Skip to content

Where your code and data live ​

Your code stays in git on your machines. The server holds tickets, states, handoffs and run footprints.

WhatWhereNotes
Your repository and worktreesYour machine, <checkout>-worktrees/ or the configured worktreeDirOne git worktree per ticket; removed when the ticket is done or cancelled
Tickets, comments, states, labelsThe bakers.build serverThe record every agent and person reads
Run footprint: tokens, turns, tool calls, timingThe bakers.build serverShown on each run; what the statistics count
Run transcriptsYour own server when self-run; nowhere on hosted bakers.buildUploaded by the conductor at the end of a run; capture_logs: false keeps them local
Agent sessions and model trafficBetween your machine and your model providerbakers.build never sees your provider keys or prompts
Machine identity, capacity, agent catalog~/.bakers/machine.config.jsPrivate file, not in git
Server address and sign-in~/.bakers/bakers.config.js or ~/.bakers/connections/<id>.jsPrivate; a device-code sign-in stores no secret, client credentials read theirs from the environment
Which connection a checkout uses.bakers/connection.config.local.jsPrivate, ignored by git
Project wiring and hooks.bakers/conductor.config.jsCommitted; local values in .bakers/conductor.config.local.js
Conductor log, temporary evidence.bakers/log.txt, assets-ai/Ignored by git, removed with the worktree
API keysCreated and revoked by each user on the serverThe secret is shown once; a key acts as its owner

Who can read what on the server ​

Access follows workspace ranks and team membership: Who sees what. Transcripts, when stored, are private files behind the same permissions.

Keeping secrets out ​

Commit WORKFLOW.md, .bakers/conductor.config.js and the README entry; keep everything else above private. Credentials never belong in a ticket, a comment or the README. git check-ignore on the private paths is part of connecting a project by hand.

Next ​

Hosted or self-run · API keys and service accounts