API keys
An API key is a workspace's credential for servers and SDKs: every request it sends is the workspace's, billed to its credit. A person's own tools, hi, HIPS Code and HIPS Provider, sign in as the person instead and need no key (Sign in with device authorization). API keys lists the workspace's keys; owners, admins and developers create and change them, and billing members read them.
| Field | What it sets |
|---|---|
| Name | Up to 48 characters: letters, digits, spaces, dots, underscores and hyphens. Unique among the active keys, whatever its case |
| Project label | Optional, up to 48 characters, to group keys in Usage; a key without one shows as Default |
| Kind | full: every route, inference, the read side of deployments and usage. inference_only: the inference routes alone. Fixed at creation: make another key for the other kind |
| Pool | Where a request runs when it names no pool: Auto, whichever pool serves the request best, first; or the Certified or the Community pool |
| Pricing mode | The price a request is quoted at when it names none; market today |
| Region | On the higher paid plan: requests run inside the region, and the region's own base URL keeps the connection in it |
| Cross-pool fallback | Off by default. If you enable it, it must still respect your region and data-use rules |
| Limits | Requests and tokens per minute for this key, at or under the workspace's; empty uses the workspace's |
- Field
- Name
- What it sets
- Up to 48 characters: letters, digits, spaces, dots, underscores and hyphens. Unique among the active keys, whatever its case
- Field
- Project label
- What it sets
- Optional, up to 48 characters, to group keys in Usage; a key without one shows as Default
- Field
- Kind
- What it sets
full: every route, inference, the read side of deployments and usage.inference_only: the inference routes alone. Fixed at creation: make another key for the other kind
- Field
- Pool
- What it sets
- Where a request runs when it names no pool: Auto, whichever pool serves the request best, first; or the Certified or the Community pool
- Field
- Pricing mode
- What it sets
- The price a request is quoted at when it names none; market today
- Field
- Region
- What it sets
- On the higher paid plan: requests run inside the region, and the region's own base URL keeps the connection in it
- Field
- Cross-pool fallback
- What it sets
- Off by default. If you enable it, it must still respect your region and data-use rules
- Field
- Limits
- What it sets
- Requests and tokens per minute for this key, at or under the workspace's; empty uses the workspace's
The four requirement defaults, pool, pricing mode, region and fallback, travel with every request of the key unless a request header overrides one. A guardrail assigned to the key applies on top of them.
The secret is shown once, at creation, with a copyable export HIPS_API_KEY= line; only its last four characters are kept. Copy it into your secret store and confirm that you stored it before you close the dialog. Keep keys on the server or in an appropriate secret store, rather than shipping them in a public browser application.
- Rotate makes a new secret with the same name, kind, project, defaults and limits. The old key keeps answering through a grace, so a deploy can switch over, then stops; revoking it ends the grace early.
- Revoke takes effect at the gateway at once: every request with the key is refused from then. A key that ran in the last 24 hours says so first, since something still depends on it.
- Edit changes the name, project, defaults and limits. A revoked key keeps everything but its name.
- Each key shows when it was last used.
Creating or rotating a key asks for a recent sign-in. A workspace that makes many keys within the hour is asked to wait; the Console says how long.
Under the keys, the limits the gateway enforces for the workspace: requests per minute, tokens per minute, concurrent jobs and the maximum request size, per model where an override sets one, each with where it comes from. A key's own cap sits at or under them. Cached input tokens count toward the tokens-per-minute limit. A request over a limit is refused with 429 and its Retry-After; nothing is queued on your behalf. Raising a limit is HIPS's to do.
Related sheets
- Sheet
- 11 of 19
- Detail
- C
- Updated
- 2026-11-11