> ## Documentation Index
> Fetch the complete documentation index at: https://docs.yambr.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Live browser
> Each sandbox runs Chromium with CDP exposed — you and the model drive the same browser in real time.
## What it is
Each sandbox container runs Chromium with the **Chrome DevTools Protocol** exposed. Open WebUI's Browser tab bridges CDP over a WebSocket, so the user sees exactly what the model is doing and can click through if needed. Not a screenshot relay — a **shared, interactive browser**.
## How the model uses it
The `playwright-cli` skill wraps Playwright with high-level helpers: navigate, fill forms, click, screenshot, extract content. The model chooses when to use Playwright vs. when to ask for human action.
## How the user uses it
Open the **Browser** tab in the chat's side panel. You're looking at the same Chromium the model is driving. You can:
* **Watch** — every navigation, click, form fill happens live.
* **Take over** — click, type, scroll. Useful for logins, 2FA codes, CAPTCHAs — the model never sees your raw credentials, only the resulting page state.
* **Leave the AI in charge** — the model picks up from wherever you left the page.
## Why a shared browser matters
Non-technical users get stuck at exactly the wrong moments: login walls, Cloudflare, banking 2FA. Shared browser means the user handles the human-in-the-loop step and the model continues — no need for a new skill every time a site adds a challenge.
## Endpoints (self-hosted)
```
GET /browser/{chat_id}/status
GET /browser/{chat_id}/json
WS /browser/{chat_id}/devtools/page/{page_id}
```
On managed Yambr, Open WebUI uses a session-scoped token before opening the WebSocket — there is no public unauthenticated access.
## Related
* [Playwright skill](/skills/reference#playwright-cli)
* [Sub-agents](/features/sub-agents) — Claude Code can drive the browser too