GitHub Copilot: Built-In Voice Dictation Arrives in VS Code
GitHub CopilotVS Code 1.131 shipped built-in, offline voice dictation for chat, the text editor, and the integrated terminal, removing the need for the separate VS Code Speech extension. Dictation runs locally on-device using an offline model, and an optional LLM cleanup setting lets Copilot tidy up the spoken transcript by removing filler words and fixing formatting. The feature answers a long-standing developer request for native, reliable voice input inside Copilot Chat.
Key Takeaways
- VS Code 1.131 added built-in voice dictation, in experimental preview, eliminating the need for the separate VS Code Speech extension.
- Dictation works across chat, the text editor, and the integrated terminal, sharing one microphone session so recordings don't overlap between surfaces.
- The feature runs on an offline, on-device model (internally called Nemotron) that downloads once and never sends audio to the cloud.
- Platform support currently covers Windows, Linux (glibc 2.34+), and Apple Silicon Macs — Intel Macs and web-based VS Code are not yet supported.
- An optional
dictation.experimental.llmCleanupsetting uses Copilot itself to strip filler words and clean up formatting in the spoken transcript. - The feature directly answers a long-standing GitHub Community feature request for reliable native voice input in Copilot Chat.
Sources & Mentions
4 external resources covering this update
Visual Studio Code 1.131 zeroes in on subagents
InfoWorld
VS Code 1.131 Adds Built-In Dictation, Hybrid Markdown Editing and Subagent Status
Visual Studio Magazine
Visual Studio Code 1.131 Adds Built-In Dictation, Subagent Visibility, and Agent Polish
NT Compatible
Feature Request: Native voice dictation for Copilot Chat in VS Code is urgently needed
GitHub
Native Dictation, No Extension Required
With VS Code 1.131 (released July 29, 2026), Microsoft added built-in dictation directly into the editor, in experimental preview. Previously, voice input required installing the separate VS Code Speech extension, which developers had long criticized for unreliable accuracy on technical vocabulary — code symbols, file names, and multi-step instructions. The new built-in transcription service works natively across three surfaces: chat inputs, the text editor, and the integrated terminal, sharing a single speech session and microphone selection across all three so recordings don't overlap when switching between them.
Runs Locally, Stays Private
Dictation is powered by an offline model, internally referred to as Nemotron, that downloads once on first use and then processes audio entirely on-device rather than sending it to a cloud service. That local-first design is reflected in platform support: dictation currently works on Windows (x64 and Arm64), Linux (x64 and Arm64, requiring glibc 2.34 or newer), and Apple Silicon Macs. It is not yet available on Intel-based Macs or in browser-based VS Code (vscode.dev) builds.
Optional Cleanup Powered by Copilot
Two new settings control the experience. dictation.showTranscript toggles a live transcript overlay while speaking, so developers can see what's being captured in real time. dictation.experimental.llmCleanup hands the raw transcript to Copilot for cleanup — stripping filler words like "um" and "uh" and applying sensible formatting — before it lands in the chat box, editor, or terminal.
Why It Matters
Native dictation has been one of the more persistently requested features from Copilot users, including an explicit GitHub Community feature request calling the lack of reliable voice dictation for Copilot Chat "urgently needed." By building transcription directly into VS Code and running it offline, GitHub sidesteps both the reliability complaints tied to the old extension and the privacy concerns of sending spoken prompts to a remote service. It shipped in the same release as improved subagent visibility, which surfaces which model a running subagent is using, how long it's been running, and what tool call it's currently executing, directly in the parent conversation.