GitHub Copilot CLI can now discover and use local Ollama models
GitHub shipped a /model command in Copilot CLI version 1.0.94 that lists local Ollama models alongside the cloud ones, so a developer can switch to a model running on their own machine without leaving the terminal.
Why it mattersA developer who already runs Ollama can send a Copilot CLI session to a model on their own machine, which keeps that session's prompts and output off a cloud provider that was never asked.
A developer with Ollama running on their own machine used to pick between it and GitHub Copilot CLI. GitHub removed the choice: the CLI's /model command now lists both sources together, so a session can be pointed at a local model and switched back to a cloud one at any time. The feature ships in Copilot CLI version 1.0.94-0 and later.
GitHub says the command "lets you discover supported local models from running Ollama instances", shows the provider and endpoint behind each entry, and lets a user add a model either for the current session or without switching to it. A selection takes effect immediately, with no restart.
What the command actually does
/model reads the local Ollama endpoint, lists each model Ollama exposes, and shows provider details next to each one so a developer can tell a cloud model from a local one before picking. A failure to reach a provider surfaces with the error GitHub returns, rather than silently falling back to a different model.
Two actions are available in the picker. "Add and use for this session" switches the current CLI session to the chosen model. "Add without switching" registers it, so a later /model call lists it alongside the others. GitHub says the behaviour mirrors the model picker already in the Copilot desktop app, so a team using both does not need to learn two conventions.
Three limits worth knowing
GitHub lists three limits worth knowing before enabling it. Ollama and the models have to already be installed; the CLI does not install a runtime or download weights. A chosen model has to support tool calling and streaming, which Copilot CLI uses to run its agent steps and to show the output as it is produced; a model missing either will not appear or will fail on use.
The feature keeps the CLI online and keeps telemetry on. A developer who wants offline mode or no telemetry has to set COPILOT_OFFLINE=true separately. That matters for a team on a laptop with no internet or inside a privacy policy, and GitHub says it on the page: a session can run against a model on your machine while the CLI still reaches the cloud for other work.
What is coming
The post previews "intelligent routing with local models" as a next step, which it describes as the CLI deciding when to use a local model and when to use a cloud one, rather than the developer picking by hand. No ship date is given on the page, and no cost figure for either side of the routing is quoted; the feature today is manual selection only.
The announcement names no price for the CLI integration itself, and no limit on how many local models a user may register in one session. The requirement that stays firm: the models have to live on an Ollama instance the CLI can reach, and they have to speak tool calling and streaming.
Source
Discover local models in GitHub Copilot CLI, GitHub Changelog.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.