API docs
The Model Store API follows the OpenAI Chat Completions format, so the official OpenAI SDKs and most LLM tools work by changing the base URL and key.
1. Get a key
Sign in to your console and make a key. You get a key ID (msk_…) and a secret (mss_…); the secret is shown once. Your API key is both joined with a dot:
msk_0123456789abcdef.mss_…
Instead of the joined key you can send the headers X-API-Key: msk_… and X-API-Secret: mss_….
2. Base URL
3. Call a model
Python
JavaScript
curl
Model IDs
Each model has one or more optimised variants. Use <model>:<variant> (for example qwen2.5-1.5b:int8-dynamic-cpu) for a specific variant, or just <model> for its default variant. GET /v1/models lists every ID your key can use; each model page shows the measured speed and quality of its variants.
Endpoints
| Method | Path | What it does |
|---|---|---|
| GET | /v1/models | Lists the model IDs you can call. |
| POST | /v1/chat/completions | Chat answer. Supports model, messages, max_tokens (or max_completion_tokens), temperature, top_p and stream. |
With stream: true the answer comes as server-sent events in the OpenAI format; it is generated whole and then sent, so the first chunk arrives with the full text. The response also has usage and a performance block (tokens/s, first-token time).
Limits and errors
| Status | Meaning |
|---|---|
| 401 | Missing or wrong key ID or secret. |
| 403 | The key is switched off, or your plan has ended. |
| 404 | No such model ID (see GET /v1/models). |
| 429 | Requests per minute reached, or your tier's tokens for the month are used up. |
| 503 | The model is not running right now. Retry after a short wait. |
| 504 | The model did not answer in time. Retry, or ask for fewer tokens. |