browserground

The local UI-grounding specialist for hybrid AI agents.
Drop in a screenshot + text target, get a strict JSON bbox.

Why this exists — the hybrid AI argument

Today, most AI agents route every screenshot to a cloud frontier model (GPT-4V, Claude Vision, Gemini) just to find click coordinates. That's a $0.01–0.05 multimodal call adding 800ms–2s of latency, repeated 20–50× per agent run. Cost and latency compound. Screenshots full of private UI leave your machine.

A general 200B-parameter LLM is overkill for "where is the Submit button?" — that's a narrow vision task. The right shape is a hybrid one: cheap fast specialist local models for the dedicated tasks they handle better, and a cloud LLM only for the planning and reasoning it's uniquely good at.

That's exactly what browserground is — the click-grounding specialist you drop in next to your Claude / GPT-4o / Codex agent.

What you get

browserground parse screenshot.png --target "Submit button"
# {"bbox_2d": [344, 612, 478, 658]}

Single-line strict-JSON bbox of the element to click. 100% format compliance on the eval set — no markdown fences, no <ref> tokens, parseable every time.

Install

npm install -g browserground
# or: bun install -g browserground

Prerequisite (one-time)

The CLI shells out to Python for inference. Install once:

# Recommended: uv
brew install uv     # or: pipx install uv
uv pip install --system 'torch>=2.4' torchvision 'transformers>=4.55' 'peft>=0.13' huggingface_hub pillow

# Or pip:
pip install 'torch>=2.4' torchvision 'transformers>=4.55' 'peft>=0.13' huggingface_hub pillow

On first browserground parse, the model auto-downloads to ~/.cache/huggingface/:

• base Qwen/Qwen3-VL-2B-Instruct (~4.3 GB)
• LoRA adapter renezander030/browserground (~67 MB)

Use

# one-shot (loads model each call — slow first time)
browserground parse screenshot.png --target "Submit button"

# daemon mode (loads once, fast subsequent calls)
browserground serve &
browserground parse a.png --target "Chrome icon"
browserground parse b.png --target "the back arrow"
browserground stop

# raw model output (for debugging)
browserground parse screenshot.png --target "X" --text

# status
browserground status

Hook into your agent stack

Claude Code

mkdir -p .claude/skills/browserground
curl -sL https://raw.githubusercontent.com/renezander030/browserground/main/plugins/claude-code/SKILL.md \
  > .claude/skills/browserground/SKILL.md

Codex CLI

Add to your AGENTS.md — spec at plugins/codex/AGENTS.md.

browser-use / Skyvern (Python)

import subprocess, json
def ground(screenshot_path, target):
    out = subprocess.check_output(["browserground", "parse", screenshot_path, "--target", target])
    return json.loads(out)["bbox_2d"]

Benchmark

ScreenSpot-v2 point-grounding accuracy (300 items, 100/split):

v0.1 = one-epoch / 12k-example LoRA. v0.2 (Tier 2, target ≥ 60%) in development.

Limitations

• v0.1 desktop & web accuracy lag mobile (training mix is mobile-heavy)
• English-only training data
• Single-target per call (batch mode in v0.2)
• No mouse-action prediction (only location — pair with an action predictor for full computer-use loops)

Links

• 🤗 Model: https://huggingface.co/renezander030/browserground
• 💻 Source: https://github.com/renezander030/browserground
• 📦 npm: https://www.npmjs.com/package/browserground

Work with me

Built by Rene Zander, freelance AI engineer (DE/EN, remote). Custom local specialists, hybrid AI agent architectures, on-prem LLM deployments — https://renezander.com.

License

Apache 2.0.