PenPal

Voice Doc Agent is an AI collaborator that lives inside your document (not a chat sidebar). You hold push-to-talk and speak an instruction like change rough draft to final draft or look up the latest figures and add them here. AssemblyAI's streaming speech-to-text transcribes you live. An LLM agent (Groq, with tool calling) reads the current document and decides what to do, they start acting through its tools: editing text, appending new content, searching the web with Tavily and writing the sourced findings directly into the page. It types its, edits a few characters at a time, like a person, and you can keep typing alongside it without any issues. If your edit moved the text it was targeting, it re-reads the live document and retries. It also talks back and acknowledges you out loud before it starts working, so there's no dead silence. You can interrupt it at any moment: the press of talk stops its voice and its typing mid-word, and your new instruction is handled against the document as it looks right now. Built with AssemblyAI Universal Streaming, Groq, Tavily, Yjs and Node.js. The AssemblyAI key lives on the server, and the browser only receives short-lived streaming tokens.