API documentation
Access supported AI models through a provider subscription, with model coverage, prices and technical conditions in one place.
Send the customer API key in the Authorization: Bearer header over HTTPS. Do not put keys in URLs. Unlimited inference requires the ai:inference scope; training quote, job, status and artifact routes require ai:training. Requests are rate-limited. A training-reference key is additionally restricted to its granted reference model.
POST /api/v1/chat/completions
Authorization: Bearer ai:inference This OpenAI-compatible endpoint accepts a JSON POST request and returns a chat completion. Message content must be a string; accepted roles are system, user and assistant.
The supplied model identifier and messages are forwarded to the currently configured default AI gateway. This does not promise direct routing to a particular provider or require a fixed SDK.
POST https://fusionpointai.com/api/v1/chat/completions · ai:inference
Use the exact provider model identifier shown in the Unlimited API catalogue. Identifiers and availability are catalogue data and can change; the example identifier below is a placeholder, not a working model ID.
Illustrative request only. Replace every angle-bracket value with your own data. Numeric reference, plan and invoice IDs in examples are placeholders, not existing records.
{
"model": "<MODEL_ID_FROM_UNLIMITED_CATALOGUE>",
"messages": [
{
"role": "user",
"content": "<YOUR_PROMPT>"
}
],
"temperature": 0.2,
"max_tokens": 512
}A successful response contains id, object, created, model and the provider choices and usage fields. Invalid request bodies return HTTP 400 with error.message, error.type, error.code and field details. Provider failures return HTTP 500 with provider_error. Authentication, scope and rate-limit failures are also possible.
400 invalid_request_error A required field is missing or a value is outside the accepted format or range.
401 The bearer key is missing, invalid, revoked or restricted.
403 The key lacks ai:inference, or a training-reference key is used for a different model.
429 The request limit has been reached; wait before retrying.
500 provider_error The inference provider did not return a completion.
A subscription provides access to the models included in its provider plan. Sending an inference request does not train or transfer model weights. Monthly and annual amounts come from the active commercial price records.
Use the documented FusionPointAI inference endpoint with a customer API key and an available model identifier. An SDK formats and sends requests; it is not the location where the provider model necessarily runs. Use the API reference for implemented request formats and authentication.
Commercial unlimited access does not remove context-window limits, output limits, model-specific modalities, service protection controls or applicable provider policies. Check the selected model and service terms before designing a workload.
Existing-customer-only plans remain restricted. A model can be available for inference without being approved as a training reference. Temporary operational restrictions do not create a right to a different provider model.
Use scoped API keys, keep secrets on the server, and follow the documented account, key-rotation and access controls. Agree any workload-specific data-handling or support requirements before production use.
Compare eligible plans, review the applicable terms, subscribe through the existing account workflow, obtain your API credentials and validate requests against the API documentation.
Supported models can help with document analysis, multilingual assistance, software workflows and structured information extraction. Choose a model by its documented capabilities rather than assuming every model accepts the same inputs.
No. It provides inference access to supported models. Use Model Training API when you need a separate trained customer model.
No. The original developer remains identified on the provider service page. FusionPointAI supplies the platform integration, commercial access, account management, billing, documentation and support.
No. Training permission, approved educational resources and execution compatibility are checked separately for each eligible configuration.