XCodex WASM
Codex running in your browser. No backend.
2 followers
Codex running in your browser. No backend.
2 followers
A real Codex runtime in the browser — with explicit control over what it’s allowed to do. XCodex runs the actual Codex core in WASM and executes locally with browser-native capabilities like DOM inspection, storage access, and network requests. The runtime is designed to be embedded — consumers define what the agent is allowed to do via capability controls, origin policies, and approval flows for high-trust actions. No backend. Execution happens entirely in your browser.
Hey everyone 👋
I built XCodex WASM around a simple idea:
What if the Codex runtime actually ran inside the browser?
Not behind an API.
Not on a server.
Right next to the page it works on.
Once you do that, everything changes.
The agent doesn’t call tools somewhere else — it operates directly on its environment:
• the current page (DOM, UI, interactions)
• local workspace and files
• browser storage and network
• its own execution context
There’s no clean boundary anymore between "agent" and "environment".
It stops being a remote assistant and becomes something you can embed into any web app:
• inspect and act on live UI
• run agents next to your data instead of sending it away
• build workflows without a backend layer
And yes — this also means the browser becomes a pretty hostile place to run an agent, so the runtime includes explicit capability control and approval for high-trust actions.
I didn’t start with a clear use case — I wanted to see if this execution model even works.
Now I’m trying to understand what it’s actually good for.
Curious what you think:
• does it make sense to run agents fully inside the browser?
• what would you build if the runtime was just a dependency?
Everything is open source — feedback is very welcome.