Look-see reads your website the way a user does — in HTML, in rendered pixels, and across the journeys between pages. Then it tells you where it falls apart.

Quick start · What it audits · Architecture · Deploy · Roadmap · Project site

Look-see is an open-source accessibility audit platform. It crawls your site, drives it in a real headless browser, and grades every page against WCAG 2.2 / 2.1 AA, ADA, Section 508, and EN 301 549. Unlike one-page scanners, it follows the journeys users actually take — clicking, navigating, and validating end-to-end flows — so the issues it reports are the ones a real visitor would hit.

Under the hood it's a monorepo of small services: a crawler API, a page builder, four specialized audit workers (content, visual design, information architecture, journeys), an Angular UI, a Chrome extension, and a VS Code extension for real-time feedback in your editor. They trade work over Google Cloud Pub/Sub and describe the site in a Neo4j graph.

Look-see is engineered in public and isn't 1.0 yet. The services below are real and runnable, but the project is still being stitched together — not a polished hosted product. Expect rough edges, roadmap items that are sketched before they're wired up, and docs that occasionally describe intent before they describe reality.

For the current cadence, skim the commits and open issues. A hosted version at look-see.com is planned but not live; self-hosting is the supported path today.

If that sounds more interesting than it sounds like a warning, you're in the right place. PRs welcome. Bug reports from real websites especially welcome.

It looks at pixels, not only at HTML. Most accessibility tools parse the DOM and stop there. Look-see also renders the page and audits the output a user actually sees — contrast measured from the paint on screen, not from stylesheet rules that may be overridden. A page can pass every rule-based checker and still be unreadable: light grey on white, a font that computed to 10px, a button that vanishes below a sticky banner. The visual-design service sees what a user sees.

It audits journeys, not just URLs. Modern applications are state machines. Auditing the landing page is like inspecting a theatre by its lobby. Look-see discovers interactive elements, drives a real headless browser through them, and audits the states you only reach by clicking — mid-login, mid-checkout, mid-search.

It's a graph, not a list. The site is stored in Neo4j, so pages, elements, journeys, and issues relate the way they actually do in a product. Ask it "which heading hierarchy violations affect which user journeys?" and it has an answer, not a spreadsheet.

It meets developers where they work. Chrome extension, VS Code extension, Angular dashboard — the same audit data, three surfaces. Live scores arrive over WebSocket as workers finish, so you watch the audit happen instead of waiting for a PDF.

It's MIT-licensed and self-hostable. One terraform apply on GCP and your data never leaves your cloud. That matters for compliance teams and for anyone who's been burned by a vendor rug-pull.

Every page is scored across four WCAG-aligned categories:

• Accessibility leads & a11y consultants who need repeatable, auditable WCAG reports across large sites.
• Product & engineering teams shipping under ADA, Section 508, or European Accessibility Act deadlines.
• Agencies running periodic audits for clients, with white-labelable reports.
• Open-source maintainers who want a Pa11y / axe-core alternative that handles full user journeys, not just static pages.

• Java 17+ (Eclipse Temurin recommended; CrawlerAPI requires Java 21)
• Maven 3.9+
• Node.js 18.19+ and npm 10+ (for the Angular UI)
• Neo4j 5.x (local Docker is fine)
• Docker (optional, for containerized runs)
• Google Cloud SDK (optional, only for GCP integrations like Vision/NLP)

All Java services depend on the LookseeCore library — build it once, locally.

git clone https://github.com/deepthought42/Accessibility-audit-platform.git
cd Accessibility-audit-platform
(cd LookseeCore && mvn clean install -DskipTests)

cd AuditManager                       # or any other service directory
mvn clean package -DskipTests
java -ea -jar target/*.jar            # -ea enables Design-by-Contract assertions

cd Look-see-UI-v3
npm install
ng serve
# open http://localhost:4200

Every backend service ships a Dockerfile:

cd <service-directory>
docker build -t looksee/<service-name> .
docker run -p 8080:8080 looksee/<service-name>

Look-see is a pub/sub-driven monorepo. A single CrawlerAPI fronts the platform; work flows asynchronously through Google Cloud Pub/Sub topics to specialized workers, then back to the UI over a Pusher WebSocket.

flowchart LR
    Browser([User Browser])

    subgraph Ingestion
        API[CrawlerAPI]
        PB[PageBuilder]
    end

    subgraph Orchestration
        AM[AuditManager]
    end

    subgraph Workers["Audit Workers"]
        CA["contentAudit<br/>alt text · readability"]
        VD["visualDesignAudit<br/>contrast · typography"]
        IA["informationArchitectureAudit<br/>headings · forms · landmarks"]
    end

    AS[audit-service]

    Browser -->|HTTP / REST| API
    API -->|url| PB
    PB -->|page_created| AM
    AM -->|page_audit| CA
    AM -->|page_audit| VD
    AM -->|page_audit| IA
    CA -->|audit_update| AS
    VD -->|audit_update| AS
    IA -->|audit_update| AS
    AS -. Pusher WebSocket .-> Browser

    classDef svc fill:#fff,stroke:#ff1a4b,stroke-width:2px,color:#111
    classDef client fill:#e8f3ff,stroke:#1e7cff,stroke-width:2px,color:#111
    class API,PB,AM,CA,VD,IA,AS svc
    class Browser client

Shared across every Java service: LookseeCore (A11yCore Maven artifact) — Neo4j models, Spring Data repositories, Selenium WebDriver automation, GCP integrations (Storage, Vision, NLP, Pub/Sub), and Pusher broadcasting.

For the full pub/sub topic list, message payloads, and journey-pipeline diagram, see the expanded Architecture details below.

Each top-level directory is a service or library. Click through for service-specific docs.

Look-see deploys to Google Cloud Run via a single Terraform package. See the full deployment guide — required variables, Secret Manager wiring, Pub/Sub topics, and a sample GitHub Actions workflow — in:

→ LookseeIaC/GCP/README.md

The monorepo CI (.github/workflows/ci.yml) uses path-based change detection to build and test only affected modules. Each service also ships its own workflows:

• docker-ci-test.yml — on PR: build, test, Docker build verification
• docker-ci-release.yml — on merge to main: test, semantic version bump, Docker image push to Docker Hub

• 🌐 look-see.com — hosted SaaS launch (coming soon)
• 📊 Trend dashboards — score deltas over time per page and per journey
• 🤖 Auto-fix suggestions — LLM-generated remediation PRs
• 🌍 PDF & Office document audits — go beyond the browser
• 🏷️ White-label reports — for agencies and consultants

Follow the Issues tab to track progress.

PRs are very welcome. Good first issues are tagged in each service's repo. Please:

1. Fork and branch from main.
2. Run mvn install -DskipTests in LookseeCore before building any dependent service.
3. Keep changes scoped to one service where possible — CI will only rebuild what you touched.
4. Enable Java assertions (-ea) when running locally; Look-see uses Design-by-Contract checks.

Look-see is released under the MIT License. © 2024–2026 Look-see contributors.

Individual subdirectories may retain their original Apache 2.0 license files from before the monorepo migration; the MIT license at the repository root governs the project as a whole going forward.

Full system view — the page-audit pipeline, the journey pipeline, dead-letter handling, and real-time broadcasting to the UI.

flowchart TD
    UI([Look-see UI])

    subgraph Ingestion["Ingestion"]
        API[CrawlerAPI]
        PB[PageBuilder]
        EE[element-enrichment]
    end

    subgraph Orchestration["Orchestration"]
        AM[AuditManager]
    end

    subgraph PageAudits["Per-page audit workers"]
        CA[contentAudit]
        VD[visualDesignAudit]
        IA[informationArchitectureAudit]
    end

    subgraph JourneyPipeline["Journey pipeline (domain-level audits)"]
        JX[journeyExpander]
        JE[journeyExecutor]
        JCL[journey-map-cleanup]
        JER[journeyErrors]
    end

    subgraph Broadcast["Real-time broadcast"]
        FB[front-end-broadcaster]
        AS[audit-service]
    end

    UI -->|HTTP / REST| API
    API -->|url · AuditStartMessage| PB

    PB -->|page_created · PageBuiltMessage| AM
    PB -->|page_created| FB
    PB -->|page_created| EE
    PB -->|journey_verified · VerifiedJourneyMessage| JX

    AM -->|page_audit · PageAuditMessage| CA
    AM -->|page_audit| VD
    AM -->|page_audit| IA

    CA -->|audit_update · AuditProgressUpdate| AS
    VD -->|audit_update| AS
    IA -->|audit_update| AS

    JX -->|journey_candidate · JourneyCandidateMessage| JE
    JE -->|journey_verified · re-enter loop| JX
    JE -->|journey_discarded| API
    JE -->|audit_error| AS

    JCL -. scheduled trigger<br/>cleans stale CANDIDATEs .-> JX
    JER -. dead-letter<br/>marks failed as ERROR .-> JX

    FB -. Pusher WebSocket .-> UI
    AS -. Pusher WebSocket .-> UI

    classDef svc fill:#fff,stroke:#ff1a4b,stroke-width:2px,color:#111
    classDef client fill:#e8f3ff,stroke:#1e7cff,stroke-width:2px,color:#111
    classDef journey fill:#fff7e8,stroke:#e39800,stroke-width:2px,color:#111
    classDef ops fill:#f3e8ff,stroke:#7a3cff,stroke-width:2px,color:#111
    class API,PB,EE,AM,CA,VD,IA,FB,AS svc
    class UI client
    class JX,JE journey
    class JCL,JER ops

Each top-level directory is a former standalone repository. Former origin URLs are listed in LEGACY_REMOTES.txt for reference. The monorepo uses a single new history at the root (per-repo commit histories are not included). To preserve old SHAs, clone the URLs in LEGACY_REMOTES.txt into separate folders or use git filter-repo to replay histories under subdirectories.