Base URLs
| Client | Base URL |
|---|---|
| OpenAI SDK / generic HTTP | https://formalengines.com/inference/v1 |
| Anthropic SDK | https://formalengines.com/inference |
Raw HTTP calls need the full endpoint path. Do not send inference requests to the website root or /api/.
Endpoints
| Method | Full path | Authentication |
|---|---|---|
| POST | /inference/v1/chat/completions | Bearer key |
| POST | /inference/v1/messages | Bearer key |
| GET | /inference/v1/models | Public |
| GET | /inference/v1/models/{model_id} | Public |
| GET | /inference/v1/health | Public |
| GET | /inference/v1/billing/balance | Account key |
| GET | /inference/v1/billing/usage | Account key |
| GET | /inference/v1/billing/usage/summary | Account key |
| GET / POST | /inference/v1/keys | Account key |
| PATCH / DELETE | /inference/v1/keys/{fingerprint} | Account key |
Use Content-Type: application/json for JSON request bodies. This reference documents the supported onboarding and inference workflows; it does not imply support for every endpoint in an upstream SDK.
Test your integration
Start with an authenticated balance request, then one non-streaming completion, then a streaming completion. Test your error handling with a deliberately invalid model ID. Add tools only after basic requests succeed.
Useful links: quickstart, troubleshooting, API keys, and post-training documentation.