Shubham

For anyone curious about the voice quality — this isn't text-to-speech layered on a chatbot. It's real-time bidirectional voice streaming through AWS Nova Sonic. You talk, it listens and responds simultaneously, like an actual phone call. Latency is low enough that it feels natural.

Quick note on how the memory works — Velur doesn't just store chat logs. It extracts specific facts from your conversations (your name, interests, what's happening in your life) and recalls them naturally in future sessions. It's not keyword matching, it's semantic search over everything you've ever shared.

Hey everyone — I'm Shubham, solo founder of Velur. I built this because every AI conversation feels like talking to someone with amnesia. You share something important, come back the next day, and it's completely forgotten. Velur remembers. You mention a job interview on Monday, it asks how it went on Thursday — without you bringing it up. I'm an MBA and AI engineer. Built the entire stack solo...

Wrestling with this exact question for [Velur](https://velur.ai), my voice-first AI companion. We went with fixed pricing ($14.99/mo) because: Credit anxiety kills the experience. You can’t have authentic conversations while mentally calculating costs per sentence. Voice usage is wildly unpredictable. Some days you talk for 30 minutes, other days it’s a quick check-in. Usage-based would...