Provider Setup

StyleMakar talks to model providers through an OpenAI-compatible HTTP API. That means the provider must expose OpenAI-style endpoints such as GET /models and POST /chat/completions. It does not mean the provider needs to publish an OpenAPI schema.

Configure The UI

The web UI stores provider settings in browser local storage.

1. Open StyleMakar.
2. Expand Advanced checks.
3. Edit Endpoint to the provider base URL, such as http://localhost:1234/v1.
4. Wait for the status and model list to refresh.
5. Select the model you want to use.
6. Run a rewrite.

The endpoint must include http:// or https://. Do not include /models or /chat/completions; StyleMakar appends those endpoint paths itself.

You can also use another provider through API requests by passing provider.baseUrl, provider.model, and optional provider.reasoningEffort to POST /api/rewrite.

Future desktop provider profile work should add editable presets, secure API-key storage, and clearer local-vs-remote labeling.

LM Studio

LM Studio is the default development provider.

1. Install and open LM Studio.
2. Download or load a compatible chat model. The current verified path is a Gemma 4 QAT model.
3. Open the Developer tab.
4. Start the local API server.
5. Keep the server port at 1234, or update request examples to match your chosen port.

The default StyleMakar provider settings are:

{
  baseUrl: 'http://localhost:1234/v1',
  model: 'gemma-4',
  reasoningEffort: 'none',
}

LM Studio can also start its server from the terminal:

lms server start

Check that LM Studio exposes models:

curl http://localhost:1234/v1/models

Then check StyleMakar's API health endpoint:

curl http://127.0.0.1:5174/api/health

A ready response includes lmStudioReachable: true, ok: true, and a model id. If gemma4Found is false, StyleMakar can still use the selected model, but the default Gemma path was not detected.

Other OpenAI-Compatible Providers

Any provider must support this minimum contract:

Requirement	Expected Shape
Base URL	HTTP or HTTPS URL without a trailing endpoint path beyond /v1
List models	GET {baseUrl}/models returning OpenAI-style data[].id values
Chat completions	POST {baseUrl}/chat/completions accepting model, messages, temperature, and stream: false
Response body	OpenAI-style choices[0].message.content text
JSON responses	The model must be able to return JSON objects when prompted

Example request using a custom local provider:

curl -X POST http://127.0.0.1:5174/api/rewrite \
  -H 'Content-Type: application/json' \
  --data '{
    "document": "Our platform leverages advanced AI to improve workflows.",
    "provider": {
      "baseUrl": "http://localhost:11434/v1",
      "model": "gemma3",
      "reasoningEffort": "none"
    },
    "options": {
      "includeDebug": true
    }
  }'

For a remote provider, use https:// and the provider's model id:

{
  "provider": {
    "baseUrl": "https://your-provider.example/v1",
    "model": "provider-model-id"
  }
}

Remote providers that require API keys are not ready for normal app usage yet, because StyleMakar does not currently provide secure API-key storage. Do not put secrets in docs, checked-in files, screenshots, or shared curl examples.

Command-Line Evals

The eval helper scripts call StyleMakar's local API server and can point that server at a different model endpoint.

STYLEMAKAR_EVAL_BASE_URL=http://localhost:11434/v1 pnpm eval:iterations

Use an explicit provider id when you want to pass a named provider or URL through the eval API:

STYLEMAKAR_EVAL_PROVIDER_ID=http://localhost:11434/v1 pnpm eval:iterations:matrix

You can combine endpoint, model, and reasoning settings:

STYLEMAKAR_EVAL_BASE_URL=http://localhost:1234/v1 \
STYLEMAKAR_EVAL_MODEL=google/gemma-4-12b-qat \
STYLEMAKAR_EVAL_REASONING_EFFORT=none \
pnpm eval:iterations

STYLEMAKAR_EVAL_PROVIDER_ID takes precedence over STYLEMAKAR_EVAL_BASE_URL. If neither is set, evals use the built-in lmstudio provider id.

Troubleshooting

Symptom	Check
provider.baseUrl must be an http(s) URL	Include http:// or https://; localhost:1234/v1 is not enough.
/api/health returns 503	Start LM Studio's server and verify curl http://localhost:1234/v1/models.
/api/models returns 502	Confirm the provider supports GET /models at the configured base URL.
Rewrite fails with a provider error	Confirm the model id is loaded and accepted by POST /chat/completions.
Empty or invalid JSON response	Try the verified Gemma QAT path or a stronger instruction-following model.

Source Docs

• LM Studio local API server
• LM Studio OpenAI compatibility endpoints