Wayfinder AI
Free version for macOS — Apple Silicon

Is your website ready for AI agents?

Compass Desktop crawls your site the way an AI assistant would, by choosing links, completing realistic user tasks. It shows you where agents get through, and where they get stuck. Free to download, free to run, unlimited audits.

We’ll email you the download link — and only use your email for product updates and feedback. No spam, no marketing list.

Unlimited auditsRuns locallyNo card needed

Crawlers map URLs. Agents complete tasks.

Traditional SEO tools tell you whether a page can be reached. Compass tells you whether an AI agent can actually get there, understand it, and use it to answer a real question.

Coverage
Does your site have the right content for AI agents?
Accessibility
Are AI user-agents allowed to access your content?
Navigation
Can agents find their way through the site or do they get stuck?

Unlimited free version

Unlimited audits. Bring your own LLM provider and run as much as you like.

Guided audits

Enter a URL, pick tasks from our 171-task library of common tasks (expanded constantly!) and run. Audits use realistic intents: find the privacy policy, locate pricing, contact the team.

AI readiness score

One score (0–100) built from three dimensions: Coverage (is the content available), Accessibility (are agents allowed in), Navigation (can they find their way). Failures grouped by cause.

Live run view

Per-task progress with outcomes, steps, and scores as they happen. Watch the audit unfold in real time.

Prioritised recommendations

Impact-rated issues with concrete fixes: broken navigation for agents, robots.txt / AI-agent blocking, pages agents can’t reach, and where they get stuck.

AI Access report

Which AI user-agents — ChatGPT, Claude, Perplexity, and more — can reach your site. Preflight verdicts plus robots.txt analysis.

Your keys, your models

Run audits through Anthropic, OpenAI, or any OpenAI-compatible endpoint (connections for ollama and llama-swap baked in). No account needed — audit data never leaves your machine except to the provider you chose.

Also included

  • Task drill-in — the exact path the agent took, the link candidates it scored, and why
  • Journeys — first-link choices from the homepage, sink pages, per-task navigation paths
  • Issues browser — deduplicated findings by severity and category
  • One-click export to JSON, CSV, or HTML
  • History — every past audit, searchable, with score tracking
  • Dark mode and JS-rendered crawling (Playwright) with a static-mode fallback

Get started in 3 minutes

Download, add a provider, run your first audit. No account, no signup, no card.

01

Download

Grab the DMG and drag Compass to Applications. It’s signed and notarised by Apple — opens with a normal double-click, no Gatekeeper right-click dance.

02

Add a provider

In Settings, paste an API key (Anthropic / OpenAI / custom endpoint) or point at a local Ollama / llama-swap server. Free audits use your own key — we never see it.

03

Run an audit

Enter a URL, choose tasks (or a full standard set), press Run. Watch it live, then read the score and recommendations. Export the report in Excel and AI-friendly structured JSON.

Compass 2.1.2-beta · ~320 MB · macOS Apple Silicon (M1+) · signed + notarised by Apple · no auto-updater yet — updates by email

SHA-256: 7ffa0d38066b4cc3847290ee2da91121e4decef1c7ed6f84758a5d65a56d8efe

We’ll email you the download link — and only use your email for product updates and feedback. No spam, no marketing list.

Four teams, one dataset

Compass turns the same AI navigation audit into answers for different disciplines.

UX Research

Map how users actually navigate your site — at scale. Run 171 navigation tasks automatically and see where tasks complete, loop, or fail.

UX Research

The problem

Traditional UX research is expensive and small-sample. You can’t manually test every task on every page for every user segment.

What Compass measures

Task completion rate
Did users find what they were looking for?
Path length
How many steps does it take? Too many?
Drop-off points
Where in the journey do users give up?
Failure type
Navigation, semantic, or depth failure?

A Compass audit found “Find pricing” completed in 2 steps, while “Cancel subscription” needed 7 clicks and a support form. That’s the kind of insight a survey won’t give you.

Accessibility

The problem

Automated accessibility scanners test individual pages. They don’t tell you whether a screen-reader user can actually find a button from the homepage.

What Compass measures

Agent compatibility
Which agents can navigate to this content?
UA_BLOCKED detection
Blocked by bot protection?
JS_REQUIRED detection
Does content render without JavaScript?
Depth failures
Are critical pages buried too deep?

A page can pass axe but still be six navigation steps from the homepage with no semantic link labels. Compass finds the journeys WCAG can’t see.

Conversion Optimisation

The problem

Analytics shows the symptom: drop-off at step 3. Compass shows the cause: a dead-end link, missing nav path, or CTA hidden behind a hover state.

What Compass measures

Task failure by goal
Which conversion-critical tasks fail most?
Path length
How many steps between intent and action?
Failure type
Navigation, semantic, or depth failure?
Specific page failure
Exactly which page is the blocker?

“Find the returns policy” failed on 3 of 5 e-commerce sites — the policy was buried under a hover-only mega-menu. Fixing the navigation took 2 hours. No A/B test needed.

Technical SEO

The problem

A page can be indexed, canonicalised, and technically perfect — but if an AI agent can’t navigate to it, it’s invisible to the users who matter.

What Compass measures

UA_BLOCKED detection
Is your site blocking AI agent user-agents?
JS_REQUIRED detection
Does content render without JavaScript?
Semantic failures
Is your link text meaningless to AI?
Depth failures
Are key pages more than 3 clicks deep?

Compass complements Screaming Frog and Search Console by answering not “Can this page be reached?” but “Can an AI agent complete a task starting from the homepage?”

Questions, answered

Does my data leave my machine?

Pages are fetched by your machine, and history is stored locally. The only calls that leave are the LLM requests to the provider you chose — never to us.

Does it work on big JS-heavy sites?

Yes — a full browser engine is bundled in the app, so JS rendering works out of the box with nothing extra to download. There’s also a static-mode fallback for sites that don’t need it.

What does an audit cost me?

Whatever your LLM provider charges per run — typically pence on cheap models. Run fully local models and it’s free.

Pro — coming soon

The free version is everything above. Pro adds the next layer when it ships — get notified, no pricing announcements yet.

Coming soon

Projects

Group a client’s site with competitors, batch audits, and track score deltas and trends over time.

Coming soon

Presets & custom tasks

Save standard audits and write your own task definitions.

Coming soon

MCP server

Drive Compass from Claude Desktop, Cursor, or your own agents.

Coming soon

Bundled local model

Zero API cost, fully offline audits. Fully post-trained to match frontier AI behaviour in site navigation.

Get an email when Pro launches — and updates on the free version as it improves.

We use your email only for product updates and feedback — no spam, no marketing list.

See where agents get stuck on your site

Run your first AI agent audit in 3 minutes — free, unlimited, and your keys stay yours.