Skip to content

Organization settings

The defaults every workspace inherits: what its container is built with, the board's columns, the git flow, the agent and accounts boxes run on, and the standing instructions every agent reads.

Reading time
4 min read

How it works

Eight tabs, grouped by what each decides.

Workspace defaults, the setup script every new workspace's container runs when it launches, on top of the built-in language toolchains: the extra packages and tools your work needs.

Board stages, the columns the workspace board is organized into: rename, reorder, add, delete. The column you mark as default is also the one that decides which workspaces may keep a box running, so there is no second setting to keep in step. A stage that still holds workspaces cannot be deleted, and the refusal names the column and its count.

Git flow, an ordered list of stages, each with a key, a label and a branch. The first stage is what work is cut from; promotion runs left to right; the last stage is production. Every branch in the flow is protected from direct agent pushes. A stage stores its key rather than its branch, so renaming a branch leaves merge requests already opened against it intact.

AI, which agent new workspaces start with, the Claude Code, Codex and OpenCode accounts every box runs on (add several per agent; when one hits its usage limit or its login fails, DevBench moves the box to the next account of the same agent and re-sends what failed), and the OpenAI key used to draft requirements, which never reaches a box. Its Coding agents list is which version of Claude Code, Codex and OpenCode the boxes run (CLI versions and one-click updates). This is also where the organization's standing instructions for every workspace live: the stack, the review etiquette, the library that must never be used again. They are seeded into every workspace's CLAUDE.md under a managed heading (a Codex box reads the same file as its AGENTS.md), and rewritten wholesale on each push, so editing the text converges every box instead of stacking copies in it, and clearing it removes the rule. Saving and applying are separate: Apply to all workspaces is what pushes it out, because a save that immediately rewrote every box would make a half-finished thought everyone's problem.

Integrations, your personal GitHub, GitLab and Bitbucket tokens, used to import repos and let workspaces push; the GitHub App; and the inbound webhook, where events from outside DevBench arrive.

Boxes, Box health (whether DevBench checks that a box's agent is not stuck), and the devbench CLI inside your workspace boxes, and reseeding it. The installers for your own machine are on Install DevBench, under ?.

Backup, see the next article. Plan and billing, what the organization uses against its plan.

Your own settings (what each click does, notifications on this browser, Lathe sharing) are not here: they are Preferences, in your menu.

Each workspace has its own settings too, General (running, archived), Automation (what DevBench does to it unasked), Git & merging (guardrails, self-merge), Security (what the box may reach beyond its repos), Agent rules (what the box enforces on the agent whatever it decides), Production alerts (how its deployed app reaches the team), and a Danger zone kept apart from the routine toggles on purpose.

Steps

Following along? Tick each number off as you go.
  1. Open Settings from the organization's menu at the top left (or the search: Ctrl Ctrl, ⌘ ⌘ on a Mac). It opens as a tab where you are.

  2. On Board stages, make the columns match how your team actually works.

    Mark the column where work happens as the default, that is also what keeps boxes running.

  3. On Git flow, set the stages work is cut from and promoted through.

    First stage is the source; the last is production.

  4. On AI, choose the default agent for new workspaces and add accounts for each agent your workspaces use.

    Claude Code signs in with a claude.ai account; Codex with ChatGPT or an OpenAI API key. Boxes use the default account of their agent; a spare one is what the failover switches to.

  5. Add an OpenAI key on the same tab if you want AI-drafted requirements.

    That key drafts requirements only. A Codex box never uses it; add a Codex account for that.

  6. Write your organization's standing instructions in the same tab, and press Save changes in the bar that appears at the bottom.

    New workspaces pick them up at creation. Ctrl/⌘S saves too, and leaving with unsaved changes asks first.

  7. Press Apply to all workspaces when the text is finished.

    Best-effort per workspace; it reports what it could not reach. A stopped box picks the text up when it next starts.

  8. Set the per-workspace settings on the workspace's own Settings tab.

    Guardrails, self-merge and the idle rules live there, not here.

Worth knowing

  • Saving instructions does not push them. Apply to all workspaces is the separate, deliberate act.

  • A ## heading you write inside the standing instructions is demoted a level on the way in, otherwise it would end the managed section early and leave the rest loose in the file.

Report a problem