect from another machine?

Follow Get started for the supported local setup. To connect from another machine to an existing cube, use Borg server 0.10.1 or later. For the operator-side steps, follow the trust and provisioning guide.

How do I inspect a cube from an agent session?

Use the Borg tools available to the running drone. borg_regen refreshes context, borg_roster shows participating drones, and borg_read-log reads coordination activity. See the tool reference for the current manifest.

Where should I report a security concern?

Follow the reporting guidance in the relevant source repository and read the security page before exposing a server beyond loopback.