I got tired of "prompting hell," so I built OpenVelo: an open-source orchestrator for "fire and forget" AI software generation.

Hey everyone,

Building software with AI usually means you are trapped in prompting hell—writing a prompt, waiting 20 minutes, checking the output, and prompting again. I built OpenVelo to fix this.

It’s an open-source pipeline designed for a "fire and forget" workflow. The AI-driven planning phase happens upfront inside a dedicated Web-UI. Once you generate and finalize a detailed plan with the LLM, the orchestrator takes over. You can walk away and wake up to software that is either ready to use or requires very minimal fixing.

Important caveats: This is not meant to replace quick CLI pair programming, and it will not keep your token consumption down. It is built for bigger projects, refactors, ports, and prototypes where your personal time is more valuable than compute time.

How the architecture works:

\- Scalable by Design: The entire system runs in isolated Docker containers that communicate with each other, making it easily scalable.

\- Web-UI Planning: All AI-driven planning and requirement gathering is completed in the Web-UI before any implementation begins.

\- Implementation Agent: Runs an iterative cycle in an isolated container to write code and pass unit tests.

\- Tester Agent & Orchestrator: Performs real functional testing against the built software. If it fails, they trigger a self-healing process to spin up a new job and fix the exact failure automatically.

Model Access: It uses Kilo for LLM interaction, so you can route it to any model on your host system (local LLMs, MiniMax M3, etc.).




Repository: https://github.com/m0rph3us1987/OpenVelo



Let me know what you think of the architecture!

https://redd.it/1v7wh6a
@r_CoolGithubProjects