Choosing a provider

What works without any provider

Search. LensHub ships with its own embedding model that runs locally, so finding things works out of the box with no API key and no external calls.

What needs a provider

Chat — the built-in interface for asking questions in the browser
Automatic tagging — sorting content into domains and topics
Summarising conversations into something searchable

All optional. Plenty of instances run search-only.

Setting it up

Either set it in the environment as a default for the instance:

LLM_PROVIDER=openai
LLM_API_KEY=sk-...
LLM_CHAT_MODEL=<model id>   # as your provider names it

or per organization in Manage → Settings → LLM provider, which overrides the environment. Pick the model there with Load models, which lists what your provider serves and checks the key at the same time. On OpenAI itself an unset model means gpt-4o-mini; other providers need one chosen, and a self-hosted server may leave it empty to use the model it runs.

Which one

If you want	Use
The simplest thing that works	Any hosted OpenAI-compatible provider
Nothing to leave your network	A self-hosted model
Lowest cost	A small hosted model — the tasks are undemanding

The work LensHub asks a model to do is bounded: tag this document, summarise this thread, answer from this retrieved material. A small, cheap model is usually fine. You do not need a frontier model here — your agent is where reasoning quality matters, and that runs wherever you put it.

What works without any provider

What needs a provider

Setting it up

Which one

On this page