The PRYSM API is OpenAI-compatible and speaks JSON over HTTPS. If you’ve used the
OpenAI API, you already know it — the only differences are an enriched response
(model: "auto" and a prysm block) and a few PRYSM-specific endpoints.
Base URL
https://api.prysm1.com/v1
/v1 version
segment.
Authentication
Every authenticated request carries your secret key as a bearer token. See
Authentication for details and environment variables.
Authorization: Bearer prysm_sk_your_key
Content type
Send Content-Type: application/json with request bodies. Responses are always JSON.
The response envelope
/chat/completions returns a standard OpenAI completion plus a top-level prysm
block describing the routing decision, cost, latency, and a
PrysmProof. OpenAI-compatible clients ignore the extra field; the
SDKs expose it via extension().
{
"id": "prysm-a1b2c3d4",
"object": "chat.completion",
"model": "deepseek-v3.2",
"choices": [ ... ],
"usage": { ... },
"prysm": { "routing": { ... }, "cost": { ... }, "proof": { ... } }
}
Errors
PRYSM uses conventional HTTP status codes. Error bodies wrap a detail object with a
machine-readable error code and a human message:
{
"detail": {
"error": "invalid_api_key",
"message": "Invalid or missing PRYSM API key."
}
}
| Status | Error code | Meaning |
|---|
401 | invalid_api_key | The key is missing, malformed, or revoked. |
404 | proof_not_found | No record matches the given request ID. |
429 | rate_limit_exceeded | You’ve exceeded your plan’s per-minute limit. |
502 | provider_error | An upstream provider returned an error. |
503 | no_provider_available | No provider API keys are configured server-side. |
The SDKs raise these as clean exceptions/errors rather than returning raw payloads, so
you can try/catch around any call.
Versioning & compatibility
PRYSM preserves OpenAI request and response shapes. Unknown fields you send are tolerated,
and the PRYSM-specific additions (routing_mode, brain_config, and the prysm response
block) are designed to be ignored safely by standard OpenAI clients. 
