alright, after about 10 days I can confidently say that the open ai realtime models aren't worth it. - biggest hurdle is cost, it's just too expensive right now for what you can charge your users (outside of some very bespoke situations aka not B2C) - the latency and

Build your entire realtime AI backend in one min No deployment required. We even give you the code $1/hr AI voice with memory, tool-calling, premium voice cloning, realtime, video input (soon)

Your AI voice should be able to laugh, cry, moan, and change emotional tone based on what's being said So should your clones $1/hr, 300ms latency, most realistic voice clones on the market, no hidden fees DMs open

Quick comparison of AI voice providers: ElevenLabs, Cartesia, and Gabber $12/hr + > 1000ms vs $1/hr + < 1000ms

Whole internet is down Gabber is up

I built an AI to teach me guitar

Turn your house into a AI smart house with a webcam

An AI personal trainer to watch your workout, correct your form in realtime, and save reps/sets

Gabber v2 is here Build AI apps that can see, hear, speak, and interact with the world - in realtime Video-based tool calls, realtime conversation, $1/hr, and a simple app builder The use cases are insane. Leave a comment if you're interested in trying it

Our CTO gave a talk at the AI Engineering World's Fair on how we pulled off making AI Voice more accessible at $1/hr Great listen on productionizing open-source models youtu.be/rD23-VZZHOo?si…

Gabber's pretty cool now. I built an end-to-end AI app that tells me my outfits are awful in realtime Repo below 👇

this is sick

What if the fellowship had access to realtime AI? - track the movements of Orcs - build a clone of Gandalf - give Frodo an AI companion to help with the ring repo in comments

What if Frodo had AI?

I built an AI to give me recipes based on ingredients I show it Uses Qwen 2.5 Omni

Using Qwen Omni 2.5 7b for my LLM, VLM, and STT Realtime conversational AI with the ability to see me Had some issues with longer conversations, but worked well for "what do you see" asks

Jack and I tested Gabber with Qwen 2.5 Omni to see a workout and give real-time verbal feedback. Close, but not there yet: - Struggled to ID exercises with a standard prompt - Mediocre form feedback - Counted objects, not reps Going to try a Qwen vision model later.