AI inference · Token Factory

Open models, served through one API.

Call open chat, code and embedding models from your application, or serve your own fine-tune, without running inference servers. [TBD: launch scope]

Talk to our teamRead the docs

One endpoint for every model.

Your application calls one API. Token Factory routes each request to the model it names.

Token Factory API

Chat model

  • 70B
  • FP8

Code model

  • 32B

Embedding model

  • embed

Your fine-tune

  • LoRA
Illustrative endpoint with sample models.

What you get.

  • A familiar API

    Send requests in the chat-completions format most SDKs already speak. [TBD: OpenAI compatibility]

  • Serverless or dedicated

    Start on shared serverless endpoints, then move to dedicated capacity when traffic grows. [TBD]

  • Your prompts stay yours

    Requests run on Supergate infrastructure, or on yours, and aren’t used to train models. [TBD: data policy]

From key to first token in one request.

Create an API key in the console and call a model from any language that can send HTTP.

  • Pay for tokens, not servers

    Usage is counted per input and output token, with prices listed per model. [TBD: pricing]

  • Bring your own weights

    Serve a fine-tune of a supported model on the same API. [TBD]

  • Usage in the console

    Follow requests and token usage per key and per model.

POST /v1/chat/completions

curl $SUPERGATE_ENDPOINT/v1/chat/completions \ -H "Authorization: Bearer $SUPERGATE_API_KEY" \ -d '{ "model": "chat-70b", "messages": [{"role": "user", "content": "Summarize this assay"}] }'
Illustrative request. Endpoint and model names to be confirmed.

Questions.

Which models are available?

The launch list is still being chosen. Tell us which models you need. [TBD]

Can we run it on our own infrastructure?

Talk to us. Supergate can run the platform on your infrastructure. [TBD: Token Factory on-premises]

How is it priced?

Per token, by model. Prices will be on the pricing page. [TBD]

More solutions.