[FEATURE] Enable voice dictation (/voice) when using AWS Bedrock, Vertex AI, and other third-party providers

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

The /voice dictation feature is currently restricted to users authenticated via a Claude.ai account. It is explicitly unavailable when Claude Code is configured to use AWS Bedrock, Google Vertex AI, Microsoft Foundry, or a direct Anthropic API key.

Many enterprise and team users access Claude Code through AWS Bedrock or Vertex AI for compliance, cost, or organizational reasons and cannot switch to a Claude.ai account login. These users are entirely blocked from using voice dictation.

Issue #58685 captured this as a bug, but it is a deliberate architectural restriction — so filing this as a feature request to ask that the restriction be lifted.

Proposed Solution

Allow /voice to work regardless of the configured model provider.

Two possible approaches:

1. Decouple transcription from Claude.ai auth — Allow the speech-to-text transcription service to be called with a separate API credential (e.g., an Anthropic API key, or a user-supplied STT endpoint), independent of the model provider used for completions.

1. Support a configurable STT backend — Let users configure an external speech-to-text backend (e.g., OpenAI Whisper API, Google Cloud STT, a local Whisper instance) so that voice input works in any provider context.

Either approach would unblock enterprise users on Bedrock/Vertex who rely on voice as part of their workflow.

Alternative Solutions

• Document a supported hook or wrapper pattern for routing voice input through a user-supplied STT service and injecting the transcript as a prompt — even without native /voice support, an officially documented integration path would help.

Priority

Medium — impacts productivity for enterprise users who cannot use Claude.ai account auth.

Feature Category

Voice / input methods

Use Case Example

A developer using Claude Code on AWS Bedrock (e.g., for data residency or enterprise procurement reasons) wants to dictate prompts hands-free. Today there is no supported path to do this. The built-in /voice command is silently absent and there is no documented alternative.

Additional Context

Related closed issue (marked duplicate, originally filed as a bug): #58685