echora README

Echora is a VS Code extension that empowers your CLI agent with STT/TTS and makes it your best companion — chatting, learning, and working, always echoing to you.

Features

Echora is a UI side extension, and we recommend to use it locally for cli agent requires working on your local file system. Though you can use SSH to connect to a remote server, and run a claude instance to actually edit server side files, but it is not vscode-native, meaning that VS Code remote extension is not working, and you must opt in to do this.

• NodeSpeech voice input (@vscode/node-speech): Local Hosted STT AI. Multi-language recognition Supported.
  • Realtime STT
  • VAD
• Voice pipeline send modes
  • Default: final voice transcription is sent immediately and pipeline enters thinking.
  • Optional: enable echora.pipeline.enableTextEditingBeforeSend to review/edit transcription in the sidebar and click Send (or Ctrl/Cmd + Enter) before entering thinking.

Requirements

• VS Code Desktop >= 1.85
• Node.js >= 18 on system PATH (required by Agent SDK for Web Streams API and global fetch)
• ms-vscode.vscode-speech installed in the current profile
• OS-level microphone permission granted for VS Code
• If using echora.pipeline.textConsumer = agent-sdk: run pnpm add @anthropic-ai/claude-agent-sdk and ensure Claude authentication is ready
• For debugging, use launch config: Run Extension (UI Host, Recommended)

Extension Settings

This extension contributes the following settings:

• echora.nodeSpeech.locale: Preferred locale for dictation. Use auto to follow VS Code UI language, or set values such as zh-CN or en-US.
• echora.stt.sessionDurationMs: Maximum duration (ms) for a voice input session.
• echora.pipeline.textConsumer: Text consumer backend (agent-sdk or mock).
• echora.agentSdk.tools: Optional SDK tool whitelist (tools). Keep empty to inherit project/default tool set.
• echora.agentSdk.allowedTools: Auto-approved tools (allowedTools) such as WebSearch, WebFetch, Write.
• echora.agentSdk.disallowedTools: Always-blocked tools (disallowedTools).
• echora.agentSdk.permissionMode: SDK permission mode. Default inherit uses project/user .claude settings.
• echora.agentSdk.allowDangerouslySkipPermissions: Required guard flag for permissionMode = bypassPermissions.
• echora.pipeline.remote: [Experimental] Remote bridge configuration (wsl:<distro> or ssh:<host>). Leave empty to run locally.
• echora.pipeline.enableTextEditingBeforeSend: When true, final voice transcription pauses in sidebar for manual review/edit and only sends after explicit user action. Default is false (auto-send).
• echora.tts.enabled: Enable or disable local TTS playback for assistant final responses in the voice pipeline.
• echora.tts.backend: TTS backend selector. Current value: local (@vscode/node-speech via VS Code Speech extension).

Known Issues

• WSL Workspace Host does not run the voice pipeline (extensionKind: ["ui"] keeps it in UI Host).
  • If VS Code Speech target does not match the current host target, runtime target mismatch errors will occur.
• In Windows + WSL UNC path + UI Extension Host debugging, custom Activity Bar file icons (viewsContainers.activitybar.icon with SVG/PNG path) may fail to render.
  • Workaround: use built-in codicons (for example $(mic)) for the Activity Bar icon in this environment.

Release Notes

Check CHANGELOG.md.