Changelog

Clicking Regenerate (or Force refresh) asked the model the same question against the same page with no memory of what it had already answered, so it tended to come back with the same, or near-identical, tools — the only suggestions it actually knew to avoid were ones you'd already saved. Every suggestion a site has ever been shown now counts, whether it was saved, left pending, or explicitly dismissed, so the next generation goes looking for something new instead of repeating itself.

Pro previously ran on Gemini 2.5 Pro for tool generation. Comparing output quality side by side against Flash, the difference wasn't large enough to justify roughly 10x the cost, so both tiers now generate on Flash. What Pro still gets is the higher monthly quota and every suggested tool unlocked, rather than only the top two — the generation itself just got noticeably cheaper to run without a meaningful quality trade-off.

A new checkbox next to the session cookie field, off by default: on a page reached with your login cookie, click a small, bounded number of things that look like they only expand or reveal more of the page — an accordion, a "show more", a dropdown toggle — so a tool hidden behind one of them can be found too. The bar for what counts as safe to click is deliberately high: the strongest signal is aria-expanded, the standards-based marker for exactly this kind of disclosure widget, or a native <details>/<summary>. Anything that looks like a form submit, or is worded like delete, pay, invite, upgrade, log out, or similar, is never clicked, no matter what else about it looks safe.

This is opt-in on purpose — it clicks things on a real, authenticated session, so it's your call whether that's worth it for a given site, not something that happens to your account by default.

Naming one page behind your login, like /dashboard, used to mean exactly one page got crawled — any subpage it links to, like /dashboard/settings, needed typing out by hand too, since a public sitemap crawl structurally can't see them. We now follow links found on a page you named into further pages, as long as they match the path globs you've allowed the login cookie to reach, bounded the same way public discovery is so this can't turn one named page into an unbounded crawl of your own app.

Tools generated from a page inside that authenticated area are now also automatically scoped to run only on the page they came from, rather than site-wide, since a tool built against one page's layout has no reason to work anywhere else.

Naming a page behind a login, like a dashboard, is exactly the case that setting exists for. Those pages are almost always marked noindex for search engines, which is unrelated to whether you want a tool generated from them, but the crawler was treating the two as the same thing and silently dropping the page, with no error, just zero tools. A page you name explicitly now overrides noindex; auto-discovered pages still respect it.

Crawl results are cached for a day so repeated generations don't re-fetch your whole site. That cache ignored your settings: if you added pages under "Pages to crawl", ticked "only crawl the pages I list", or pasted a login cookie, the next generation still replayed yesterday's snapshot of whatever it had found on its own. The cache now keys on those settings, so changing them re-crawls immediately.

Two smaller fixes alongside it. A wildcard pasted into "Pages to crawl" (that field wants real paths; the cookie allowlist below it is the one that takes globs) is now trimmed to the page it points at, so /dashboard/** crawls your dashboard instead of looking for a page literally named **. And the headless renderer reports it when Chrome cannot start on our side, rather than quietly returning nothing and leaving you with "no usable tools".

Corrected an inaccurate claim: we previously advertised "90-day data retention" on Pro and "7-day" on Free, but no data was ever actually deleted on either plan. The date range picker on Pro now goes up to 1 year or all-time (bucketed by day, week, or month so it stays fast regardless of how much history a site has), and the pricing page reflects reality: unlimited analytics history on Pro.

Enterprise workspaces can now invite teammates from Dashboard → Team. Invite by email with an Admin or Member role, and everyone with access sees the same sites, tools, and AI generation quota, shared across the team instead of tied to one person's account.

You can now sign in with Google, alongside the existing email magic link and GitHub login. Until Google credentials are configured on our end, clicking it shows a short notice that it's coming soon instead of failing silently.

The dashboard, including the admin panel and every popup and modal, now works on phone-sized screens. This fixed a header overlap at narrow widths, several rows that didn't stack or wrap, tables that clipped without scrolling, and a real accessibility bug: the per-tool action buttons (test, edit, history, enable/disable, delete) relied on hover to appear, which meant they were invisible on touch devices. They're now always visible on mobile.

Our Privacy Policy previously stated tool call logs were purged after 90 days. That purge was never actually implemented, so we corrected the policy instead of quietly building a deletion job for it: logs are retained for as long as we operate the service, which is proportionate given arguments are already scrubbed of personal data and IP addresses are stored only as a one-way hash.

The WebMCP readiness checklist download now has a separate, unticked checkbox for occasional product-update emails, kept independent of the checklist itself. Ticking it sends a confirmation email; nothing further goes out until that link is clicked, and every future email includes a working unsubscribe link. We record the exact wording you agreed to as proof of consent.

We updated our Privacy Policy and Termsto describe, in plain language, how we use aggregated and de-identified data to improve crawling and AI generation quality, and to confirm that tool-call arguments are stripped of personal data before storage. We do not sell personal data, and anything we publish externally is aggregated so it doesn't identify you or your users.

You can now dismiss an AI-generated tool suggestion you don't want, straight from the generate panel, instead of leaving it sitting in the list.

On the privacy side: tool-call arguments recorded for your analytics are now scrubbed of obvious personal data (emails, phone numbers, card and account numbers) before they are stored. The useful part, such as a search query, is kept; anything that looks personal is replaced with a placeholder.

Tools whose executeJs uses await, for example a fetchcall, now run correctly instead of failing with an "await is only valid in async functions" error. The embed script was constructing tools as plain functions; it now constructs them as async functions, matching what the docs already described.

Connecting Shopify is now free, no card required. You get 5 storefront search-and-browse tools (searchProducts, getProductDetails, and similar) immediately after install. Upgrade for $19/month, 14-day free trial, to unlock the other 10, cart, checkout, discounts, live inventory, and order lookup.

A one-click Upgrade to Probutton is available from your site page and the Shopify onboarding screen whenever you're ready.

Pro access is now consistent no matter how you pay for it: an active Stripe subscription or an approved Shopify recurring charge both grant the same Pro features (full analytics, higher generation quota, unlimited sites/tools). Cancelling one no longer affects the other if you have both.

Also tightened billing enforcement on connected Shopify stores: cart, checkout, and Admin API tools now stay off until the Shopify billing charge is actually approved, instead of going live immediately after the OAuth install step.

Added an Enterprise plan for teams and agencies managing many sites: team accounts with roles, SSO/SAML, custom AI generation quotas, audit log export and compliance, and dedicated support with an SLA. Custom pricing, contact sales to get started.

Also fixed the pricing page, which listed Pro as "Custom, contact us" even though it's been a self-serve €19/month plan for a while. It now shows the real price and links straight to sign-up.

New Detect from schema.org button in the generate panel builds WebMCP tools directly from JSON-LD markup already on your pages, no LLM call, no AI generation used. Free on every plan, including Free-tier accounts with no generations left.

• Covers Product, Article, FAQPage, Recipe, Event, LocalBusiness, and JobPosting, the same set Google's Rich Results docs cover.
• Suggestions appear in their own "Auto-detected from schema.org" group, saved with the same one-click flow as AI suggestions.
• Skips anything that would duplicate a tool you already have.

See Generate Tools with AI for the full list of supported types.

A tool whose executeJs navigates the page (window.location, a form submit) can have its result lost if the page unloads before the call returns. Aigentably now does best-effort detection of this: interrupted calls show up as Interrupted in Recent Calls instead of silently vanishing.

• Analytics beacons now use navigator.sendBeacon, which is built to survive a page unload, instead of a plain fetch.
• The editor's live security warning for navigation now also catches .href, .assign(), .replace(), and form .submit(), not just bare window.location =.

See Write Tools by Hand for details.

Pro users who burn through their 20 monthly generations no longer have to wait for the reset: buy a top-up pack of 50 more generations for €9 right from the generate panel or Settings. Purchased generations expire with your next monthly reset.

Separately, the generator now knows about tools your site already has (saved or hand-written) and is instructed to propose new tools that fill gaps instead of re-suggesting near-duplicates on every regeneration.

See Generate Tools with AI for details.

Don't want to wait for the weekly automatic scan? The Agentic Browsing Score panel now has a Scan now button that runs a fresh Lighthouse check on demand.

• Limited to one manual scan per site every 5 minutes.
• Queues behind any scan already in progress, so it still never runs more than one Chrome instance at a time.

See Track Your Lighthouse Agentic Browsing Score for details.

Every site now gets a real Lighthouse Agentic Browsing scan roughly once a week, run as a background job so it never slows down your dashboard. See your score, a trend across recent scans, and a pass/fail breakdown for all six audits, no manual DevTools runs required.

• Covers all six real Lighthouse 13.4 audits: WebMCP tools registered, schema validity, form coverage, accessibility tree, Cumulative Layout Shift, and llms.txt.
• Runs sequentially, one Chrome instance at a time, so it doesn't compete with your app for resources.
• Self-hosted instances can wire it up with a single cron line. See the docs for setup.

See Track Your Lighthouse Agentic Browsing Score for details.

Every tool now keeps a full revision history. See exactly what changed between versions, mark a change as reviewed, and revert to any earlier version in one click, without losing the trail of what happened.

• Field-by-field diffs for name, description, page path, allowed origins, input schema, and executeJs.
• Reverting creates a new version instead of deleting history, so the audit trail never has gaps.
• "Needs review" flag on unreviewed changes, useful if more than one person has dashboard access.

See Write Tools by Hand for details.

Tools now support WebMCP's exposedTo option, so you can scope a sensitive tool, like applyCoupon or submitSupportTicket, to specific agent origins instead of leaving it open to every caller.

• New Allowed origins field on every tool, comma-separated, blank by default (open to all callers, unchanged from before).
• Enforced by the browser via WebMCP's native registerTool() options, not just in our dashboard.
• Scoped tools show an origin-count badge in the tool list.

See Write Tools by Hand for details.

Shopify merchants can now connect their store in under a minute via OAuth, without copying a script tag or editing a theme. Aigentably auto-seeds 15 WebMCP tools spanning the storefront and Admin API.

• OAuth install with encrypted token storage for the Admin API connection.
• Storefront: searchProducts, addToCart, applyDiscount, goToCheckout, and more.
• Admin API: live getInventory and getOrderStatus lookups with email verification.
• Theme App Embed (Online Store 2.0) with a vintage-theme fallback for stores that haven't migrated.

Read more in the Shopify + WebMCP guide.

AI tool generation now crawls up to 5 representative pages per site (homepage plus product, cart, checkout, search, account) instead of analyzing the homepage alone. Pages are discovered via sitemap.xml with fallback to homepage link extraction. Robots.txt is honored.

• Per-page LLM call with deduplication across pages.
• Each suggestion stamped with its source URL (visible in the dashboard).
• Crawl results cached for 24h per site. Force refresh button bypasses the cache.
• Crawl report panel shows which pages succeeded and how many tools came from each.
• Pro: Gemini 2.5 Pro. Free: Gemini 2.5 Flash.

Read the deep dive on the pipeline.