🦎 From French Army officer to open-source dev, say hi to Gecko Agent

[Gecko51]

Introduce yourself

I'm Guillaume, a solopreneur based in France. After 21 years in the French Army, I made a hard pivot into tech, taught myself to code, fell deep into the AI/automation rabbit hole, and never looked back.

I just open-sourced a project:
Gecko Agent a Chrome extension that turns your browser's side panel into an AI-powered automation assistant.

What it does:
You chat with an AI agent and it can actually do things in your browser : read pages, click buttons, fill forms, extract data, navigate across tabs. It runs in a think → act → observe loop, up to 40 iterations, until the job's done.

Where are you in your GitHub journey?

Learning the basics

And where are you going next on GitHub?

Growing my first open-source project Gecko Agent an AI-powered Chrome extension for browser automation. Looking for early feedback and contributors.

What technical skills or projects are you working on?

Building Gecko Agent with React 19, TypeScript and the Chrome Extensions API (Manifest V3).

I'm also deep into AI agent design my extension runs an autonomous think → act → observe loop to automate browser tasks like data extraction and form filling.

[queenofcorgis]

Welcome @Gecko51 ❤️

Thanks for sharing Gecko Agent with the community. Unfortunately, we currently do not allow self-promotion or advertising on the Community Discussions so please edit your post to remove any asks for Stars. We want to make sure there is space for users to ask questions without overwhelming them with other conversations. Thank you for helping us maintain a productive and tidy community for all our members.

[Gecko51]

Thanks for the warm welcome and for letting me know! I'll edit the post right away to remove the ask for Stars. Totally understand the policy, happy to keep things focused on helping the community. Appreciate the heads-up ! 🙏

[Gecko51]

Done ;)

[ajm19826]

Hi @Gecko51 - is there any way that I could embed your AI Agent into my project, still giving you full credit tho? We are looking for an AI agent, and can't find any open source ones until I came across this discussion & your repository.

[Gecko51]

Hey, thanks for the interest, really glad you found the project useful !

Absolutely, feel free to embed it. The repo is open source, so you're welcome to integrate it into your project. Just keep the license and attribution, and you're good to go.
If you run into any issues during integration or have questions, don't hesitate to open an issue, happy to help.

[sanath-kumar-s]

Hi Gecko51, this is a really impressive project.

I went through the repo, and the way you’ve structured the interaction between the agent and browser actions is particularly interesting. Turning the Chrome side panel into an execution layer instead of just a UI is a smart design choice.

The use cases you’re targeting especially things like multi-step navigation, form handling, and data extraction are exactly where this kind of agent can provide real value beyond simple chat interfaces.

One thing I’m curious about: how are you handling reliability across different websites, especially with dynamic DOM changes or unexpected states during those multi-step iterations? It seems like that would be one of the harder problems to solve at scale.

Overall, this looks like a strong foundation for a very practical automation tool definitely something I’d like to follow as it evolves.

[Gecko51]

Hey Sanath,
thanks for the kind words and for actually digging into the repo, always nice when someone reads the code before commenting haha.

To answer your question about reliability with dynamic DOMs, honestly it's one of the trickiest parts.
Here's how I'm approaching it right now:
The agent doesn't rely on fixed selectors or hardcoded paths. Instead, it reads the page state fresh at every iteration of the think.

So if the DOM shifts after a click or a navigation, the agent re-evaluates what's on screen before deciding the next move. It's more like a human adapting to what they see than a brittle script following a recipe.

That said, it's not bulletproof yet. Some edge cases still trip it up, things like heavy client-side rendering where elements take a while to appear or overlays/modals that block interactions. I'm working on better wait strategies and fallback logic for those situations.

The 40-iteration cap also helps, it prevents the agent from getting stuck in infinite loops when something unexpected happens. If it can't figure out the next step, it stops and tells you what went wrong instead of silently failing.

But yeah, scaling reliability across random websites is definitely the long game here.
Every site is its own little puzzle. That's partly why I open-sourced it, more eyes on edge cases means faster progress.