Documentation
oMLX
Browse

oMLX

Connect RewriteBar to oMLX to rewrite text using models running on your Mac. The dedicated oMLX provider supports streaming responses and lists models available on your server.

Setup oMLX with RewriteBar

  1. Install and start oMLX using the instructions in the official oMLX repository.
  2. Download a text-generation model in oMLX. Note the server address, port, and API key if authentication is enabled.
  3. In RewriteBar, open Settings → AI Providers and add oMLX.
  4. Enter the server connection details:
SettingValue
Base URLhttp://127.0.0.1:8000/v1 by default. Use the address and port configured in oMLX.
API KeyThe server API key from oMLX if authentication is enabled. Otherwise, leave it empty.
Default ModelRefresh the model list and select a downloaded text-generation model.
StreamingEnable to see the answer as it is generated.
  1. Click Verify and Enable to test the selected model.
  2. Select oMLX and its model for your rewriting actions, then try a short piece of text.

If oMLX is missing from the provider menu, check for a RewriteBar update that includes oMLX support.

Choosing a Model

The model list comes from your oMLX server, including model aliases. Select a text-generation model rather than an embedding or reranking model. No OpenAI account or cloud model name is needed.

Start with a model that fits your Mac's available memory. The first request can take longer while oMLX loads it. You can change models in RewriteBar after downloading them in oMLX and refreshing the list.

Supported Features

  • Text rewriting with normal and streaming responses.
  • Text context and command instructions.
  • Thinking behavior controlled by oMLX's per-model settings.

Screenshot context is currently disabled for this provider because the model list does not report which models support images.

RewriteBar connects through the Responses API and requests store: false. This disables Responses API conversation storage for these requests; oMLX's own logging and cache settings are managed separately on the server.

Troubleshooting

Connection refused

Make sure oMLX is running and that the base URL matches its server settings. Include /v1 after the port. The default is http://127.0.0.1:8000/v1.

API key required or authentication failed

Copy the server API key from oMLX into RewriteBar's API Key field, then click Verify and Enable again. Leave the field empty only when server authentication is disabled.

No models or model not found

Download a text-generation model in oMLX, refresh the model list in RewriteBar, and select it as the default model. Refresh again if you rename an alias or remove a model.

Slow first response

Allow time for the model to load. If responses remain slow, select a smaller model or free memory by closing other applications.