Your vector store can answer questions. It just can't talk. A Pinecone index. A Postgres with pgvector. A dataset parked in HuggingFace. Today you reach them with code. Vec lets you just ask — out loud. https://voice-vec.vercel.app Because Vec holds no corpus of its own. Nothing to ingest, no schema to match, no data handed over. You connect your own store with your own keys, and the question is answered from that. → Retrieve — Pinecone, Astra, or your own Postgres + pgvector It reads your table's columns and builds the search from what it finds. → Compute — a HuggingFace repo, or any .csv / .parquet URL Paste the link and it's answerable in SQL, over rows nobody embedded. → Act — Gmail, Slack, Notion, GitHub, via your own Composio project "Check my inbox" stops being a question and becomes a call. The agent is never handed a menu of what you own. It holds one tool and goes looking — so the prompt grows with the question, not with your account. All of it spoken in 22 Indian languages, answered in the one you asked in — whatever language your index was written in. 1.9s from the moment you stop speaking to the first audio out. Open source. Every credential stays yours. 🎙️ https://voice-vec.vercel.app 💻 github.com/Hitesh-s0lanki/voice-vec Connect your index and ask it something you'd normally write a query for. #VectorDatabase #AgenticRAG #pgvector #VoiceAI #AIAgents #OpenSource

Vec

Speak, and be answered out loud, in the language you spoke.

Speak, and be answered out loud, in the language you spoke.