Home / Integrations / Zed

Use Zed with TurboLLM

Zed is a fast editor with a built-in AI assistant that speaks to OpenAI-compatible providers. TurboLLM is one of them — point Zed's assistant at your local endpoint and its AI runs entirely on your own GPU, with nothing leaving the machine.

What you need

Two things, both from a running TurboLLM instance:

  • The OpenAI-compatible base URL: http://localhost:6996/v1
  • A model — load one from your library so the assistant has something to talk to.

An API key is optional. You only need one if you've turned on Require API key in TurboLLM. If that's on, create a key under Developer → API Keys and use it in Zed; if it's off, any non-empty value works.

Point Zed at TurboLLM

  1. Start TurboLLM and load a model

    Make sure the daemon is running on http://localhost:6996 with a model loaded.

  2. Add an OpenAI-compatible provider in Zed

    In Zed's assistant / agent settings, add an OpenAI-compatible provider pointed at http://localhost:6996/v1. If Zed asks for an API key, paste the one from Developer → API Keys (or any non-empty value when Require API key is off).

  3. Select your model

    Choose the model you loaded (for example, local). The assistant will now route its requests to TurboLLM instead of a cloud provider.

Check Zed's docs for the exact fields

Zed's assistant settings UI and the exact field names change across versions, so we keep this high-level on purpose. For the current config path — where to add an OpenAI-compatible provider and what each field is called — see Zed's own documentation. The values you plug in stay the same: base URL http://localhost:6996/v1, an API key only if Require API key is on, and your loaded model.

Recommended models

Coding assistants lean on tool-calling and long context. See the TurboLLM model catalog for good local picks by GPU size, then load one before pointing Zed at it.

Previous← Cline