# Vartari Hub (experimental)

> Choose which model does which job in DevHub, from model servers you run and clients you already pay for.

> **Experiment:** The page carries the badge **Experiment** and the words "this may change". It answers one
> question, which model does which job, and is planned to become an application of its own.

Vartari Hub is in the **System** group of the sidebar. DevHub can hand some of its own work to a model: grouping skills, naming repeated failures, naming the parts of your code, and turning notes into vectors for search. Here you say which model does each job.

## Where a model can come from

- **Servers**: a model server you run, such as Ollama or LM Studio. **Add model server** and DevHub asks it what it has. **Test** and **Test all** check that servers answer.
- **Clients that lend their models**: a connected client that can answer as a model, for example Claude Code, Codex or Copilot. A client lends its models only after it answered a test on [Harnesses](/devhub/features/harnesses/index.md) and you switched on **use its models**. Using them spends that subscription.

## Assign a model to a job

1. Add a server, or test a client so that it offers its models.
2. In the list **What uses what**, choose the **Model** for a job. A job that needs an embedding model offers only those.
3. Choose the **Effort**, or keep the model's own default.

Each job says what the work is and how big a model it needs, in three levels: **Enough**, **Better** and **The ceiling**. A bigger model is not always worth it, and the page says so for each job. The header counts how many jobs are assigned.

## Search by meaning

The job for the Business Context index turns your notes into vectors so that [Business Context](/devhub/features/business-context/index.md) finds them by meaning. Without a model assigned, a banner says the index is waiting for one and search stays text-only. Choose it once: vectors from two models cannot be compared, so changing the model means embedding the notes again.

> **Caution:** Which models work well for search was not checked for these pages. Naming the parts of your code
> does not wait for a model: the index builds without one and uses DevHub's own names.

## Related

- [Harnesses](/devhub/features/harnesses/index.md)
- [Settings](/devhub/features/settings/index.md)