Zed with Ollama, LM Studio and llama.cpp: Local Model Setup and Context Length

ollama pull mistral
ollama serve

• Install Ollama from ollama.com/download.
• On macOS you can open Ollama.app instead of running ollama serve.
• In Zed, choose an Ollama model from the model dropdown.

Zed sends the context length to Ollama as num_ctx. The default is 4096 tokens. Raise it for all Ollama models, or per model with max_tokens.

{
  "language_models": {
    "ollama": {
      "context_window": 8192
    }
  }
}

If your Ollama server needs a key, enter it in the provider UI or set OLLAMA_API_KEY. For a remote service, set api_url to the remote endpoint and add the key.

llama serve

lms get qwen2.5-coder-7b
lms server start

• llama.cpp settings go under language_models with the key llama.cpp. The docs example uses api_url http://localhost:8080, and a key can come from LLAMACPP_API_KEY.
• LM Studio keys can come from LMSTUDIO_API_KEY or the provider UI.
• Local models work with Zed Agent and the Inline Assistant. External Agents and Terminal Threads need their own config.