Account, memory, tokens and usage

Settings separates your identity, stored memory, API tokens and device notifications. Usage provides the accounting view. Administrator grants and limits continue to apply to the features configured here.

Inspect your account

Open /settings to see your signed-in email, internal user ID, identity-provider roles and resolved AIplane roles. These roles can differ because the installation maps identities to its authorization model. Ask an administrator about missing rights; this page does not edit your identity-provider roles.

Manage memory

Open Settings → Memory (/settings/memory). The three cards have different purposes:

Kind	Use	How the assistant receives it
Preferences	Response style, units, language and other standing preferences	Included in the standing context when Memory is granted and enabled, subject to its context budget.
Project context	Background about ongoing work	Retrieved on demand through memory tools.
Facts	Stable details to recall when relevant	Retrieved on demand through memory tools.

Select Add on the appropriate card, enter the content and save. The add dialog lets you correct the kind before saving. Edit existing text inline and select Save; use Delete and confirm to remove an entry. The browser editor limits each entry to 2,000 characters.

When the memory tools are available, you can also ask the assistant to remember, recall, correct or forget information. Check the memory page to verify the resulting stored entries. Personal Dots keep their own private memory; in a Dot chat, recall can read both that Dot's memory and your shared memory. Dot memory writes stay with the Dot, and your shared memories are read-only from a Dot chat. The same Memory permission and setting control whether a Dot can use these tools.

Manage personal Dot notes in that Dot's Memory panel. New or edited Dot notes are limited to 2,000 UTF-8 bytes, which can be fewer than 2,000 characters. Organization Dot memory is separate from your profile; see organization memory.

Preferences use a bounded standing section: the driver reads up to 50 recent preferences and applies a 3,000-character budget, while keeping at least the first preference. Older or omitted entries can be looked up through recall. Turning Memory off stops its use and standing-context injection; it does not delete the stored entries. Deleting an entry does not remove text already present in earlier chat messages.

Create an API token

Open Settings → Tokens (/settings/tokens).

1. Select Create token.
2. Give the token a name that identifies the application or device.
3. Choose its lifetime in days. The UI accepts 1–1,825 days and starts with 90.
4. Choose Inherit from groups or a priority up to your inherited user priority. Higher values require an administrator.
5. Create it and copy the plaintext value from the reveal panel immediately.
6. Follow the Guides tab for the client you want to configure.

For OpenCode v2, the Guides → OpenCode tab shows a prompt and a command. The prompt is for OpenCode itself: paste it into a session — in the terminal or the Desktop app — and the agent downloads the AIplane plugin, already set up for your gateway, into ~/.config/opencode/plugins/. The command does the same from a terminal. Either way, paste the token when OpenCode asks for it. The running session picks the plugin up without a restart, and OpenCode then lists every chat model the token may use under croit AIplane, with the context window and capabilities the gateway knows. Signing in with a different token, or revoking it, changes that list.

OpenCode Desktop uses the same flow: on a machine it shares the global configuration directory and background service with the terminal, so the plugin and the sign-in apply to both. Paste the prompt into the Desktop app's chat and the models appear under croit AIplane there just like in the terminal; both pick up gateway-side changes such as a newly granted model without a restart.

The plaintext is revealed at creation or rotation; it cannot be retrieved from a normal token listing. Store it as a credential, not in a document, screenshot or source repository. New tokens have gateway-owned tool execution disabled by default.

Each token card provides four configuration areas:

• Models: restrict the token to selected models within its effective permissions. The picker groups models by kind, such as chat, embedding and reranking. Administrator restrictions are displayed and remain applicable.
• Tools: enable gateway-owned tools and set their capability states. A separate MCP policy switch controls MCP access for the token.

• Budget: add request, token or cost quotas with hourly, daily, weekly or monthly windows. Owner limits still apply and are displayed separately; user configuration cannot remove administrator-owned quotas.

• Priority: open the priority tile, choose inheritance or a permitted priority, then select Save. The dialog shows the inherited user priority. Cancelling leaves the token unchanged; a failed save keeps the draft open.

Use the token's menu to rotate or revoke it. Rotation reveals a replacement credential; update clients that use the old value. Revocation prevents further use. A revoked token card offers removal. Expired and revoked status are shown explicitly.

Read usage and limits

Open /usage. Choose a period and optionally filter by source, backend or API token. The page summarizes usage and groups it by backend, source, model and token. Users with permission to view all usage also receive an all-users scope and per-user table; other users see their own scope.

Your applicable limits appear in the personal view; a rule replaced by a more specific one for you is left out. A model-scoped limit counts only calls on its models, and an all-models limit next to it counts only the remaining models, unless it is an overall ceiling. A spent cost limit does not block models priced at zero. Cost depends on the installation's configured pricing. An unpriced-model warning means the displayed cost cannot account for those models reliably. If usage collection is disabled, the page reports that condition; an empty accounting view is not proof that no requests ran.

When a limit blocks a request, inspect the window and dimension and contact the administrator if necessary. Token quotas are additional constraints, not permission to exceed owner-level limits.

Enable device notifications

Open Settings → Notifications (/settings/notifications). When the operator enables push notifications and your browser supports them, select Enable and allow the browser permission prompt. Notifications can alert this device when an assistant turn you started finishes while you are away.

Use Disable to stop this device's subscription. If browser permission was denied, change the site's notification permission in the browser first. An unavailable warning indicates installation configuration or browser support must be addressed.

Report a problem

Use the feedback button in the composer, or the floating feedback button on other pages. The dialog lets you describe the problem, attach images and review or annotate a screenshot. Review diagnostic and conversation-sharing consent controls before submitting. The public-tracker confirmation matters: do not submit confidential material to a public issue.

Request priority

Your administrator can prioritize chat and API requests from any application using Low, Medium and High. Tokens inherit your groups unless a token-specific priority is saved. Choose it when creating a token or edit its Priority tile under Settings → Tokens. Owners can select Low, Medium or High up to their inherited user priority, or restore inheritance. Administrators can select any priority. If prioritization is disabled, the UI explains that the saved setting applies when an administrator enables it. Rotation preserves the saved priority. Higher-priority waiting requests receive the next available model slot; already running requests finish normally. Sustained higher-priority load may cause a lower-priority request to time out. Priority does not change model permissions or usage limits.

---

Documentation build: 2610.1.108 · commit b774e1f9e2c2