Posted by VladimirIvlev 4 hours ago
Context: Astra computer-use building PCBs in KiCad was a gpt-3 kind of moment for us. Seeing an agent doing work that’s always required a dedicated professional was both exciting and unsettling. Results are still coarse, but we all saw how it went with coding — if the first take looks like this, we can expect it to be fast, cheap, and reliable in a year.
Our vision:
Most engineers using human tools ← we're here
Most engineers using machines using human tools ← one year from now
Most engineers using machines using tools built specifically for machines ← future
So we basically try to build a tool straight from the future
First, we built the EDA itself - Browser-based - KiCad/Altium imports and exports supported - Shared workspace with real-time Figma-like collaboration
But it's built around MCP workflows - MCP covers all features; anything you can do in traditional EDA, your agent can do in Nomoreda
- MCP is optimized for product design work
- built-in project logic, workflows, and checks - better outcomes with less trial and error compared to bare LLM
- You can get decent outcomes starting from around DeepSeek 4 - level. Frontier models are obviously superior, but not a must.
- Direct tool use cuts token spend compared to computer use
- We don’t resell tokens - you use the harness subscription you already have