ROUTEAgents & automation
Cognitive Framework with JEV for CLI Development
Dragos Roua@dragosroua
0:42This project pairs the 'Assess Decide Do' cognitive framework with a JEV implementation to automate the creation of a CLI tool that counts words in text files.
TakeSo it first routed to Assess, which wrote specs, then to Do, then it evaluated everything and move the project to Done.
I paired my cognitive framework "Assess Decide Do" with a JEV implementation.
In the movie below, "Assess" and "Do" are LLMs, they output text. "Decide" is JEV, it evaluates the state (which is dynamically updated by LLMS) and route to:
- Assess, if we need more clarity, so specs are updated
- Do, if we have clarity and just need to write code
- Done, if the code manages the specs and the tests are passing
The goal was "CLI that counts words in a folder of text files" - that's it, that was the entire text that was passed to JEV.
So it first routed to Assess, which wrote specs, then to Do, then it evaluated everything and move the project to Done.
The whole flow took less than one minute.
The LLM was DeepSeek on OpenRouter, JEV was the official @typesafeai API.
First iteration was a local model on my 16GB M1 MBP, for both text/code generation, and JEV. It took about 29 minutes, but it succeeded. The context windows was 4k, the model was Ternary Bonsai 2.
I am absolutely fascinated by the possibilities this System One models are opening!
Related builds
