Skip to main content
POST
Run a credit-metered inference request

Authorizations

X-Api-Key
string
header
required

A connection-bound gk_agent_ credential. REST deliberately accepts it through either X-Api-Key or Authorization: Bearer; X-Api-Key remains supported. Its fixed scopes are intersected with the current live grant before route authorization.

Headers

Idempotency-Key
string
required

A unique key for this logical mutation. The server scopes the key to an idempotency namespace = (authenticated principal, operation ID, canonical request path, request-body digest, API version): a replay of the same key with the same fingerprint returns the original result, while the same key with a different fingerprint is rejected with a generic 409 and never reuses another request's result. Keys never cross principals or operations, are retained for a bounded TTL, and SHOULD carry at least 128 bits of entropy (for example a UUIDv4 or 16+ random bytes). Reuse a key only when retrying the exact same request.

Required string length: 1 - 255

Body

application/json
model
string
required
Minimum string length: 1
messages
object[]
required
max_budget_micros
string
required
Pattern: ^(0|[1-9][0-9]{0,18})$
max_output_tokens
integer
required
Required range: x >= 1

Response

Inference result.

kind
enum<string>
required
Available options:
completed,
refused
request
object
required
body
object
reason
string