⚡ OnyRouter — Automatic Model Selection

Let the system pick the best model per request: set the model field to onysoft/auto.

OnyRouter analyzes the incoming message, classifies the question type and routes the request to the best-fit model. There is no special endpoint — the familiar /v1/chat/completions with the model name "onysoft/auto". Routing is completely free; the request is billed only at the selected real model's catalog price.

POST /v1/chat/completions "model": "onysoft/auto"

The standard chat completions endpoint — OnyRouter activates when the model field is onysoft/auto. All parameters, including streaming, work unchanged.

How It Works

There are two layers. The first is instant heuristics with zero added latency: messages carrying code signals (code blocks, error output, technical commands) go to the coding category; very short greetings go to the quick category.

Messages the heuristics can't settle are shown to a light, fast classifier model that returns a one-word category. Classification runs on the platform's own infrastructure — you pay neither tokens nor fees for it. The category is then mapped to a real model through a continuously updated routing table.

Categories and Routes

Category	When is it chosen?	Routed class
`hizli`	Short/simple questions, greetings, one-line lookups	Leading speed-class model
`genel`	Normal conversation, writing, summarization, translation	Balanced flagship model
`kod`	Programming, debugging, technical commands	Strongest balanced coding model
`derin`	Multi-step analysis, strategy, math/logic	Reasoning flagship
`ekonomik`	High-volume mechanical work such as batch classification/labeling	Best price/performance model
`gorsel`	Image/photo analysis requests	Strong vision-capable model

The routing table is updated as the model catalog evolves — your code keeps saying onysoft/auto while we keep the routes current.

Example Request

                Python
                
            

from openai import OpenAI

client = OpenAI(
    base_url="https://api.onysoft.com/v1",
    api_key="sk-ony-your-api-key",
)

response = client.chat.completions.create(
    model="onysoft/auto",
    messages=[{"role": "user", "content": "Bu Python kodundaki hatayı bul: ..."}],
)
print(response.model)  # yönlendirilen gerçek model
print(response.choices[0].message.content)
            

                cURL
                
            

curl https://api.onysoft.com/v1/chat/completions \
  -H "Authorization: Bearer sk-ony-your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "onysoft/auto",
    "messages": [{"role": "user", "content": "Merhaba!"}]
  }'
            

Transparency: Which Model Was Selected?

The model field of the response shows the routed real model (in streaming, the model field of the chunks). Your usage reports also list the request under the selected model's name. onysoft/auto additionally appears as a virtual model in the /v1/models list.

Billing

Routing and classification are free. The request is billed at the catalog price of the real model OnyRouter selects; there is no separate "router fee".

Limits and Notes

On keys with a model lock (allowed_models), the request returns 403 if the selected model is not on the allow-list — use explicit model names on locked keys.
The classifier layer can add a few seconds to time-to-first-token; for clear-cut cases like code or greetings the zero-latency heuristics kick in, so most requests won't feel it. Keeping latency-critical endpoints on fixed model names is a good middle ground.

For the full guide and architecture, see our blog post: OnyRouter