Settings

Settings holds your preferences for the orchestrator — open it from the sidebar. Most of it is per-user (your tokens and variables, your NUCLEUS.md and NOVA.md, theme); the DevBox Security profiles are tenant-level — shared across your team. The page is organised into the sections below (the right-hand nav jumps between them).

Environment Variables

Everything injected into the DevBoxes you launch, in one honest place. Three groups:

• Service access — the provider tokens your runs need to reach repositories and tickets: GitHub, Azure DevOps, GitLab, and Jira. Each row names the env var it injects (GH_TOKEN, AZURE_DEVOPS_EXT_PAT, GITLAB_TOKEN, JIRA_API_TOKEN) and the scopes the token needs. A configured token shows a check mark; a missing one is exactly what makes a run against that provider skip before spending AI tokens — see run readiness & skip reasons.
• Agent Secrets — the AI coding agents’ own API keys and tokens, an alternative to signing each agent in per worker: the Claude OAuth token (CLAUDE_CODE_OAUTH_TOKEN), Anthropic API key (ANTHROPIC_API_KEY), GitHub Copilot token (COPILOT_GITHUB_TOKEN), and OpenAI API key (OPENAI_API_KEY). The Claude and Anthropic keys additionally carry a Use for DevBox authentication toggle, which injects the key into the DevBoxes you launch; the Copilot and OpenAI tokens are plain injected tokens. Running Nova, Nucleus and the prompt improver on your Claude credential is a separate switch that lives in App AI Assistants, not on the credential row. See agent secrets.
• Global variables — free-form variables injected into every DevBox you launch. Mark a variable secret to encrypt it at rest and redact its value from run traces. Project variables override global ones on a name collision — see sources & precedence.

Values are encrypted at rest and are only ever handed to your running DevBox instances — secret values are never shown back to you once saved (you replace a value by pasting a new one).

Agent sign-in (Claude, Codex, Copilot) is per worker, not per user — set it up from any worker’s page via Manage agent sign-in, or see the agent sign-in guide.

App AI Assistants

The two agents that run inside ClusterCode — the assistant you chat with and the coordinator that executes your work — are configured here. Each has its own personal instruction file, conceptually like a CLAUDE.md: set tone, standing preferences, or context the agent should always keep in mind. Both editors carry the ✦ prompt improver, and both can be reset back to the default at any time.

• NUCLEUS.md — Nucleus, the assistant you talk to in the dashboard. See Nucleus Configuration.
• NOVA.md — Nova, the coordinator that executes your Runs, Schedules, Loops and Prototypes. See Nova Configuration.

Both files are stored against your account, so they follow you to whichever browser you sign in from — and an unattended Schedule firing at 3 a.m., with no browser open at all, still applies NOVA.md. Both cap at 10,000 characters, and both count toward the prompt of every message or execution they apply to — the editors show a live token estimate.

Nova and Nucleus are both built on Anthropic (Claude) models. Every plan includes a monthly AI allowance, but you can bring your own key for usage beyond your plan — or to run entirely on your own account. Store the credential under Environment Variables → Agent Secrets (an Anthropic API key or a Claude OAuth token), then switch it on for both agents from the Bring your own key control in this section, which also lets you pick the Claude model they run on. The OAuth option routes them through your personal Claude Code subscription — make sure your usage complies with Anthropic’s terms. See also AI Credits for how plan allowance and BYOK interact.

Appearance

Pick the app theme: Light, Dark, or System (follow your OS). The choice applies immediately and persists on this device.

DevBoxes

Two groups of settings for your DevBoxes:

Display

• Hide external and system DevBoxes — show only the DevBoxes managed by ClusterCode, hiding anything else running on your workers.

DevBox Security — configure your tenant’s security profiles. Each tenant defines a default security profile applied to every DevBox launched on its workers (it governs what the DevBox can reach — egress, allowed domains, and so on). You’ll see one card per tenant you belong to, showing its current default profile and your role; from here you set that default. You can still override the profile or add extra allowed domains per DevBox at launch time. Changing the tenant default requires the right role (e.g. owner/admin).

Session persistence

Keep your engine logins, session history, and settings so a new DevBox picks up where the last one left off — on any worker. Persist my sessions is on by default; Include my self-hosted workers is off by default (config only, your code never leaves your machine); Delete stored data wipes everything with a 14-day recovery window. See Session persistence for the full picture, including how it’s encrypted end-to-end.

Notifications

Control which email notifications you receive:

• DevBox Alerts — DevBox lifecycle events (created, started, stopped, deleted).
• Worker Alerts — worker connection and disconnection events.

Billing & Plan and Account notifications are always on — they cover events you shouldn’t miss (payment, plan changes, account security).

About

Shows the running application version (click to copy) — handy when reporting an issue so we know exactly which build you’re on.