API reference

Base URL https://api.infersia.com/v1. Shapes match the OpenAI specification; anything additive is marked.

Chat completions

POST/v1/chat/completions

Parameter	Type	Description
`model`required	`string`	A catalogue id, optionally with a routing suffix — see model routing.
`messages`required	`Message[]`	The conversation. Roles: system, user, assistant, tool.
`stream`	`boolean`	Stream the response as server-sent events. Defaults to false.
`stream_options`	`object`	Set `include_usage: true` for a final chunk carrying exact token counts.
`max_tokens`	`integer`	Output ceiling. `max_completion_tokens` is also accepted. Defaults to the variant’s own limit.
`temperature`	`number`	Sampling temperature, 0–2.
`top_p`	`number`	Nucleus sampling.
`stop`	`string \| string[]`	Stop sequences.
`seed`	`integer`	Best-effort determinism.
`tools`	`Tool[]`	Function definitions, where supported.
`response_format`	`object`	`{ "type": "json_object" }` for JSON mode, where supported.
`user`	`string`	Your own end-user identifier. Recorded on the usage row for attribution.
`provider`	`object`	Infersia extension — per-request routing controls.

Parameters not listed here are forwarded to the inference engine unchanged. That includes vLLM extensions like top_k, repetition_penalty, min_p and guided_json. We validate only the fields we make decisions from — the engine is the authority on its own parameters.

Response

{
  "id": "chatcmpl-...",
  "object": "chat.completion",
  "created": 1785116409,
  "model": "qwen/qwen3-8b",
  "choices": [
    {
      "index": 0,
      "message": { "role": "assistant", "content": "..." },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 13,
    "completion_tokens": 3,
    "total_tokens": 16,
    "prompt_tokens_details": { "cached_tokens": 0 },
    "cost": 0.0000011
  }
}

usage.cost is an addition: the exact USD charged for this request. The model field always reports the catalogue id you asked for, not the upstream repo path.

Legacy completions

POST/v1/completions

Prompt-in, text-out. Supported for older clients; prefer chat completions for anything new.

Embeddings

POST/v1/embeddings

Parameter	Type	Description
`model`required	`string`	An embedding model id.
`input`required	`string \| string[]`	Text to embed. Arrays are batched.
`dimensions`	`integer`	Output dimensionality, where the model supports truncation.

Models

GET/v1/models

Unauthenticated. Includes our extensions — quantization, context_length, pricing, supported_parameters — alongside the standard fields.

Key introspection

GET/v1/key

{
  "data": {
    "label": "Production backend",
    "limit": 25.0,
    "usage": 3.417,
    "limit_remaining": 21.583,
    "is_free_tier": false,
    "rate_limit": { "requests": 200, "tokens": 400000, "interval": "60s" },
    "balance": 21.58,
    "balance_formatted": "$21.5834"
  }
}

Credits

GET/v1/credits

Lifetime credited, lifetime spent, and the current balance.

Health

GET/health

Liveness only — no dependencies checked. /ready additionally verifies the database, and /metrics exposes Prometheus metrics.