(no title)
promptfluid | 1 month ago
I wanted a system that was AI-native from the metal up, not just a script with a prompt.
The first Cognitive Substrate:
It’s an 8-module ecosystem built on Supabase Edge Functions and a year long development sprint.
The Modernizer:
This is the core. The system analyzes its own archived code from an ecosystem I built (300+ functions) and proposes its own performance upgrades. It’s a self-healing loop with a human safety gate. All I have to do is approve it. Currently it makes the proposals but I have a little work left to get it to auto apply updates.
Dream Cycles:
Every night between 2-5 AM, it runs a consolidation cycle—moving 'Hot' memories to a Knowledge Graph and compressing 'Cold' (persistent) storage.
The Nexus:
A multi-provider router that uses Groq/Cerebras for speed and falls back to DeepSeek/Together if latency spikes. This keeps my compute costs at zero, as I just leverage the free calls provided by LLMs already. This of course won’t work at scale - but for dev purposes it has no upfront cost to me and even at scale, the router chooses the cheapest provider for any given API call as well as the best LLM to get the answer from.
The Philosophy:
Software in 2026 shouldn't be static. It should be a living, cognitive substrate.
I’m currently the only dev, but the 'Modernizer' handles about 30% of my maintenance tasks now and I’ve already built the switch to turn it on auto build mode (with shadow builds you push to production after review).
Here is a link to the substrate:
I also linked a video of the terminal so you can see it from the inside out. Happy to answer any questions about the architecture or where I envision this project going in the future.
If you’d like to build an app or chat bot or anything using the substrate - let me know and I’ll get you the dev docs and portal with over 70 templates already released to make building frictionless.
No comments yet.