Show HN: Ved AI Voice Assistant
TL;DR
Ved is a local offline AI agent for Windows, Linux, and macOS. The repo lists 12 GB VRAM and 32 GB RAM for local models; with an OpenRouter cloud fallback, it claims to run on 4 GB VRAM and 16 GB RAM. It includes a Tkinter desktop UI, FastAPI server with SSE, LangGraph orchestration, standard, turbo, and coder modes, per-thread RAG, file uploads, script execution, and human approval for risky tool actions.
Nauti's Take
Good direction, but still clearly a maker project rather than a finished home assistant. The README is more concrete than the HN pitch: hardware targets, model paths, API, UI, and telemetry are spelled out.
That also raises the bar. If the promise is local privacy, remote access, telemetry, tool permissions, and update paths need careful design, otherwise offline just becomes self-hosted risk.
Briefingshow
Ved hits a real demand: people want a voice assistant that does not send every request to Big Tech servers and still runs on consumer hardware. The interesting part is not the Jarvis framing, but the mix of local voice, RAG, coding mode, and tool approvals. The catch: multi-room audio, remote access, and local tool execution quickly turn this into a security and maintenance project.
Sources