iFurySt/open-codex-computer-use

A self-hostable agent framework that lets a language model actually control a desktop or browser to complete multi-step tasks — clicking, typing, navigating, running code — without routing everything through a closed proprietary API. The core idea is to give you the infrastructure layer that commercial computer-use products keep behind paywalls or waitlists, so you can wire it to whatever model backend you prefer and run it on your own hardware. For AI engineers building internal automation tools, QA pipelines, or research prototypes, this removes a meaningful bottleneck. You are not locked into OpenAI's rate limits or usage policies when you want an agent to interact with a real GUI environment. The self-hosting angle also matters for teams handling sensitive workflows where sending screenshots to a third-party API is a non-starter. The honest reservation is that this is early-stage, meaning reliability and documentation are still catching up to the ambition. Expect rough edges, incomplete coverage of edge cases, and a setup process that assumes you are comfortable reading source code. -> Best for: AI engineers who need computer-use automation on their own infrastructure without API gatekeeping.