October 7, 2026
Local models are easier to connect to GitHub Copilot CLI: Ollama now appears in /model
An installed Ollama model can be selected directly in an agent session, without manually configuring environment variables.
As of Oct 7, the `/model` command in Copilot CLI 1.0.94-0 discovers supported models in a running Ollama instance. The version was released as a preview.
GitHub Copilot CLI has supported local models since Apr 7, 2026. At the time, GitHub described connecting to Ollama, vLLM and Foundry Local through environment variables. Now the agent lists Ollama models in the same menu as configured models and Copilot's cloud models.
Connect within a session. After running `/model`, select a discovered model and check the provider and connection URL. `Add and use for this session` adds the model and immediately switches to it. `Add without switching` adds it without switching. There is no need to restart the CLI.
What to install first. Model discovery does not install Ollama or download anything. Ollama runs on macOS, Windows and Linux: the official Quickstart instructs you to download the app and choose a local model. On Linux, you can start the server with `ollama serve` if needed.
Install the Copilot CLI preview with `npm install -g @github/copilot@prerelease`. It requires Node.js 22 or later.
The model must support tool calling and streaming responses. If either capability is missing, the CLI returns an error. GitHub recommends a context window of at least 128,000 tokens.
Manual Ollama connections still work. Before starting `copilot`, set `export COPILOT_PROVIDER_BASE_URL=http://localhost:11434` and `export COPILOT_MODEL=YOUR-MODEL-NAME`, substituting the name of your installed model. Local Ollama does not require an API key.
Selecting a local model leaves GitHub telemetry enabled; it can be disabled separately with `COPILOT_OFFLINE=true`. A remote provider still receives prompts and code context over the network.
Source