
I kept hitting the same issues with big AI tools: nerfed “thinking mode”, hidden chain-of-thought, flaky memory. So I built QonQrete: a local-first, file-based agent system where all reasoning + memory live in your repo, with new dual-core context to cut cost & latency.
Check out the product page for in-depth explanation/description and the whole flow of the QonQrete system + architecture: https://www.indiehackers.com/product/qonqrete