You can now talk with Puck in realtime:
Realtime chat with Puck is powered by gpt-realtime-2.1. It delegates work to the Puck agent you already know, powered by GPT-5.6 Sol. Once the agent responds, gpt-realtime-2.1 summarizes the answer out loud while the full response appears in the thread.
You can now have a proper back-and-forth conversation with Puck without waiting for text responses to stream in. Just talk. That makes it easier to coordinate parallel work, get progress updates, send follow-up instructions, or talk through big ideas without typing, whether you are sitting in front of your computer or on the go.
Here are some conversation starters we have used:
- "Check my active threads and tell me which ones need input."
- "Review the messages in the #issues Slack channel where I am tagged. Read them to me one by one so we can talk through how to fix each one."
- "Start an agent to fix the CI failure, tell me what caused it, and keep me updated on the fix."
- "The executor lease reconciliation pipeline broke. Tell me about the changes made to it yesterday."