Sled - Run your coding agent from your phone, with voice
Sled lets you run your coding agent from your phone using voice.
Coding agents need frequent input, but when you step away from your desk they just sit idle. Sled solves this by giving you a voice interface that connects securely to your local agent over Tailscale.
Your code never leaves your machine. You talk, the agent works locally, and you hear the result read back. Works with Claude Code, OpenAI Codex, and Gemini CLI. Fully open source.
Replies
For longer agentic workflows that require input at irregular intervals, this is definitely the way. I'm interested to see how you guys implemented this.
This hits a very real technical pain point for local coding agents.
Current agents (Claude Code, Gemini CLI, Codex, etc.) don’t fail because of model quality — they fail because they need frequent, low-latency human steering. Once you step away from the terminal, the feedback loop breaks.
Sled’s approach is clean and opinionated:
Voice as a lightweight control channel
Agent execution stays fully local
Secure networking via Tailscale instead of proxying code or context
That design choice matters. No cloud relay, no prompt mirroring, no hidden data paths.
This feels less like a “voice feature” and more like missing infrastructure for agent orchestration. Especially useful for long-running tasks, iterative refactors, or high-level task decomposition when you already trust your agent.
Very strong foundation — excited to see how review, confirmation, and agent-state inspection evolve next.
Interesting idea! I’ve been thinking for a long time about making something similar, where you could take a walk in a park but complete tasks at the same time. Правда, my idea wasn’t about programming but about controlling phone apps by voice… but technically it turned out to be almost impossible.
This is such a clever solution for the 'away from desk' bottleneck, Jack! 🚀 Coding agents are powerful, but that idle time while grabbing coffee or stepping away really adds up
Love that it’s fully open-source and uses Tailscale for security — that definitely lowers the barrier for devs to try it out.
Quick question on the voice interface: how does it handle complex technical terms or specific variable names during the 'hear back' phase? Is it clear enough to follow a logical flow without looking at the screen?
Upvoted! Great to see tools that actually respect a dev's time