Skip to main content
API keys authenticate your requests to the Runcrate Models API and Infrastructure API. Each key is scoped to a workspace.

Creating an API Key

1

Navigate to Keys

Go to Dashboard → Keys and select the API Keys tab.
2

Click Create Key

Click Create API Key.
3

Name Your Key

Give the key a descriptive name (e.g., production, dev-testing).
4

Copy the Key

Your full API key is displayed. Copy it immediately — it will not be shown again.
The full API key is shown only once at creation. If you lose it, you will need to create a new key.

Key Format

API keys follow the format:
The rc_live_ prefix identifies it as a Runcrate API key. The prefix is always visible in the dashboard for identification purposes.

Managing Keys

Activate / Deactivate

Toggle a key between active and inactive states. Deactivated keys will reject all API requests until reactivated.

Delete

Permanently delete a key. Any applications using the deleted key will immediately lose API access.

Last Used

The dashboard shows when each key was last used, helping you identify unused or stale keys.

Limits

Each key can carry three limits, set from its row’s menu under Limits on the API keys page. The API checks them before a request runs, so a request past a limit is refused with the reason and costs nothing. No per-minute cap on paid models beyond a high abuse guard. Set a per-key limit on the API keys page if you want one. Your balance is shared by the requests running at once: each paid request holds what it may cost (its prompt and its max_tokens) until it finishes, so requests running together don’t spend more than the balance. When less is free than a request asks for, it still runs with its answer shortened to what is left (its max_tokens lowered; the answer ends with finish_reason: "length"). When running requests hold all of it, a new one gets 429 in_flight_budget_exhausted with a Retry-After header. An environment’s monthly budget (Settings, Environments) is checked the same way: its model calls stop as soon as the environment, GPU instances included, has spent it. How the monthly limits count:
  • Spend this month starts from what the key (or environment) had already spent when the limit was set, and resets at 00:00 UTC on the 1st.
  • A free request costs nothing and counts nothing. A refunded request (a failed video, say) gives its cost back.
  • Each request is counted at what it may cost before it runs (its prompt and the most it may write, or an image’s or video’s price) until it finishes, so requests running at once can’t add up past the limit. When less than one request’s worth is left, one more may run, so spend can pass the limit by at most that one request’s cost. Close to the limit, requests on the key run a few at a time and the others are asked to retry in a moment (key_limit_near).
  • A change to a key’s limits applies within five minutes.

Error codes

Every refusal has the OpenAI error shape, { "error": { "message", "type", "code", "param" } }, and an X-Request-Id header to quote to support.

When the balance runs out

A call refused because the workspace balance can’t cover it answers 402 and says how to get paid up. Model calls, the REST API (instances, prepaid terms, jobs, schedules, clusters, volumes) and the MCP server all carry the same fields: Where they are:
  • Model calls: inside error, beside its code (insufficient_balance, or insufficient_credits for a balance under $1). The image, video and speech endpoints, whose error is a plain string, carry them beside it with code: "insufficient_balance"; a video that needs more held than is free keeps code: "video_hold_unaffordable".
  • REST API (/api/v1): in error.details, with reason: "insufficient_balance". error.code stays insufficient_credits.
  • MCP: the tool’s error text ends with next_step.
If the balance could not be read just then, the refusal comes without these fields; check GET /api/v1/billing/balance before asking anyone to add credit. What an agent should do:
  1. Stop retrying the call: it fails the same way until credit is added.
  2. If its key or sign-in has the billing permission, add the credit itself with top_up or POST /api/v1/billing/topups. Otherwise tell its user next_step: the amount and the link to tap.
  3. Retry once the user says the credit is in, or once GET /api/v1/billing/balance shows it.
  4. To stop it happening again, suggest auto-recharge (auto_recharge_docs).

Default Key

When you create a workspace, a default API key is auto-provisioned for use with the Playground. You can use this key for testing or create additional keys for your applications.

Security Best Practices

Never Commit Keys

Do not hardcode API keys in source code or commit them to version control. Use environment variables instead.

Rotate Regularly

Create new keys periodically and deactivate old ones to limit exposure.

One Key Per Application

Use separate keys for different applications or environments (dev, staging, production).

Deactivate Unused Keys

Deactivate keys that are no longer in use rather than leaving them active.
API keys are scoped to the workspace they were created in. They cannot access resources in other workspaces.