Ollama v0.32.0 makes agent UI the default

Ollama shipped v0.32.0, the stable release, on July 14. It confirms changes that had been sitting in the v0.32.0-rc0 preview, covered here earlier this month: running the bare ollama command now launches an interactive agent by default, built to code, search the web, and delegate work, rather than opening a plain chat prompt with a single model.

The Codex App integration is renamed to ChatGPT. Launch it with ollama launch chatgpt; a new --restore flag brings back a previous ChatGPT profile’s settings. The ollama launch menu is simplified too, now showing only the most popular integrations, with the fuller list still one command away.

Ollama also added deprecation warnings for older agent models. Launching CodeLlama, Qwen2.5 (or Qwen2.5-coder), Llama 3.x, Mistral, StarCoder, or a base DeepSeek-R1 tag now prompts a warning before the model runs, a clear signal that Ollama considers these superseded for agent use.

What it means

None of this changes what a model can do on its own. It changes what happens by default when you type ollama in a terminal, and it nudges long-time users of the older models toward newer ones. If you run local models for coding tasks, the guide on coding agents for non-developers covers where a tool like this fits next to hosted alternatives.