> ## Documentation Index
> Fetch the complete documentation index at: https://docs.hopscotchlabs.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Cline

> Point Cline's OpenAI Compatible provider at Hopscotch: the Base URL field, the model id, and the model metadata Cline asks you for.

export const FeatureStatus = ({missing, feature, instead, detail, children}) => <Info>
    <strong>Not available yet: {missing ?? feature}.</strong>
    {instead ?? detail ? ` ${instead ?? detail}` : null} This page documents what
    the product does today. When that changes, this page changes with it.
    {children}
  </Info>;

[Cline](https://cline.bot) is a VS Code extension that edits your code through a
tool-calling loop. Read [coding harnesses](/integrations/coding-harnesses)
first: the model id edit and the spend rate cap are the same on every harness and
are not repeated here.

<FeatureStatus missing="a proof run of Cline against the production hostname" instead="The settings below were read from the extension's own provider component and its model-refresh handler in the cline/cline repository on 2026-09-08, at the commit its main branch then pointed at. No Cline session has been run against this API." />

## The configuration surface

In Cline's settings, choose the provider named **OpenAI Compatible**. It then
shows a **Base URL** field, an **API Key** field and a **Model ID** field.

**Cline does not append `/v1`.** Its model-refresh handler appends only `/models`
to whatever you type, so the value you paste has to carry the version segment
itself:

```text Base URL theme={"theme":{"light":"vitesse-light","dark":"vesper"}}
https://api.hopscotchlabs.ai/v1
```

Paste the key into **API Key**. Cline reads no environment variable for this
provider: configuration is through the extension, and the settings it writes live
under your home directory rather than in the repository. `.clinerules` in a
project is an instructions file Cline reads into its prompt, not a place
credentials go.

## The model id

**Model ID** is the field, and it takes an arbitrary string, so a pinned id goes
in as it is:

```text Model ID theme={"theme":{"light":"vitesse-light","dark":"vesper"}}
openai/gpt-4o-mini
```

The field is a dropdown when Cline has fetched a model list and a plain text box
when it has not, and it always offers a **Use custom model ID** option, so you
are never blocked by the list.

## Cline will show you our catalog

This is the one behaviour on this page worth knowing in advance. When you change
the Base URL or the API Key, Cline calls `GET` on `/models` under the base URL
you gave it, with your key on an `Authorization` header, and populates the Model
ID dropdown from what comes back.

That call is [`GET /v1/models`](/concepts/models). So the dropdown fills with our
catalog rather than with a vendor's list, already in the pinned `provider/model`
shape, and picking from it gives you a valid id without typing one. It also means
the list is only as long as what your key can call, which is the intended
behaviour rather than a truncated view.

## The model metadata Cline asks you for

The OpenAI Compatible provider has a **Model Configuration** section, and the
fields in it are ones Cline cannot discover for a model it does not recognise:
**Supports Images**, **Context Window Size**, **Max Output Tokens**, **Input
Price** and **Output Price** per million tokens, and **Temperature**.

Two of those matter here.

**Max Output Tokens is the spend rate cap lever.** Cline sends `max_tokens` on
its requests by default, and that figure is what sizes the amount we hold against
your balance before the call. Setting it to something realistic for the work,
rather than leaving it at whatever it defaults to, is what keeps the cap from
binding. See [rate limits and spend
controls](/concepts/rate-limits-and-spend-controls).

**The price fields are Cline's own display, not our billing.** They drive the
cost figure Cline shows you in its chat. What a request actually cost is in your
usage record, which is computed from the rates in force when the request ran.
Filling them in wrongly gives you a wrong number in the extension and changes
nothing about the charge.

Take the context window and maximum output for a model from
`GET /v1/models`, which publishes both per model, rather than guessing.

## What we did not verify

* That a Cline session against this API succeeds. Nothing was run.
* Whether Cline has any feature that calls an embeddings endpoint. Cline's own
  writing says it indexes no codebase and uses no embeddings or vector store,
  reading files directly instead, and we found no embeddings call while reading
  the provider path. We did not audit the whole extension for one, so we do not
  state it as a fact about the current version. If a feature of it fails against
  this API, [no embeddings endpoint](/integrations/coding-harnesses) is the first
  thing to check.
* Whether the extension's documentation still matches its source. It listed a
  capability toggle we could not find in the component, so where the two
  disagree we went with the source and you may see a different set of fields.
* Anything about prompt caching on this provider path. We found no caching
  mechanism in it and no setting for one, which is an absence rather than a
  confirmed negative.