Scheduled tasks: ask your Mac to run a prompt on a schedule, a nightly test run or a Monday morning dependency check, and it runs while you are away. Approvals and results reach your phone exactly as they do for a live turn, and a task that missed its time while the Mac was asleep says so. Also new in 1.3: - The home screen groups every conversation under its project, with the freshest project already open. - Long conversations open on their newest turn at once instead of scrolling in from the top. - The context meter now reads the current step rather than the session total, and shows a ring at a glance. - The Mac companion's new Model Sources keeps OpenCode's model list in step with your own OpenAI-compatible endpoints.

Choose your tools: a new menu in the composer turns each MCP server on or off per conversation, so a thread only carries the tools it needs and your context stays lean. Your choice rides with the session and follows you across devices. Also new in 1.2: - See how full the agent's context window is, live while a turn runs. - The transcript renders markdown tables properly. - Select and copy your own prompts straight out of the history.

Your agent's work is now live in the app, even when it starts somewhere else. - Live sessions: a session you drive from OpenCode in a terminal on your Mac now streams into the app as it happens, with the prompt, the agent's progress, and the finished diff. - Code previews: when the agent writes or edits a file, the code appears in the conversation as a syntax-highlighted block you can expand. - Add any folder as a project: pin a folder on your Mac as a place to start sessions, straight from your phone, and remove it with a swipe. - Fresher lists: sessions and projects now refresh automatically when you return to the app, alongside the existing pull-to-refresh.