Ollama v0.24.0 lands with one headline feature: native support for the Codex App, OpenAI's desktop experience for managing coding threads. You launch it with a single command.
ollama launch codex-app
That gets you a full desktop workspace with built-in worktree support and git functionality. For builders running multiple tasks at once, the parallel thread model means you are not bottlenecked on one context window per session.
Built-in browser with annotation
The Codex App ships with a built-in browser that can load local servers and sites. You can annotate directly on a rendered page to request changes. That tightens the loop between running your dev server and describing what needs fixing. No switching windows, no copy-pasting URLs into a chat interface.
Inline code review
Review mode lets you read code, leave comments, and iterate inside the same workspace. The goal is to keep you from bouncing between your editor, a chat window, and a review tool.
Model selection
The release ships with a clear split between cloud and local model options. For difficult coding and agentic tasks, the recommended models are kimi-k2.6 (which includes vision support) and glm-5.1. For local use without an Ollama Cloud subscription, the options are nemotron-3-super, gemma4:31b, and qwen3.6.
The vision support in kimi-k2.6 matters here specifically because of the built-in browser. Annotating a rendered page and sending that context to a vision-capable model is a more direct workflow than describing layout issues in text.
Restoring your previous setup
If you need to roll back, there is a restore flag:
ollama launch codex-app --restore
That brings back your previous Codex App configuration without manual cleanup.
MLX sampler rework
Beyond the Codex App, the release includes a reworked MLX sampler aimed at improved generation quality on Apple Silicon. No benchmark numbers are provided, but if you are running local models on an M-series Mac, it is worth testing your usual prompts against the previous version to see if output consistency improves.
What to do today
If you are building or prototyping on a local machine, pull the update and run ollama launch codex-app. Try the browser annotation feature against a local dev server. If you are on Apple Silicon, run a few generation tasks and compare output quality to v0.23. The restore flag means the cost of trying it is low.