QonQrete
100% file-based, local-first, secure Agentic AI construction yard for planning, writing, reviewing, and iterating on code inside safe sandboxes on your own machine.
About This Project
QonQrete is not another cloud tab with a chatbot veneer. It is a local-first, file-based agent runtime that drops a structured build loop into your workspace, plans work into briqs, generates and reviews code inside a hardened container, and can run with user-gated checkpoints or autonomous cycles. The current repo ships the core CLI/runtime plus VS Code and IntelliJ integrations.
Current Highlights
- Core CLI/runtime with structured tasq → briq → qage workflow
- Workspace deployment into .qonqrete with root tasq.md syncing
- VS Code and IntelliJ / JetBrains integrations
- Hardened container execution with dropped capabilities and non-root runtime
- Support for cloud providers plus local llama.cpp style backends
- Checkpointed or fully autonomous multi-cycle execution
Current State
The attached repo reflects QonQrete v1.2.0. Right now it ships the main runtime, a deploy-to-workspace flow, auto-init on first run, and IDE integrations for both VS Code and JetBrains so the same file-based workflow can be triggered without hand-rolling shell commands every time.
How the Yard Works
- TasqLeveler sharpens the task before the heavy build loop starts
- InstruQtor breaks work down into briqs and generates the qontract
- ConstruQtor writes or modifies code inside the qage
- InspeQtor reviews the result and pushes the next correction loop
- Qontextor and Qompressor keep context and structure manageable across cycles
Why It Hits Different
The whole thing is designed around local ownership. Your code, context, prompts, and pipeline files stay on your own box, inside a deterministic file tree you can inspect and version, instead of disappearing into some SaaS black box.