I also built a hosted option for the OmniParser network since running it locally might be more setup than the average person wants to deal with, and there&#x27;s a simple interface to create keys on the website. Give it a shot if you wanna try the tool but please don&#x27;t abuse it too hard I&#x27;m paying out of pocket for the GPU. This ain&#x27;t some VC funded project it&#x27;s literally just me.And please let me know if&#x2F;when you find bugs, I&#x27;m definitely expecting you&#x27;ll encounter some and I&#x27;ll do my best to fix them ASAP.

Yeah speed is definitely a weakness right now. One option is to switch the models around and try a faster one, that will probably give you some speed-up at the potential expense of accuracy. You can do that in the .skipperrc file. There&#x27;s some gains to be made for prompting most likely, as well as in the Omniparser model.A more future thing I&#x27;d like to do is have a model specifically for the action selection task which could be a much smaller network. That would really improve the latency.

tried it out and honestly it worked pretty well, but it’s way slower than me just doing the task myself, needs to be faster

As I started to use Claude Code to do more random tasks I realized I could basically build any CLI tool and it would use it. So I built one that controls the browser and open-sourced it. It should work with Codex or any other CLI-based agent!I have a long term idea where the models are all local and then the tool is privacy preserving because it&#x27;s easy to remove PII from text, but I&#x27;d definitely not recommend using this for anything important just yet. You&#x27;ll need a Gemini key until I (or someone else) figure out how to distill a local version out of that part of the pipeline.Github link: <a href="https:&#x2F;&#x2F;github.com&#x2F;moonshinelabs-ai&#x2F;skipper-tool" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;moonshinelabs-ai&#x2F;skipper-tool</a>

Show HN: Give Claude Code control of your browser (open-source)