sled ai

This repository introduces Sled, a tool designed to let you interact with coding agents like Claude Code, Codex, or Gemini CLI from your mobile device using voice commands.

For the Non-Technical Reader

Imagine you're a chef, but your cookbook (the coding agent) is stuck on your desktop. Sled is like a remote kitchen assistant that lets you access that cookbook from anywhere using just your voice. Instead of typing instructions on your phone, you can simply tell it what you want, and it will relay the instructions to the cookbook. This means you can code on the go, hands-free, making it easier to brainstorm and execute ideas, even when you're away from your computer.

For the Technical Reader

Sled is a web UI built with Typescript that locally spawns coding CLI agent processes. It leverages ACP (Agent Control Protocol) to wrap CLI agents, running them as subprocesses and exposing them via JSON-RPC. The architecture includes transcription and text-to-speech capabilities, enabling voice interaction. It runs locally using wrangler, with no cloud deployments required, except for optional voice processing via layercode.com's API. When voice mode is enabled, audio and agent conversations are sent to Layercode.com for transcription and text-to-speech, but this can be disabled in the settings.

• Key Technologies: Typescript, ACP (Agent Control Protocol), JSON-RPC
• Agent Support: Claude Code, OpenAI Codex, Gemini CLI
• Voice API: Optional, via Layercode.com (Cloudflare hosted)

Why It Matters

Sled addresses the challenge of interacting with coding agents while mobile. By enabling voice input, it significantly speeds up the interaction process compared to typing on a phone. The tool prioritizes local processing, ensuring that code, prompts, and session history remain on the user's machine, enhancing privacy. The optional voice API introduces a trade-off between convenience and data transmission, which users can control via settings.

The "Voice AI Space Lab" Idea

Imagine building a voice-controlled code debugging station. You could verbally walk through code, set breakpoints, and inspect variables, all without touching a keyboard. This could revolutionize how developers approach debugging, making it more intuitive and accessible.

The Collaborative CTA

How can we enhance the security of tools like Sled when exposing local services for remote access, ensuring both convenience and robust protection against unauthorized control?

GitHub Repository

#VoiceAI #CodingAgents