Token Factory overview
The Token Factory serves every open-weight and custom model behind one governed, OpenAI-compatible API, metered by the token and running on Navon metal, so no prompt or output ever leaves the country.
What you can do
Section titled “What you can do”- Chat, embeddings, and rerank through one OpenAI-compatible API.
- A model catalog of open-weight models, including African-language models hosted nowhere else (Swahili, Amharic, Hausa, Yoruba).
- Serverless shared endpoints to start fast, and dedicated endpoints with an SLA when you need guaranteed capacity.
- Batch inference at roughly half the price of live requests.
- Fine-tuning, LoRA, and distillation, deployed to an endpoint in one step.
- Retrieval that resolves through the Data Lake ontology.
Why it is different
Section titled “Why it is different”The economics of a token factory, pay per token with no idle GPU to carry, with in-country data residency, local billing, and zero retention by default. The thing you cannot get by calling an inference API hosted abroad.
- Quickstart — your first request.
- Product reference — capabilities, generated from the registry.