Custom Providers

Add any OpenAI-compatible endpoint with a name, base URL, and API key. The provider and its key are stored only in your local database.

Custom models appear in the model picker alongside catalog models; a conversation can switch between them freely.

Fetch the provider's model list automatically or add model ids by hand. Each model is typed as text or image; image-typed models feed the in-chat image generator instead of the chat dropdown.

Model entries carry their own settings, including the tokenizer used for budgeting.

Token counting drives history trimming and lorebook budgets, so the right tokenizer matters for long chats. Every custom model picks from 13 presets:

auto cl100k o200k claude glm5 glm4 deepseek deepseek-v4 llama3 gemma qwen mistral cohere

Or point at any HuggingFace tokenizer by slug or URL:

hf:openai-community/gpt2
hf:https://huggingface.co/Qwen/Qwen2.5-7B/resolve/main/tokenizer.json

HuggingFace tokenizers download once and cache locally. Counting never blocks a send: if a tokenizer fails to load, an approximate fallback counts instead.

Catalog models can route through a billing group that seeds new chats. Custom-provider models bypass routing entirely, so the group has no effect on them.

Requests to custom providers go straight from your browser to your endpoint. UnoRouter servers never see the request, the reply, or your provider key.