Where UNARY AI is going, and when. Dates are targets we intend to hit; if one moves, this page moves with it.
The agent, the tool surface and the desktop app in the hands of early users. Focused on everyday coding work: reading and editing a real repository, running the build, and keeping the session coherent across long tasks.
Hardening the parts that only show up under sustained use — long sessions, parallel workers, and the cost of every request. The feature surface settles; the emphasis moves to reliability and measured performance.
Open-weights models pretrained by us and specialised for tool use: choosing the right tool, filling its arguments correctly, and recovering when a call fails. Tool use is treated as the training objective rather than a prompt-time add-on.
Every release along the way is listed in the release notes.