datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
weblinx-browsergym
WebLINX: Real-World Website Navigation with Multi-Turn Dialogue
Xing Han Lù*, Zdeněk Kasner*, Siva Reddy
💾Code
📄Paper
🌐Website
📓Colab
🤖Models💻Explorer
🐦Tweets
🏆Leaderboard
Your browser does not support the video tag.
This dataset was specifically created to allow WebLINX to be used inside the BrowserGym and Agentlab ecosystem. Please see the browsergym repository for more information.
[!NOTE]
The version associated with this library is WebLINX… See the full description on the dataset page: https://huggingface.co/datasets/McGill-NLP/weblinx-browsergym.fake_browser_state_zooBrowserSparseAttention
BrowserSparseAttention — WebVoyager End-to-End Trajectories
Browser-agent trajectories for Qwen3-VL-8B running the WebVoyager benchmark under
sparse-attention configurations. 60,567 files, ~2.4 GB (trajectories + per-step screenshots +
observation JSONs).
Variants
Summarized set (8 variants × 200 tasks = 1,600 runs) — layout <variant>/webvoyager__<Site>--<id>/:
dense, quest, block_sparse, tsa (default page size) and their *_b64 counterparts (page/block size 64).… See the full description on the dataset page: https://huggingface.co/datasets/JhLU/BrowserSparseAttention.browser-tool-use-trajectories
Browser Tool Use Trajectories
Rights & intended use: legacy public research corpus / portfolio
artifact. Hosted frontier-model outputs are research-only inputs under
project policy (synthetic-factory#161):
intended_use: research_only, project_training_policy: blocked. Not
training data for any model-weight update. Machine-readable record:
rights.json.
Release status: The raw, uncurated payload is now published under
data/raw/. It is available for inspection and… See the full description on the dataset page: https://huggingface.co/datasets/rmems/browser-tool-use-trajectories.BrowserART
Refusal-Trained LLMs Are Easily Jailbroken As Browser Agents
Paper PDF
Homepage
Github
This project contains the behavior dataset in BrowserART, a red teaming test suit tailored particularly for browser agents.
Abstract
For safety reasons, large language models (LLMs) are trained to refuse harmful user instructions, such as assisting dangerous activities. We study an open question in this work: Can the desired safety refusal, typically enforced in chat… See the full description on the dataset page: https://huggingface.co/datasets/ScaleAI/BrowserART.BrowserAgent-Data
BrowserAgent ChatML Dataset (SFT/RFT)
This dataset contains ChatML-style multi-turn dialogues for a browser agent task. The data is prepared as JSON Lines so it can be previewed directly with the Hugging Face Hub Data Visualizer and loaded with the datasets library.
Links
Paper
Github
Files
sft.jsonl — SFT split (one JSON object per line)
rft.jsonl — RFT split (one JSON object per line)
Schema
Each record is a JSON object containing:
messages:… See the full description on the dataset page: https://huggingface.co/datasets/TIGER-Lab/BrowserAgent-Data.browser-agent-tasks
Browser Agent Tasks
Moderate, multi-step browser tasks for collecting agent trajectories and evaluating screenshot, DOM, and DOM-diff evidence.
Files
tasks.jsonl: one task definition per line.
Each task contains:
task_id: stable identifier
category: task family
instruction: complete instruction given to the browser agent
start_url: suggested public starting page
stopping_condition: when the agent must stop
constraints: safety and scope restrictions
These tasks… See the full description on the dataset page: https://huggingface.co/datasets/ishagarg1103/browser-agent-tasks.browser-agent-failure-corpus
Browser Agent Failure Corpus — case-row view
This view contains 30 controlled local regression cases from one owner-produced deterministic run on 2026-08-31 using Qwen/Qwen3-8B-GGUF:Q4_K_M, llama.cpp b10103-c588c4f47, the Codex in-app browser and the historical Mutant Web guard v2, at temperature 0, reasoning budget 0 and at most eight steps per case. It preserves model proposals, guard decisions, executed actions and recorded outcomes for inspectability and reuse. Allowed… See the full description on the dataset page: https://huggingface.co/datasets/mutantweb/browser-agent-failure-corpus.omnimcp_stealth_browser_scraping_teaser
🔬 INSPECT THE DEEPSEEK-R1 REASONING CHAIN LIVE:
Zero hallucinations. Null syntax errors. 100% AST compiler validated.🌐 Live Interactive Reasoning & Code Inspector: https://emgena.com/trainingslager🎁 Claim your Free Starter Kit (Code: STARTER100): https://emgena.com/trainingslager🏷️ Launch Discount: Get 20 € OFF any 500-incident production suite with code LAUNCH20!
📜 Enterprise Compliance: EU AI Act Articles 50 & 53 certified • 100% DSGVO / GDPR clean • Commercial EULA… See the full description on the dataset page: https://huggingface.co/datasets/emgena/omnimcp_stealth_browser_scraping_teaser.BrowserBenchSee: https://github.com/Halluminate/browserbench
browser-agent-tasks
Browser Agent Tasks
Moderate, multi-step browser tasks for collecting agent trajectories and evaluating screenshot, DOM, and DOM-diff evidence.
Files
tasks.jsonl: one task definition per line.
Each task contains:
task_id: stable identifier
category: task family
instruction: complete instruction given to the browser agent
start_url: suggested public starting page
stopping_condition: when the agent must stop
constraints: safety and scope restrictions
These tasks… See the full description on the dataset page: https://huggingface.co/datasets/WootzappLab/browser-agent-tasks.hermes-flight-recorder-browser-tool-calling-trajectories
Hermes Flight Recorder Browser Tool-Calling Trajectories
This dataset repository publishes the exact public-synthetic artifacts used by
the Qwen3-4B browser LoRA case study.
data/browser/flightrecorder_action_sft.jsonl: governed browser train view.
data/development_action_sft.jsonl: frozen multi-scope development file; the
evaluator selects the browser task scope.
data/sealed_final_action_sft.jsonl: original frozen multi-scope final file;
the evaluator selects the browser task… See the full description on the dataset page: https://huggingface.co/datasets/zwright/hermes-flight-recorder-browser-tool-calling-trajectories.whisper-browser-benchmarks
whisper-browser-benchmarks
Measurements from a Whisper transcription pipeline running entirely inside a
browser tab: which audio and video containers the browser will actually decode,
how accurate the smallest usable Whisper size is on clean synthetic speech, how
long transcription takes relative to the length of the clip, what the first
load pulls over the wire, and what happens to clips longer than the model's
30-second window.
Everything here was measured, not quoted from a… See the full description on the dataset page: https://huggingface.co/datasets/ruanjiange/whisper-browser-benchmarks.betterwright-agentic-browser-50k
BetterWright Agentic Browser — 6,093-row stopped checkpoint
This is the public checkpoint of a generation run originally planned for 50,000 rows. Generation was stopped at the account owner's request and the exact 6,093 accepted rows were packaged. It is synthetic training data, not live browser recordings.
Contents
5,971 BetterWright demonstrations and 122 Playwright demonstrations.
32 task domains and 21 browser feature categories.
Harness-shaped conversations… See the full description on the dataset page: https://huggingface.co/datasets/ProCreations/betterwright-agentic-browser-50k.BrowserAgent-SeedData
BrowserAgent-Data
Dataset used in https://github.com/TIGER-AI-Lab/BrowserAgent.
Summary
Total rows: 230,015
Total size: ~29.47 MB
Subsets: 2wiki, bamboogle, hotpot, musique, nq, popqa
Splits and Sizes
2wiki: 22,576 rows (~2.73 MB)
bamboogle: 125 rows (~0.03 MB)
hotpot: 97,852 rows (~14.20 MB)
musique: 12,417 rows (~0.92 MB)
nq: 82,778 rows (~10.14 MB)
popqa: 14,267 rows (~1.45 MB)
Files are stored as Parquet under each subset directory with dev/train/test… See the full description on the dataset page: https://huggingface.co/datasets/TIGER-Lab/BrowserAgent-SeedData.omnimcp_browser_dom_structured_extractor_teaser
🔬 INSPECT THE DEEPSEEK-R1 REASONING CHAIN LIVE:
Zero hallucinations. Null syntax errors. 100% AST compiler validated.🌐 Live Interactive Reasoning & Code Inspector: https://emgena.com/trainingslager🎁 Claim your Free Starter Kit (Code: STARTER100): https://emgena.com/trainingslager🏷️ Launch Discount: Get 20 € OFF any 500-incident production suite with code LAUNCH20!
📜 Enterprise Compliance: EU AI Act Articles 50 & 53 certified • 100% DSGVO / GDPR clean • Commercial EULA… See the full description on the dataset page: https://huggingface.co/datasets/emgena/omnimcp_browser_dom_structured_extractor_teaser.omnimcp_browser_session_pool_rotator_teaser
🔬 INSPECT THE DEEPSEEK-R1 REASONING CHAIN LIVE:
Zero hallucinations. Null syntax errors. 100% AST compiler validated.🌐 Live Interactive Reasoning & Code Inspector: https://emgena.com/trainingslager🎁 Claim your Free Starter Kit (Code: STARTER100): https://emgena.com/trainingslager🏷️ Launch Discount: Get 20 € OFF any 500-incident production suite with code LAUNCH20!
📜 Enterprise Compliance: EU AI Act Articles 50 & 53 certified • 100% DSGVO / GDPR clean • Commercial EULA… See the full description on the dataset page: https://huggingface.co/datasets/emgena/omnimcp_browser_session_pool_rotator_teaser.browserbench-live
BrowserBench Live Tasks
This private dataset contains 292 live-browser task-generation records derived only from the starting_url field of Halluminate/BrowserBench at revision aa56ce5e6331425c29878037fa8c169507deecdc.
Pages were visited live at generation time. Historical prompts, results, and ground-truth URLs were not used. Every source row is retained, including rejected pages, with its rejection reason. Fresh page screenshots are stored under screenshots/.
The dataset… See the full description on the dataset page: https://huggingface.co/datasets/merve/browserbench-live.omnimcp_browser_playwright_stealth_teaser
🔬 INSPECT THE DEEPSEEK-R1 REASONING CHAIN LIVE:
Zero hallucinations. Null syntax errors. 100% AST compiler validated.🌐 Live Interactive Reasoning & Code Inspector: https://emgena.com/trainingslager🎁 Claim your Free Starter Kit (Code: STARTER100): https://emgena.com/trainingslager🏷️ Launch Discount: Get 20 € OFF any 500-incident production suite with code LAUNCH20!
📜 Enterprise Compliance: EU AI Act Articles 50 & 53 certified • 100% DSGVO / GDPR clean • Commercial EULA… See the full description on the dataset page: https://huggingface.co/datasets/emgena/omnimcp_browser_playwright_stealth_teaser.omnimcp_browser_turnstile_solver_teaser
🔬 INSPECT THE DEEPSEEK-R1 REASONING CHAIN LIVE:
Zero hallucinations. Null syntax errors. 100% AST compiler validated.🌐 Live Interactive Reasoning & Code Inspector: https://emgena.com/trainingslager🎁 Claim your Free Starter Kit (Code: STARTER100): https://emgena.com/trainingslager🏷️ Launch Discount: Get 20 € OFF any 500-incident production suite with code LAUNCH20!
📜 Enterprise Compliance: EU AI Act Articles 50 & 53 certified • 100% DSGVO / GDPR clean • Commercial EULA… See the full description on the dataset page: https://huggingface.co/datasets/emgena/omnimcp_browser_turnstile_solver_teaser.awesome-browser-use-prompts
Awesome Browser-Use Prompts
A curated collection of effective prompts for Browser-Use, the framework that enables AI agents to control web browsers. This repository aims to provide examples, templates, and best practices for crafting prompts that maximize the capabilities of Browser-Use agents.
Introduction
Browser-Use allows language models to interact with web interfaces through natural language instructions. Effective prompting is crucial to achieving successful… See the full description on the dataset page: https://huggingface.co/datasets/metehan777/awesome-browser-use-prompts.browser-agent-phase1-sft-action-only
Browser Agent Phase 1 SFT Action-Only
What this is
Action-only step-level chat SFT data for browser-agent training.
Each example teaches the model to predict the next BrowserGym action from:
the original generation-time system prompt used for data collection
task goal and URL
short recent history
current observation text and diagnostics
Assistant targets contain only the next action.
Why this format
This is the primary training format for small-model SFT… See the full description on the dataset page: https://huggingface.co/datasets/saital/browser-agent-phase1-sft-action-only.mimi-browser-binariesTukTak_BrowserLexBench-Browser
LexBench-Browser
LexBench-Browser is a public browser-agent dataset for evaluating agents on real-web workflows.
The v1.0 snapshot contains 210 no-login tasks across 107 distinct websites, with Chinese and
English instructions, task-level reference steps, key points, common mistakes, scoring rubrics,
and robustness tags.
Repository: https://github.com/lexmount/browseruse-agent-bench
Dataset page: https://huggingface.co/datasets/Lexmount/LexBench-Browser
Docs:… See the full description on the dataset page: https://huggingface.co/datasets/Lexmount/LexBench-Browser.functiongemma-browser-edit-e14-4-scale-51200-datasettool-calling-browser-agent-tasks
Dataset Card
Created by: DataCreator AI
Overview
Tool Calling for Agentic Tasks with Multi-Step Workflows contains 1,062 synthetic multi-turn conversations between a user and an AI assistant. The examples primarily focus on practical agentic tasks such as train ticket booking, dynamic form filling, and payment processing. It provides diverse scenarios including successful execution, context retrieval, tool integration, and failure recovery.
The dataset is… See the full description on the dataset page: https://huggingface.co/datasets/DataCreatorAI/tool-calling-browser-agent-tasks.browser-world-models-transitions
Browser World Models — Transitions
(before screenshot, action, after screenshot) transitions from real websites, for training and
evaluating a world model ("simulator") that predicts the consequence of a web action. Part of
the browser-world-models project.
How it was collected
An LLM policy (gpt-5.4-mini) drove vercel-labs/agent-browser
over the WebVoyager task set (642 tasks / 15 sites); the
task questions are the goals. For every action we saved a screenshot… See the full description on the dataset page: https://huggingface.co/datasets/sudac/browser-world-models-transitions.browseragent-sftbrowserbench-live-expanded
BrowserBench Live Expanded
Live-browser tasks derived only from the starting_url field of
Halluminate/BrowserBench at revision
aa56ce5e6331425c29878037fa8c169507deecdc.
Each source page was revisited live at 800×600. Up to three tasks were generated
from the current page observation: information, navigation, and interaction.
Historical prompts, results, screenshots, and ground-truth URLs were not used.
The canonical dataset contains all 876 task slots from 292 source pages. A… See the full description on the dataset page: https://huggingface.co/datasets/merve/browserbench-live-expanded.