AI Fusion docs

Quickstart

Point any OpenAI-SDK-compatible client at https://ai.viktorarsov.com/v1 and supply your project token as the API key.

from openai import OpenAI

client = OpenAI(
    base_url="https://ai.viktorarsov.com/v1",
    api_key="afp_live_…",
)
resp = client.chat.completions.create(
    model="gateway:fast-free",
    messages=[{"role": "user", "content": "Say hi."}],
)
print(resp.choices[0].message.content)

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://ai.viktorarsov.com/v1",
  apiKey: "afp_live_…",
});
const resp = await client.chat.completions.create({
  model: "gateway:fast-free",
  messages: [{ role: "user", content: "Say hi." }],
});
console.log(resp.choices[0].message.content);

curl https://ai.viktorarsov.com/v1/chat/completions \
  -H "Authorization: Bearer afp_live_…" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gateway:fast-free",
    "messages": [{"role": "user", "content": "Say hi."}]
  }'

Models & aliases

You can address a model three ways:

• Alias — gateway:fast-free, gateway:best-free-chat. Resolves to a ranked list. Recommended.
• Provider-qualified — groq/llama-3.1-8b-instant. Pin to one provider.
• Bare — llama-3.3-70b. Matches across providers by display name.

Aliases

Raw models

42 models. Free models show $0.00. Full OpenAPI schema at /docs.

Error shapes

All errors return a uniform JSON envelope and an HTTP status matching the failure class.

{
  "error": {
    "type": "rate_limit_exceeded",
    "message": "Provider quota exhausted; try again in 30s.",
    "request_id": "req_01H…",
    "retry_after_s": 30
  }
}

Error types

Headers & rate limits

Every successful response carries x-gateway-* metadata so you can see which key answered and at what cost without enabling extra logging.

Response headers

Rate limits

Per-tenant limits come from your plan: RPM, monthly requests, monthly tokens. When the RPM bucket is full, the gateway either waits up to max_wait_ms and retries, or returns 429 with next_slot_eta_s.

Idempotency & retries

The gateway retries transient upstream failures automatically (up to 2 per attempt, up to N providers). Your own retries should use an Idempotency-Key header so we can deduplicate if our SSE disconnects mid-stream.