Skip to main content

The touchstone for your agent's web tools.

Assayers tested gold on a touchstone: rub the metal on the dark stone and the streak shows what it is. Webstone does the same for the browsers and web tools that AI agents rely on. Every tool runs the same frozen tasks, every call is kept as evidence, and the report shows what each one left behind.

Benchmark browsers and Web tools for AI agents through Web Fetch, Web Search, Web Extract and Browser Trajectory Replay. Compare correctness, latency, resource use and recorded cost. Optionally turn saved results into local reports you can inspect, rebuild and share, with drilldown to every individual result.

Quickstart

Install uv, then run the offline example:

git clone https://github.com/lexmount/webstone.git
cd webstone
uv sync --frozen
uv run --frozen webstone validate
uv run --frozen webstone run --delay 0

Open the report path printed in the terminal. The example has one expected pass and one expected failure; it needs no browser, Docker or API key. Python 3.11 or newer is required; CI tests 3.11.

Next, try HTTP, Moli or Chromium on a local article. Optional adapters, service credentials and live evaluation commands are covered in the track guides.

Evaluation tracks

Choose a workflow for its task contract, setup and runnable example.

Track Benchmarks and task suites Guide
Web Fetch LexBench-Web-Fetch, LexBench-WebMainBench, LexBench-WCXB Local example · Full guide
Web Search LexBench-Web-Search: scenario queries and LexBench-SimpleQA-Verified retrieval tasks Local example · Full guide
Web Extract Zyte Product Extraction: 140 product pages; SWDE: 800 pages from 80 sites; Expanded SWDE, NewsXLM and LiveWeb-IE: independently versioned datasets for frozen-page and live-URL extraction Structured field extraction
Browser Trajectory Replay LexBench-Headless-Browser scenarios; curated fixed-action tasks with explicit checks; WebArena replay with the official evaluator Headless inputs and upstream runner · Curated suite · WebArena replay

Use these suites or bring your own tasks. Add browsers and tools through the existing adapter contracts.

Experimental Browser Use gives an Agent a task, lets it choose browser actions, and checks the recorded outcome with the task evaluator. It includes model and planning variation and is evaluated separately.

How each evaluation workflow runs

Web Fetch

The reusable ten-method Web Fetch cohort is selected with --providers configured --provider-config configs/providers.example.json; shared defaults are 30 seconds per call and one failure retry. See the provider contract.

flowchart LR
    A["Frozen HTML or URLs<br/>with references"]:::input --> B["Call tools<br/>or browsers"]:::run
    B --> C["Save responses, files<br/>and measurements"]:::evidence
    C --> D["Programmatic<br/>scoring"]:::score
    classDef input fill:#eff6ff,stroke:#60a5fa,color:#1e3a8a
    classDef run fill:#f5f3ff,stroke:#a78bfa,color:#4c1d95
    classDef evidence fill:#f0fdf4,stroke:#4ade80,color:#14532d
    classDef score fill:#fffbeb,stroke:#fbbf24,color:#78350f
flowchart LR
    A["Fixed queries<br/>Requirements and rubric"]:::input --> B["Call search tools<br/>Save native rankings<br/>and execution evidence"]:::run
    B --> D["Retrieval and<br/>source checks"]:::score
    B -. If graded .-> E["Relevance grades<br/>Human or opted-in LLM"]:::score
    D --> F["Programmatic<br/>metric aggregation"]:::score
    E --> F
    classDef input fill:#eff6ff,stroke:#60a5fa,color:#1e3a8a
    classDef run fill:#f5f3ff,stroke:#a78bfa,color:#4c1d95
    classDef evidence fill:#f0fdf4,stroke:#4ade80,color:#14532d
    classDef score fill:#fffbeb,stroke:#fbbf24,color:#78350f

Relevance grading is scoring; optional verification reviews the resulting evidence and scores.

Browser Trajectory Replay

Headless scenarios and comparison reports

Headless scenario replay checks browser/driver compatibility on local fixtures. For retained Headless comparisons, the upstream runner owns execution and grading; Webstone builds reports from its results. The original benchmark also contains individual CDP and Web-platform probes, so not every task is a multi-step trajectory.

flowchart LR
    A["Local fixtures<br/>Fixed scenarios and driver bindings"]:::input --> B["Replay fixed actions<br/>through upstream drivers"]:::run
    B --> C["Save responses<br/>and measurements"]:::evidence
    C --> D["Inline checks<br/>Compatibility results"]:::score
    classDef input fill:#eff6ff,stroke:#60a5fa,color:#1e3a8a
    classDef run fill:#f5f3ff,stroke:#a78bfa,color:#4c1d95
    classDef evidence fill:#f0fdf4,stroke:#4ade80,color:#14532d
    classDef score fill:#fffbeb,stroke:#fbbf24,color:#78350f

WebArena

flowchart LR
    A["Self-hosted sites<br/>and starting state"]:::input --> C["Fixed Playwright<br/>replay"]:::run
    B["Qualified recordings<br/>and task revisions"]:::input --> C
    C --> D["Save responses,<br/>observations and HAR"]:::evidence
    D --> E["WebArena-Verified<br/>programmatic checks"]:::score
    classDef input fill:#eff6ff,stroke:#60a5fa,color:#1e3a8a
    classDef run fill:#f5f3ff,stroke:#a78bfa,color:#4c1d95
    classDef evidence fill:#f0fdf4,stroke:#4ade80,color:#14532d
    classDef score fill:#fffbeb,stroke:#fbbf24,color:#78350f

Task completion is checked in code, without an LLM judge.

Repository-curated tasks

flowchart LR
    A["Fixed tasks and actions<br/>Expected outputs"]:::input --> B["Fixed-action<br/>replay"]:::run
    B --> C["Save outputs<br/>and measurements"]:::evidence
    C --> D["Programmatic<br/>assertions"]:::score
    classDef input fill:#eff6ff,stroke:#60a5fa,color:#1e3a8a
    classDef run fill:#f5f3ff,stroke:#a78bfa,color:#4c1d95
    classDef evidence fill:#f0fdf4,stroke:#4ade80,color:#14532d
    classDef score fill:#fffbeb,stroke:#fbbf24,color:#78350f

Evaluation approach

Webstone aims for stable, reproducible results and a controlled, trustworthy evaluation process. Three principles guide the evaluation.

  1. Prefer definite answers and programmatic scoring wherever possible, with minimal reliance on LLMs as judges

    When a task has clear answers and acceptance criteria, use code to check the result and reduce variation from model judgment. This does not exclude LLM judges entirely: graded Web Search relevance is a suitable use, with an explicit rubric and retained scoring evidence and grades.

  2. Do not introduce LLMs or Agents to execute the evaluation

    Web Fetch and Web Search call tools directly; trajectory replay follows predetermined actions to check browser compatibility or website task completion. Participants follow the same task contracts, avoiding extra variation from LLM decisions and Agent planning. Each run retains its task inputs, code revisions and execution settings.

  3. Make results easy to review and verify

    Use an LLM as a verifier to add a check of programmatic verdicts and their supporting evidence. Trigger verification only on request; it does not run by default. Generate local HTML reports with drilldown from aggregate metrics to every individual result, including raw responses, action traces, logs and scoring evidence. Reviewing every result manually is expensive, so reports must also support human sampling. Upload to Artifact Site only when publication is requested and authorized.

See evaluation design for detailed scoring and verification requirements.

Reports

Results and evidence are saved locally under ~/.webstone/runs/; use --output /absolute/path/new-run to choose another directory. HTML reports include comparisons, task results, full responses and failure evidence. Local evaluation, scoring and HTML inspection need no Artifact Site account or publishing authorization.

The run command builds a local report by default. Add --no-report to save results without HTML; rescore --no-report updates scores without rebuilding HTML. Build or rebuild a report later from saved evidence without repeating calls:

uv run --frozen webstone run --delay 0 --no-report
uv run --frozen webstone report /absolute/path/to/run

English is the default; add --language zh when running or rebuilding for Chinese. A rebuild without that option keeps the saved language.

Optional sharing

When sharing is requested and publication is authorized, open Artifact Site (Global .com · China .cn), install the skill offered there, and follow it to publish the local report and choose who can access it. Uploading is a separate action; creating a local report does not publish it.

Scoring code ships with Webstone; dataset commits and file hashes are selected in the data lock. Each report records its Webstone build and selected benchmark edition. See versioning for comparison rules, reference inputs for historical data, and the report guide for local builds and formal releases.

Report drilldown, rebuilding and sharing
flowchart LR
    A["Saved results<br/>and evidence"]:::evidence -->|Optional build / rebuild| B["Local HTML report"]:::run
    B --> C["Drill down to each result<br/>Responses · actions · logs"]:::evidence
    B -. Requested and authorized .-> D["Share through<br/>Artifact Site"]:::evidence
    A -. On request .-> V["LLM verifier"]:::score
    V -. Review findings .-> B
    classDef input fill:#eff6ff,stroke:#60a5fa,color:#1e3a8a
    classDef run fill:#f5f3ff,stroke:#a78bfa,color:#4c1d95
    classDef evidence fill:#f0fdf4,stroke:#4ade80,color:#14532d
    classDef score fill:#fffbeb,stroke:#fbbf24,color:#78350f

Rebuild from valid saved evidence without new participant calls. Verification adds review findings and does not automatically replace scores.

Datasets and code

Use LexBench-Web-Fetch (LB-WF) on first mention and LB-WF in subsequent text, tables and charts. LB stands for LexBench. When reporting results, cite the dataset URL and revision. Credit third-party sources by their original names.

Hugging Face datasets

These datasets supply Webstone's versioned tasks, acceptance contracts and reference inputs.

Dataset Short name Role
LexBench-Web-Fetch LB-WF Versioned Web Fetch tasks, acceptance contracts and reference bindings
LexBench-Web-Search LB-WS Scenario search queries and source references; factual QA is provided by SimpleQA Verified
LexBench-WebMainBench LB-WMB Reference-corrected WebMainBench tasks and correction provenance
LexBench-SimpleQA-Verified LB-SQAV Length-qualified SimpleQA Verified retrieval tasks, with original answers
LexBench-WCXB LB-WCXB Source-qualified frozen extraction inputs and references
LexBench-Headless-Browser LB-HB Executable CDP, driver and web-platform tasks; scenario specifications retained as source inputs
LexBench-Trajectory-Replay LB-TR 30 tasks written in Webstone for live public websites, with fixed actions and exact checks
LexBench-WebArena-Replay LB-WAR Full official task catalog; smaller replay editions retained as versioned source inputs
LexBench-Zyte-Product-Extraction LB-ZPE Frozen product pages with price, SKU and stock references
LexBench-SWDE LB-SWDE Multi-domain frozen-page field extraction from SWDE
LexBench-Expanded-SWDE LB-ESWDE Expanded SWDE field-extraction sample
LexBench-NewsXLM LB-NXLM Multilingual news-page field extraction
LexBench-LiveWeb-IE LB-LWIE Record and list extraction from live websites

Webstone wrote the 30 public-web replay tasks. Headless scenario designs informed some interactions; its executable cases and WebArena's tasks are not included in that suite. The task JSON fixes actions and expected outputs, while the target websites remain live.

Evaluation code and upstream sources

Web Fetch and Web Search scoring code lives in this repository under src/lexbench_webfetch/ and src/lexbench_websearch/. Install and run it through the quickstart above.

Component Purpose Code or source
Headless Browser Original CDP, driver and local-fixture runner Original repository
WebMainBench Pinned extraction scorer and upstream attribution Maintained fork · Upstream
WebArena-Verified Official evaluator for replay and experimental Agent integration Maintained fork · Upstream
WCXB Original data source for the qualified HF edition Upstream

Dependency sources and pins cover Playwright, extraction libraries and supporting runtimes. Third-party notices cover attribution and data rights. Cite the benchmark and upstream data used alongside Webstone.

Browsers and tools
Project Type Use
Moli Browser Evaluated through the same task contracts as other participants
WebFetch CLI CLI client Calls the Lexmount WebFetch service
CowAgent Agent framework Web Fetch tool baseline and experimental Agent integration
OpenClaw Agent framework Web Fetch tool baseline and experimental Agent integration
DeepSeek Harness Agent framework Web Fetch tool baseline and experimental Agent integration

Versions

Published versions, changes and dependency snapshots are listed in GitHub Releases. See version rules for release policy and report provenance.

Contribute

Read CONTRIBUTING for code ownership, tests and PR conventions, or open an issue with a reproducible example.

make setup-test
make ci

Architecture · Security · Third-party content · Apache 2.0

Cite Webstone

Project citation: CITATION.cff.

@software{webstone_2026,
  author = {{Webstone contributors}},
  title = {Webstone: Benchmarking Browsers and Web Tools for AI Agents},
  note = {Evaluation harness},
  year = {2026},
  url = {https://github.com/lexmount/webstone}
}

Submit benchmark results

Internal and external contributors use the same reviewed publication flow in webstone-site. Install GitHub CLI and use your existing GitHub account; no separate website account is required.

webstone results template --benchmark lexbench-webfetch --definition split-media --output report.json
# Fill actual measurements, sample scope and public evidence before validation.
webstone results validate report.json
webstone results submit report.json --provider-url your-method=https://example.com/project --dry-run
webstone results submit report.json --provider-url your-method=https://example.com/project
webstone results status 123 --verify-published

Submissions create result-only PRs from personal forks. CI validates the format and immutable history; maintainers review evidence before merging. A merge is distinct from a completed deployment. See the CLI/SDK and PyPI guide for authentication, version-conflict rebasing, public-data requirements and release setup. These commands are available from source until a PyPI release containing them is published; no software version or tag is changed by this feature.

Result submissions use independently versioned vMAJOR.MINOR.PATCH editions with reviewed UTC publication dates; see the result submission guide. These do not bump the Python package version.

Metadata

Release files for webstone 0.8.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for webstone 0.8.0
File Size Uploaded Provenance Actions
webstone-0.8.0.tar.gz 11.0 MB With provenance Details for webstone-0.8.0.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for webstone 0.8.0
File Interpreter ABI Platform Actions
webstone-0.8.0-py3-none-any.whl Python 3 none any Details for webstone-0.8.0-py3-none-any.whl

Total release size: 21.3 MB

Release files / webstone-0.8.0.tar.gz

Download URL webstone-0.8.0.tar.gz
Size 11.0 MB
Tags Source
SHA-256 checksum
How to use checksums
c3e8972c2037072c652698e4b4f4454b29ee74b48c385623a06da672dd4e3885
BLAKE2b-256 checksum
How to use checksums
5d5b6b5e91f9248a6e86445abd5337c340c7b9a86586ee592009163f61c2d32b
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 18, 2026.

Transparency log

Release files / webstone-0.8.0-py3-none-any.whl

Download URL webstone-0.8.0-py3-none-any.whl
Size 10.4 MB
Tags Python 3
SHA-256 checksum
How to use checksums
e3cc1354304d2298b2f2d2b7b15ad0a4e93a263107512b86261309ef44b3ef0b
BLAKE2b-256 checksum
How to use checksums
547076514e8350dc28dab4901cdc83f70995308ce993cde19fbc3640d73b5e1a
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 18, 2026.

Transparency log

Security for webstone 0.8.0

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

Both files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for webstone-0.8.0.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 18, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Release history Release notifications | RSS feed

This release

0.8.0 This release

2 release files Full provenance

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page