Browser Use HUD Environment

A fully functional HUD environment that wraps the Browser Use Python SDK into tool calls. Every Browser Use action (navigate, click, input, scroll, evaluate, etc.) is exposed as a top-level HUD tool.

Includes two scenarios out of the box:

• answer — browse to a URL, complete a task, optionally compare against an expected answer.
• wiki-game — navigate Wikipedia from a start page to a target page using only link clicks (fewer clicks = higher reward).

Deployed environment: hud.ai/environments/98905607-59c6-4f38-a176-43f7b5944a0f

Using this with HUD

Install the HUD CLI / SDK:

pip install hud-python

Run locally — install deps and launch the MCP server:

pip install -e .
hud dev env:env

Deploy — push to the HUD platform:

hud deploy

Train — create tasks from your scenarios on hud.ai, run evals across models, then train directly on successful traces. Fork a base model at hud.ai/models, select a taskset, and the platform handles fine-tuning — each run produces a new checkpoint you can set as HEAD and use immediately.

See the HUD docs for the full workflow: Environments · Tasks & Training · Models

Extending

Fork this repo and edit env.py to add new tools, scenarios, or scoring logic. The environment auto-registers every Browser Use action at import time — add your own with @env.tool() or @env.scenario().