LiminalisRuns in the browser. Sits on the desktop when you need the microphone.
The workspace is a web application and always will be — same projects, same documents, same account. The desktop client is a thin native shell around it, and it exists for one reason that a browser can't do well: transcribing a table full of people talking, locally.
Liminalis in the browser
Nothing to install. Everything on this site except local speech-to-text works here — all three lenses, the full retrieval stack, sharing, live co-presence, billing and the admin surfaces.
Requires: a current desktop browser. Some views — the weave board, the escalation matrix, the corpus browser — want real screen width.
Sign-in link coming
Liminalis Desktop
A native shell wrapping the same application, so your session, projects and account are identical. What it adds is on-device speech-to-text — audio is transcribed locally and never leaves the machine. It isn't uploaded, isn't buffered on a server, and isn't sent to a transcription provider.
In development — builds not yet published
The desktop client is fully built, complete, and working well. Speech‑to‑text, streaming, and installer packaging are finished, but no signed build has been published yet. When an installer is available, its links will appear here. See what's shipped for the full picture.
The microphone is the one place “send it to the cloud” isn't obviously fine.
It's a room, not a user
A live session microphone picks up everyone present, including people who never agreed to anything and aren't using the product. Keeping that audio on the machine is the only version of this that's straightforwardly defensible.
Latency at the table
Play doesn't pause for a round trip. Local transcription keeps the loop short enough that speaking is a usable input method rather than a novelty.
It still works offline
Transcription doesn't depend on connectivity, so a bad hotel wifi degrades the assistant rather than the microphone.