Ollama 0.32.0: bare CLI now launches an agent; legacy models flagged deprecated

▼ WHAT HAPPENED

Ollama v0.32.0 (released 2026-07-11) changes default CLI behavior: running bare `ollama` now opens an interactive agent session — "Chat, Code, & Work" — that chats with models, writes code, searches the web, and delegates tasks. The release notes' example shows the agent defaulting to `glm-5.2:cloud`, a cloud-hosted model. The Codex App integration is renamed to ChatGPT (`ollama launch chatgpt`, with `--restore` to return to your usual ChatGPT profile), and the `ollama launch` menu now lists only the most popular integrations. The release also adds deprecation warnings for older agent models: CodeLlama, Qwen2.5(-coder), Llama 3.x, Mistral, StarCoder, and base DeepSeek-R1 tags now trigger a confirmation prompt before `ollama launch` continues.

▼ OPERATOR ANGLE

Bare `ollama` no longer prints help — it opens an agent session, and the release example defaults it to a cloud model (`glm-5.2:cloud`), so verify which model the agent targets if you want inference to stay on your hardware. If you drive `ollama launch` with Qwen2.5, Llama 3.x, Mistral, CodeLlama, StarCoder, or base DeepSeek-R1, expect an interactive deprecation prompt that can stall scripted workflows; plan a migration to newer agent-capable models.

SOURCE: https://github.com/ollama/ollama/releases/tag/v0.32.0[GITHUB-RELEASE]

▼ ENTITIES REFERENCED

TOOLOllama

[pulse item] · runlocalai.co/pulse/ollama-0-32-0-default-agent-cli-legacy-model-deprecation-warnings