@1569126506sudo260

SpeakToDebug is a voice-first debugging partner built on the AssemblyAI Voice Agent API. Every developer knows the moment: you hit a bug while your hands are busy in the editor, or mid-review on a second screen. Chat assistants can talk about your code, but they have never seen it. SpeakToDebug is the colleague with hands on your keyboard: you speak the problem — no context switching, no pasting snippets — and the agent investigates for real. It searches your actual files, opens the suspicious lines, and runs your actual test suite, then answers out loud with the root cause, verified against your code. Under the hood, one WebSocket connection to the AssemblyAI Voice Agent API handles Universal-3.5 Pro speech-to-text, LLM reasoning with tool calling, voice output, and turn-taking. The agent's tool calls (search_code, read_file, run_tests, list_files) are forwarded by the browser to a local server that executes them inside your real repository — paths are jailed to the workspace, so nothing outside the project can be touched. Every spoken answer is grounded in actual tool output from your machine, never invented. The result feels less like a chatbot and more like a teammate sitting next to you: hands-free investigation while you stay in flow. A 90-second demo video with English subtitles is included in the repository (demo/demo.mp4), and the whole project runs locally with npm start after adding an AssemblyAI API key.

SpeakToDebug is a hands-free debugging partner built on the AssemblyAI Voice Agent API. Instead of pasting code into a chat window, the developer simply describes a bug out loud. The agent restates the problem in one precise sentence, then investigates for real: it calls interactive function tools that execute locally on the developer's machine - list_files to map the project, search_code to locate suspicious symbols and error strings, read_file to inspect the exact lines, and run_tests to verify behavior with the project's own test suite. Nothing is uploaded anywhere; the tools run against the real workspace through a local Node server, and results flow back to the agent as tool.result messages. The agent then speaks a root cause and concrete fix grounded in the actual tool output - never invented. Replies are streamed sentence-by-sentence and spoken with native neural voices in Chinese or English, with barge-in support so the developer can interrupt at any time. The universal speech-to-text pipeline handles multilingual input, keyword boosting keeps domain terms accurate, and a live event log plus billing meter make the whole session transparent. All audio is relayed through a local WebSocket bridge so the voice stays smooth even behind restrictive networks. SpeakToDebug turns debugging into a conversation with a colleague who has their hands on your keyboard.

Six Matches. One leaderboard. Build on AMD, earn points and compete for prizes.

🌎 Online Hackathon | 💻 Month-long challenge | Dates: Sep 1–30, 2026 | 🏆 $10,000 Prize Pool ($5k cash + $5k in AAI credits) A month-long online challenge run by lablab.ai together with AssemblyAI, the Voice AI infrastructure company for builders. Every participant builds on AssemblyAI. You can join at any point — registration stays open for the whole build window, so you can start on day one or pick the project up halfway through and still submit. 🌍 Fully online hackathon. Join and build from anywhere in the world.