Windsurf: Devin's Voice Calls Get a Live Speech Model
WindsurfWindsurf's Devin Cloud moved voice calls onto a live speech model, cutting the gap between starting a call and actually talking to Devin and dropping users straight into call controls with a ringtone while the call connects. Holding Space to talk now works immediately after muting, Devin's spoken replies stay in the correct order in the transcript, and Slack-connected sessions show who joined the call and how long it lasted instead of exposing Devin's internal voice notes.
Key Takeaways
- Live speech model replaces Devin's older voice backend, aimed at cutting the lag between placing a call and actually talking to Devin.
- Calls now drop straight into call controls with a ringtone while connecting, instead of leaving users looking at a blank screen during setup.
- The hold-Space-to-talk shortcut now works immediately after muting, removing a reset step that previously interrupted conversations.
- Devin's spoken replies stay in order in the call transcript, fixing a race condition that could scramble responses under the live model.
- For Slack-connected sessions, the thread now shows who is on a call and its duration instead of exposing Devin's internal voice notes.
- The upgrade shipped inside a broader September 16 Devin Cloud release that also added wake-as-you-type sessions and a configurable send shortcut.
Voice Calls Move to a Live Backbone
On September 16, 2026, Windsurf rebuilt Devin's voice pipeline so that calls now run on a live speech model rather than the older request-response voice stack. The change touches nearly every part of what a call feels like, from how quickly it connects to how reliably the transcript reads back afterward.
Faster, Less Awkward Call Starts
Previously, starting a voice call could leave a user staring at a mostly blank screen while the connection set up. Now, starting a call drops straight into the call controls, with a ringtone playing while the connection completes, so it is immediately clear that the call is in progress rather than stalled.
Fixing the Push-to-Talk Handoff
A small but frequently hit annoyance also got fixed: holding Space to talk now works correctly right after muting. Previously, muting could leave the push-to-talk shortcut in a broken state until the call was restarted, interrupting the back-and-forth rhythm of a conversation.
Replies Stay in Order
Devin's spoken replies now stay in the correct order within the call transcript. On the live model, responses can start generating and streaming before earlier ones fully finish, which previously risked scrambling the order shown in the written record of the call.
Slack Threads Show Who Was on the Call
For teams that connect their Devin sessions to Slack, the thread tied to a session now displays who is on a voice call and how long it lasted, replacing the previous behavior of surfacing Devin's internal voice notes in the thread.
Part of a Wider September 16 Release
The live speech model upgrade landed alongside a larger batch of Devin Cloud changes on the same day. Sleeping sessions now start waking up as soon as a user begins typing in the composer, so Devin is typically ready to work by the time a message is sent. Users can also now choose whether Enter or Cmd/Ctrl+Enter sends a message, and copied Devin messages preserve their formatting when pasted into email or documents while still pasting as Markdown into code editors.