Skip to content
Start free

Changelog

What's new in Findlane, newest first. Follow along with the RSS feed.

  1. Findlane is in ChatGPT's plugin directory

    Add Findlane to ChatGPT from its plugin directory, then sign in and choose the workspaces ChatGPT can manage. Every tool that changes something is marked as a change, so ChatGPT asks you before it changes your search.

    AI agents
  2. Updates in the API, the server SDK and MCP

    What has come into an index, as its Updates page shows it, is now in the API, @findlane/server 0.39.0 and the MCP server, and every write's receipt says what it did.

    • Receipts count the records a write added, changed, left unchanged and removed, so a sync can tell at once whether it changed anything.
    • GET /writes and /writes/log, or index.writes.get() and .log(), with a key that has the analytics scope: for a sync's status in your own admin.
    • The MCP tool get_writes lets an agent check what an upload did, and why records were refused.
  3. See what's coming into an index

    A new Updates page under Records shows what has been sent to an index, who sent it, what it changed and what was refused.

    • When data last arrived and when it last changed anything, which aren't the same when a sync finds nothing new.
    • Records added, changed and removed by hour or day, for up to 90 days, by API key, product feed, Shopify, WooCommerce, the dashboard or a connected app.
    • Refused writes with the reason and an example object ID, so a bad record is easy to find.
    • Full replacements under way, with how far they've got.
  4. Map search: stats that skip values, and failed tiles that say so

    Fixes from building a large map on map search, in @findlane/browser 0.34.0 and @findlane/server 0.38.0.

    • A cluster's stat can be { field: "price", gt: 0 }, so a price of 0 that means no price doesn't make every area's range start at 0. Those records stay in the count and on the map.
    • In `useMapSearch`, a failed tile now keeps error set while other tiles show. The next move asks for it again, and retry() asks at once.
    • clusterFor finds a hit's cluster. A cluster's bounds are rounded, so comparing a place with them directly misses records on an edge.
    • tilesFor and useMapSearch read a map's own getBounds(), and tilesFor and clusterFor are on the script tag's Findlane global.
    • @findlane/browser can be loaded with require() on Node 22.12 and later.
    Maps
  5. Map search: clusters, spread-out pins and result bounds

    Searches can now draw a large map. Ask for fixed map tiles, which repeat across visitors where a box doesn't, so they're answered from the cache.

    • clusters groups a tile's matches by cell, with each group's count, average place, box to zoom to, and the min, max and average of up to three number fields.
    • spread makes the page each cell's best records rather than the top of one list, so pins spread over the map.
    • bounds gives the box around every match, for a map to fly to.
    • tilesFor says which tiles a view needs, and React's `useMapSearch` does the rest.
    • The homes playground shows all of it: bubbles with counts and price ranges, then each area's best home as you zoom in.
    • Tile searches show on the Traffic tab and count toward your plan, and search analytics leave them out.
    Maps
  6. Facet matches can filter the search, and show in React and the search box

    A search that names a brand can now go straight to it: with apply in the facet matches setting, diablo finds the brand Diablo's products, and diablo whey searches for whey among them. It's off until you choose it, per field, in the dashboard, and only searches that ask are filtered.

    • Only exact matches apply. A rule that replaced words wins, and a search that then finds nothing runs as typed. rewrittenQuery says what was searched, and compare_settings shows what would change before you save.
    • In React (browser SDK 0.32.0), SearchBox and useAutocomplete take facetMatches for rows above the results. On a results page, SearchProvider takes it, FacetMatches offers the values, and CurrentRefinements shows an applied one as a chip to remove.
    • The search box widget 1.5.0 shows them with data-facet-matches. The server SDK 0.36.0 has the types.
    Applying a match
  7. Facet matches in the dashboard, and the filters shoppers open

    Set up facet matches under **Search settings → Filters and facets**. Turn a facet on and its names are found in your records. Then go through its values, hide the ones not worth offering, and try a search right there.

    • **Filters opened from search** on the Analytics page shows which matches shoppers pick, and from which searches. Send client.events.facetMatch(match, { query, position }) when one is picked, in the browser SDK 0.31.0 and the server SDK 0.35.0.
    • The search playground shows a search's matches above its results. Click one to filter by it.
    • On the MCP server, get_search_analytics lists the matches picked.
    Facet matches
  8. Findlane Search for WooCommerce is on WordPress.org

    The WooCommerce plugin has passed the directory's review and is out of beta. In WordPress, search for it under Plugins → Add Plugin, and updates arrive like any other plugin's.

    • Translations now come from WordPress.org instead of files in the plugin. A language whose translation isn't ready shows English, as Icelandic does for now.
    Set it up
  9. Facet matches: the categories, brands and places a search names

    A search can ask which facet values its query names, such as the category Prótein for prót, a brand, or the zips of a town, so your search box can offer them beside the products. Each comes with its name, its products, how it matched, a filter and a link. Searches that don't ask don't change.

    • Set it up with facetMatches in an index's settings: names come from your records beside each value (a category's name beside its slug), from names you write, such as a town for each zip, or from the values themselves. Hide the values that aren't worth offering.
    • Ask with facetMatches: true in a search; matches come back in facetMatches, best first. Typing, typos, synonyms and plurals count, and each match says whether it took the whole query.
    • On the MCP server, get_facet_matches lists a field's values with their names and counts, and update_settings hides them one by one.
    • In the browser SDK 0.30.0 and the server SDK 0.34.0.
    Facet matches
  10. Admins, and local development from trusted addresses

    Workspaces have a third role: **Admins** change everything about search, as owners do, without billing. Under **Settings → Members**, invite someone as an admin, or change a member's role from their row's menu.

    • Admins change indexes, records, settings, API keys, allowed origins, trusted addresses, feeds, merchandising, alerts and shared reports, and invite and remove members and admins. Owners alone manage billing, the plan, the spending limit and the workspace's name, and make someone an owner.
    • Admins get the alert and feed emails owners get, and their agents on the MCP server can change search too.
    • From a trusted address, pages on localhost, 127.0.0.1 or your local network now search every index in the workspace without being added to its allowed origins, so you can build your site against its live indexes.
    Workspace members
  11. Optimized images come from images.findlane.dev

    Optimized images now load from images.findlane.dev, the host your uploaded banners already use, so your searches and your photos don't share one connection. Search answers name the address in x-findlane-image-origin, so it can change later without an SDK update.

    • The browser SDK 0.28.0 and the search box script 1.4.0 load images from it. They remember it in the browser, so later pages open its connection with the first search, before the photos ask.
    • If your site sends a Content-Security-Policy, allow https://images.findlane.dev in img-src. Pages on earlier versions keep loading from api.findlane.dev, which keeps working.
    • The Shopify app's search box is updated.
    Content Security Policy
  12. Optimized images on every paid plan

    Findlane now resizes your results' photos to the size each page shows them at and sends them as WebP, over the connection searches already use, and every hit carries a soft, blurred preview of its photo until the photo loads. On the first stores to use it, the resized photos were 88% smaller than the originals.

    • Turn it on under an index's **Access** page: choose the hosts your records' images are on, and the field previews are made from. Agents use set_image_hosts, and servers the API with a write key.
    • The React Image component and the search box script load optimized images by themselves. Your own markup uses client.imageUrl and the hit's placeholder, with the browser SDK 0.27.1.
    • Only images in the index's records are resized. Each paid plan includes 5 new copies a month for each record it includes, and copies already made load free; Settings shows the month's copies.
    Optimized images
  13. A/B tests count each order once

    A store that sends an order from the shopper's browser and again from its server had it counted twice in a test's orders and revenue. Now the test counts the server's, when it sends them with the shopper's group, and otherwise the browser's.

    • The start dialog works out how many more searches a test makes miss the cache from the index's own last 7 days, and says about how many a day: the more of a store's searches its cache answers, the more a split costs.
    A/B test results
  14. Findlane is in Claude's connector directory

    Add Findlane to Claude from its connector directory, then sign in and choose the workspaces Claude can manage. Every tool that changes something is marked as a change, so Claude asks you before it changes your search.

    • preview_query_suggestions only reads now, and the new rebuild_query_suggestions builds an index's suggestions again.
    • A new Support page says how to reach us, and the Privacy Policy says what a connected AI app keeps.
    AI agents
  15. A/B tests of search settings

    Test a change to an index's search settings on some shoppers before everyone gets it. Choose **Test this change** instead of saving: a share of shoppers search with the change, the rest with today's settings, and the Search settings page compares their clicks, add-to-carts, orders and revenue per search, then you keep the change or drop it.

    • Ranking, rank first, typo tolerance, synonyms, rules and search field weights can be tested. Each shopper stays in one group for the whole test, and nothing about them is stored.
    • Each figure shows the range it likely lies in, and a result is called only at the end, once each group has 1,000 searches. An A/A test shows how much the numbers move by chance.
    • A test ends at its date, or early if its change makes searches find nothing far more often, and owners get an alert. Agents use start_ab_test, get_ab_test and end_ab_test, and servers read results from /analytics/ab-test.
    • The browser SDK 0.26.0 carries the shopper's group from search answers onto events. The server SDK 0.32.0 sends it with your server's own orders, and reads tests and their results.
    A/B tests
  16. Agents can change API keys

    The MCP server's new update_api_key changes a key's name, scopes, the index it's limited to, or its searches a minute, as the dashboard's Edit key does. The secret stays the same, so servers using the key keep working, and new access applies within a minute.

    AI agents
  17. Workspace settings at a glance

    The workspace Settings page now fits on about one screen. Its settings are grouped into Workspace, Plan and usage, and Reports and alerts, and each shows what it's set to on one line, such as Email and Slack: 2 of 3 alerts, and opens to change it. The settings themselves work as before.

    • Members are near the top, under Workspace, with trusted addresses.
    • Each limit on Plan and usage keeps its bar on its line, and opens to what it counts.
    • Links to a setting, such as /settings#trusted-addresses, open it.
  18. A monthly demand report

    On the 1st, buyers and category managers get what shoppers looked for last month that the store doesn't sell: searches that found nothing, checked against the products, grouped when they mean the same, with the month before.

    • Also searches the store sells under another word whose synonym isn't added yet, and the month's rising searches.
    • By email to members with the weekly report, and to shared reports' addresses, whose page shows last month's too.
    • The first, November's, goes out on 1 December.
    Monthly demand report
  19. Alerts by email, Slack and Teams

    Findlane now tells you when something changes in search: more searches finding nothing, a search shooting up, or revenue from search falling. Owners get an email, and alerts can go to Slack or Microsoft Teams too.

    • Checked each morning for every index with enough searches, at most once a day each; revenue on Mondays, against the week before.
    • Under Settings, Alerts: choose which, turn the email off, connect Slack or Teams with a webhook URL, and send a test.
    • Webhook URLs are stored encrypted and never shown again.
    Alerts
  20. Brands' share of search

    See how much of search goes to each brand: its clicks on search results, and its products' add-to-carts, purchases and revenue found through search, with the searches naming it. For supplier talks and co-marketing.

    • On the Analytics page under Revenue, beside sales by filter, with a CSV, and as a spreadsheet on shared reports.
    • Brands come from the field your records name them in, found by its name or picked from your fields.
    • The analytics API's /analytics/brands gives the same numbers.
    Brands
  21. Search terms for ads and landing pages

    Download what shoppers search for as spreadsheets: the search terms that sell, with clicks, add-to-carts, orders and revenue, for ad keywords; and the searches that found nothing, with why, for landing pages and buying.

    • On the Analytics page beside revenue from search and searches with no results, and on shared reports, so marketing gets them without signing in.
    • Searches that found nothing are marked Not sold here or Sold under another word once they've been checked. Searches still being typed and junk are left out.
    • The analytics API's /analytics/search-terms gives both lists as JSON, or one as CSV.
    Search terms
  22. The weekly report, store by store

    A workspace with several indexes now gets a section for each store in its Monday report, the busiest first, instead of lists that mixed every store together.

    • Each store's searches and the share that found nothing against the week before, clicks on results and the share in the top 3, and revenue from search.
    • Its own top searches, searches without results and rising searches. Stores with fewer than 100 searches in the week are named in a line.
    • Searches without results leave out keystrokes on the way to a search that found something, so the list holds what's worth a synonym.
    Weekly report
  23. Monthly ceilings on events, record changes and AI

    Each plan has monthly ceilings far above what stores use, so a script sending events or rewriting records in a loop can't run up costs: from 100,000 events and 300,000 record changes on Free to 50 million and 30 million on Scale, and an AI allowance from $1 to $15.

    • Past the events ceiling, new events aren't stored until the next month; past the record changes, writes wait. Searches keep working. Owners get an email at 80%.
    • The AI allowance pays for reviews of searches that found nothing and field suggestions; past it, they wait for the next month.
    • Settings shows each beside the records and searches. The busiest store sent about 280,000 events in October, so this changes nothing for anyone today.
    Limits
  24. Plans include data, and records up to 100 KB

    Each plan now includes data as well as records: 4 KB of JSON for every record it includes, 40 MB on Free up to 4 GB on Scale. Records can be any size up to 100 KB, as long as the workspace's data fits.

    • Free stops at its data as it stops at its records. On paid plans, data past the plan's counts each 4 KB as an extra record, billed at the usual $1 per 10,000.
    • Settings shows the data next to the records. Today's catalogs average 1–3 KB a record, so this changes no one's bill.
    • The largest record goes from 1 MB to 100 KB, Algolia's limit. Searched text already stopped at 16 KB a field, and the largest record we hold is 17 KB.
    Limits
  25. Learn from clicks

    Search can now learn what shoppers pick. Turn on Learn from clicks under Search settings, and each search with 10 or more clicks in the last 30 days puts its most clicked and bought products first, updated daily.

    • Up to 5 products a search, each with at least 3 of its clicks. Recent clicks count more, and with visitor IDs a sale credited to the search counts as several clicks.
    • Learned products go first within their rank first group, so in-stock products stay ahead, and after products that rules pin or boost.
    • The settings list each search's learned products with their clicks and sales; leave out a search or a product with one click.
    • 1 in 10 shoppers each day search without it, so the settings compare clicks, add-to-carts and purchases per search for both groups.
    • In the API and the server SDK (0.31.0) it's learnedRanking: true, with learnedExclusions. The MCP server's update_settings takes both, and get_learned_ranking lists what was learned.
    Learn from clicks
  26. Shared reports, and rising searches

    Owners can share a read-only report of what shoppers search for with people who don't sign in, such as marketing: a page at a secret link with top and rising searches, searches that found nothing and why, empty category pages, and sales found through search. Add email addresses and they get the weekly report on Mondays too.

    • Rising searches on the Analytics page, in the weekly report and on shared reports: searches that rose most against the period before, for its traffic.
    • Searches that found nothing are marked Not sold here or Sold as another word on shared reports, once they've been checked.
    • New link stops a shared report's old link; each address is told who added it, and every email can unsubscribe.
    • get_search_analytics on the MCP server returns rising searches.
    Shared reports
  27. A status page

    findlane.dev/status shows whether search, the API, index updates, the dashboard, the SDK CDN and the website are working. Each is checked every five minutes, with each day's results kept for 90 days and outages listed.

    • It's served apart from the API, so it still loads while the API is down.
    • An outage is two failed checks in a row; a single failed check shows in that day's bar.
    Status
  28. Icelandic word forms

    Icelandic stores can now match every case and plural of their words. Turn on Icelandic word forms under Search settings → Languages, and a search word your products don't have finds its other forms: kjólar finds Kjóll, jokerinn finds Joker, and ömmu finds Amma.

    • Nouns and adjectives from BÍN, the Árni Magnússon Institute's database of Icelandic word forms.
    • Searches that find results with their words as typed don't change, and a search that finds nothing tries every form of its words.
    • It applies within about a second, without a rebuild. In the API and the server SDK (0.30.0) it's inflections: true, and the MCP server's update_settings takes it too.
    Icelandic word forms
  29. Signed keys on the compatible API, and in Traffic

    Sites that search through the compatible API can give logged-in visitors signed keys too, made the way their servers already make secured keys, with a Findlane signing key as the parent. And the Traffic tab now shows signed keys' searches as their own kind.

    • A secured key's filters are forced onto every search, facet counts included, and its other search parameters replace the page's. The index needs no allowed sites.
    • Traffic → Who searched has a Signed-in visitors row; refused keys, such as expired ones, are a reason of their own; and a heaviest visitor a key named is marked Signed in.
    • In the MCP server: list_signing_keys and revoke_signing_key. Signing keys are still made only in the dashboard, and shown once.
    Secured API keys
  30. Trusted addresses, and unblocking heavy visitors

    Add your own addresses, such as your office or your servers, under Settings → Trusted addresses, and browser searches from them skip the per-visitor rules on every index: heavy visitors, visitor and script limits, the data-center rule and the 1,000-deep limit. And when someone lands on the heavy-visitor list, an owner can let them back in from Analytics → Traffic.

    • Trust an address or a range, up to 20; Trust my address adds the one you're using. Changes apply within a minute.
    • Trusted addresses still count toward your plan and each index's limit, the crawler rules still apply, and each index's public settings still decide what they can read.
    • Heaviest visitors marks today's blocked and slowed visitors, and you; Unblock lets one search again until the day ends (UTC).
    • In the MCP server: set_trusted_addresses and unblock_visitor.
    Trusted addresses
  31. Signed keys for logged-in visitors

    Your server can now sign a short-lived key for each logged-in person, and their browser searches with it, straight to Findlane. The key says what they may search: filters forced onto every search, the fields they get, and who they are. Its limits replace the index's public settings, so a catalog behind a login needs no allowed sites.

    • Create a signing key under Search for logged-in visitors on the API keys page, and sign keys with signSearchKey in the server SDK (0.29.0).
    • Pass the key, or a function that fetches one, as signedKey to the browser SDK or React's FindlaneProvider (0.25.0). It asks for a new key before the old one expires.
    • With a userToken, rate limits and heavy visitors count the person rather than their address, so people sharing a mobile carrier's address don't share a limit.
    • Forced filters hold for facet counts and stats too, and keys last 7 days at most.
    Logged-in visitors
  32. A dashboard for listings, docs and articles, not just shops

    The dashboard now speaks the language of each index's kind of data: visitors and listings for listings, readers and pages or articles for docs and articles, and plain words for anything else. Shops read as before. An index that isn't a shop gets What search led to on the Analytics page, in place of revenue it doesn't have.

    • What search led to shows how many results were opened from search, the share of searches that led to a click, and the results opened most from search.
    • Clicks by filter shows which filter values, such as an area or a number of rooms, lead to clicks.
    • Revenue, sales by filter value, add-to-carts, orders and Bought together show for any index once it sends carts or orders, whatever its kind.
    • Choose an index's kind in its settings, under What's in this index.
    Analytics
  33. Browser search can return only what your pages show

    Under What browsers can see on the Public access page, Only some fields ticks the fields browser search returns, starting from the fields your pages asked for in the last week, or what a page of the index's kind usually shows. Searching, filters, facets and sorts keep following the index's settings, so nothing else needs listing.

    • Fields that look like they're for you, such as stock counts, costs, ranking signals and internal IDs, are marked, and so are searches whose pages named no fields, since they'd get fewer.
    • A field a page asks for that the policy doesn't return is now left out of the results, where it used to fail the search.
    • Turning browser search on asks first which fields browsers get. Indexes already public don't change until you choose.
    • Scripts and data-center networks, on indexes that limit them as they do by default, now get results up to 1,000 deep. None went further in the last 30 days, and shoppers aren't affected.
    • From a server, setPublicSearchPolicy({ returnFields }) does the same, and so does the MCP server's set_public_policy.
    What browser search can read
  34. Heaviest visitors on the Traffic tab

    Analytics → Traffic now lists the browser visitors who took the most results in a day, with their network, country, kind of traffic, searches, and how far they paged, so you can see when someone copies your catalog through search rather than shopping it.

    • A visitor is a key made from their address with that day's random value, deleted after the next day, so it can't be traced back to an address. Visitors get a new key each day and on each index.
    • The same list is in the MCP server's get_search_analytics.
    • It only shows what happened: nothing about how searches are answered has changed.
    Heaviest visitors
  35. Replacing catalogs of hundreds of thousands of records

    A staged replacement of more than 20,000 records now commits in steps, so a catalog of 200,000 records replaces in about a minute, where it used to fail. The commit still answers once it's done, with the same receipt, and smaller catalogs commit as before.

    • While a big one commits, other writes to the index wait for it, and record writes with keepNewer go ahead and keep their state.
    • Searches through the Algolia-compatible API use changed settings within seconds, where they could use the old ones for a minute.
    • An index's indexed revision no longer stays behind after a search server reconnects.
    Replace the whole catalog
  36. Search as you type, and Algolia's stricter typo modes

    With shortPrefixes on, a search's last word matches the start of longer words from its first letter, numbers too, so a search box that searches with each letter shows results from the first one: "sk" finds Skerplugata and "iphone 1" the iPhone 13. Without it, one- and two-letter words and numbers match only whole words.

    • typoTolerance takes "min", which keeps only the results with the fewest typos, and "strict", which keeps those and the ones with one more, ranked after them in any sort. Totals and facets count only what's kept.
    • Import from Algolia carries both over: Algolia's min and strict, and its prefixes from the first letter.
    • Both apply at once, without a rebuild, and indexes that don't turn them on search as before.
    Typo tolerance
  37. Servers can search through the Algolia-compatible API

    A server that searches with Algolia's client, such as a route that keeps the key out of the browser, can now point it at Findlane with a secret API key that has the search scope. Its indexes answer to their Algolia names whether or not public search is on, and replicas' names to their sorts.

    • Pass the shopper's address in X-Forwarded-For, and their User-Agent in findlane-visitor-agent, so each index's limits and bot rules apply to the shopper and verified crawlers search free.
    • filters without parentheses now read as Algolia reads them: OR binds first, so zip:101 OR zip:105 AND forSale:true keeps either postcode for sale. Before, it was refused.
    • A query that highlights nothing (attributesToHighlight: []) gets hits without _highlightResult, as from Algolia.
    Search from your server
  38. Cancelled orders are left out of revenue from search

    An order your server or the WooCommerce plugin cancels, for a cancellation or a refund, no longer counts in revenue from search or sales by filter, whenever the cancel comes. Analytics says how many orders were left out and their value.

    • Orders per 100 clicks shows once a query or filter value has 20 clicks, so a value clicked once and bought once doesn't top the list.
    • The API's /analytics/revenue and the MCP server's get_search_revenue give the cancelled orders as cancelled.
    Revenue from search
  39. Sales by filter: which filter values lead to sales

    Under revenue from search, Analytics shows each facet value shoppers filtered on, such as a brand or a price range, with the clicks on its results and the add-to-carts, orders and revenue that followed, on searches and on category pages apart.

    • Clicks carry their search's filters, from the search box, the WooCommerce plugin, React, and InstantSearch through the compatible API.
    • The API's /analytics/filter-sales, the server SDK's analytics.filterSales() (0.26.0) and the MCP server's get_sales_by_filter give the same numbers.
    Sales by filter value
  40. Searches opened by ads and links aren't counted as revenue from search

    When an ad, an email or another site links straight to a search page, the shopper didn't search, so what they buy after it now shows on its own line, by kind, instead of in revenue from search. Only the kind and the ad's network are kept.

    • The search box and the browser SDK (0.24.0) tell such pages by their address and referrer; client.events.landed(q) marks a page's search, and React's SearchProvider with routing does it for you.
    • WooCommerce plugin 0.2.3 links each paid order to the shopper's searches, so revenue from search counts WooCommerce orders, and adding a product to the cart from the results counts as a click on it.
    Searches opened from outside your site
  41. A request that searches several indexes counts as one search

    A multi-index search, and an InstantSearch request through the compatible API, now count as one search for each workspace they search, however many of their queries ask for results. The cache answering all of it still makes it free.

    Several indexes in one request
  42. WooCommerce: clicks on the search results page, and the store on the index's Overview

    With the WooCommerce plugin 0.2.2, clicks on products on the store's own search results page count in Analytics and revenue from search, as clicks in the search box do, when the shopper allows statistics.

    • An index connected to a WooCommerce store shows it on its Overview: the store, the plugin's version, when it last reported, and whether paid orders count, with a warning when it has gone quiet.
    WooCommerce
  43. Categories and pages in the search box: subindexes, and several indexes in one request

    An index can have subindexes, such as a store's categories and pages beside its products. A search asks for them with the products and counts as one search; the search box shows them above the products.

    • Up to 5 subindexes an index, not counted among the plan's indexes; their records share the index's records limit, and the index's keys open them.
    • A multi-index search runs up to 10 searches of separate indexes in one request, each counted: client.multiSearch() in the browser SDK (0.22.0).
    • React's SearchBox takes subindexes; the server SDK (0.25.0) has index().subindex() and the subindexes search option; the search box script on any site takes data-subindexes (1.1.0).
    • The WooCommerce plugin (0.2.0) keeps the store's categories and pages in subindexes and shows them as shoppers type.
    • A results page can suggest categories above its results: React's SearchProvider takes subindexes, and useSubindex reads them (browser SDK 0.23.0). The playground suggests the demo store's.
    Subindexes
  44. Findlane Search for WooCommerce, in beta

    A WordPress plugin connects a WooCommerce store in a click: Findlane reads its products and keeps them in step, its product search box shows products as shoppers type, and the theme's own search results page lists Findlane's results.

    • Changes reach Findlane as you make them, and a daily full read catches the rest, with the product feeds' safety check.
    • Paid orders count for Bought together and revenue from search, if you choose; never who bought them.
    • Feeds can now be read a page at a time from a JSON endpoint, and staged replacements can keep writes that arrive meanwhile (keepNewer).
    Set it up
  45. Product feeds: your store's catalog without code

    Connect the Google Shopping or Meta catalog feed your shop platform makes, or a JSON file, and Findlane reads it every hour and keeps your index in step. Only products that change are written.

    • A product's variants, the items sharing an item_group_id, become one record that still points to the variant a search matched.
    • A run that would empty the index, remove more than half its products, or refuse a fifth of its items is held, and the owners are emailed.
    • Check a feed before connecting it, and see why any item was refused, on the Records page.
    • The API, the server SDK 0.24.0 (index.feed) and the MCP server's feed tools do all of it from your own app.
    Product feeds
  46. Findlane search on any site, with one script tag

    The Shopify app's search widget now works on any site: add its script tag with your index's public ID, and your own search box shows products as visitors type, with a results view, filters and sorting.

    • It finds WordPress's search box and most others by itself, or the one data-input names.
    • It shows each record's title, image, price, sale price and brand, and links to its url.
    • No events are sent until your site says the visitor allows analytics, with data-consent="granted" or findlane.consent(true) from your cookie banner.
    Add search to any site
  47. Merchandising from your own app, and from AI agents

    Everything the Merchandising page does now has an API: which products come first on a category page or a search, hidden products, a banner and a campaign's times, banner images, and each banner's numbers.

    • The server SDK 0.23.0 has index.merchandising, index.images and analytics.banners.
    • The MCP server's new tools let an agent arrange pages, upload banner images from an address, and read each banner's click rate.
    Merchandising API
  48. A clearer Merchandising page

    The Merchandising page now opens on every arranged category page and search, with whether shoppers see it: live, scheduled, ended or off. Opening one shows your order first, then a line, then your store's ranking.

    • Each product has a menu with words: put first, move earlier or later, move to a place, take out of your order, hide.
    • Products found by name come from the page's category, and a product that won't show is marked on its card.
    • Unsaved changes show in a bar that stays in view, and saving says what shoppers see and from when.
    Merchandising
  49. Revenue from search for Shopify stores

    The Shopify app's checkout pixel sends completed checkouts and add-to-carts, with prices, so a Shopify store's Analytics shows what search sold. Shopify runs it only for shoppers who allow analytics.

    • Stores that installed the app before approve its two new permissions, customer events and pixels, the next time they open it.
    • With visitor IDs on, a purchase links to the same shopper's search clicks in the 24 hours before.
    Purchases and revenue from search
  50. Did you mean suggests only what looks like a typo

    Did you mean now corrects only what looks like a typing mistake: letters left out, two swapped, a doubled letter, or a key next to the right one. Words a store doesn't have, such as a brand it doesn't sell, no longer get an unrelated word that happens to be close in spelling.

    • On five live stores, wrong suggestions for their most common searches that found nothing fell from about 60 to about 8, while the real typos (jelyct → jellycat, womaizr → womanizer) are still corrected.
    • A word is no longer split into a stop word, such as "boxer" into "box er".
    Did you mean
  51. Why searches found nothing, and fixes

    On the Analytics page, with one index chosen, each search that found nothing now says why: a typo, another word for something you sell, something you don't sell, or not a real search. Typos and other words come with a fix, checked with a search, that you add with one click, or all at once.

    • A fix is a one-way synonym from the query to the word your records use, so only searches for that query change.
    • An AI model only chooses among words your index already has, its categories, brands and searches that found results, and nothing changes until you add a fix.
    • Agents read the same with the MCP server's get_no_result_fixes, and the fix-no-results prompt starts from it.
    Why, and fixes
  52. InstantSearch pages: optional filters, facet order and Insights events

    Stores that keep their InstantSearch pages lose less in the switch. Optional filters now boost and bury results as they did, DynamicWidgets show their facets, and click and conversion events from Algolia's insights library count as the index's events.

    • A search can send its own boost and bury, filters or a list of up to 10, each counting once, so records matching more come first. Optional filters use them.
    • Point the insights library at Findlane and its clicks keep their query and place, and add-to-carts and purchases their prices, for analytics and revenue from search.
    • Facet filters negated with a minus after the colon, such as brand:-Puma, now leave that value out, as they do in Algolia.
    • Browser SDK 0.20.0 and server SDK 0.21.0 have boost and bury, and responses' banners and redirect; React's Hits shows banners and SearchBox follows redirects.
    Click and conversion events
  53. Merchandising: arrange category pages and searches on a grid

    A new Merchandising page arranges a category page or a search on a grid of its results, with their images: drag a product to a place to pin it, hide products, add a banner, and give it start and end times. Saving keeps it as a rule that applies within seconds.

    • Rules can show a banner among the results, and send a search to a page, such as "returns" to your returns page.
    • React's Hits shows banners at their places and SearchBox follows redirects on Enter, as does the Shopify app's search once you update it.
    • The Algolia-compatible API returns banners in userData and redirects in renderingContent, and Import from Algolia carries them over.
    Merchandising
  54. Pin products at places

    A rule can now pin products at places, such as a new arrival first on the shoe category page or a campaign's product second for "black friday". Pins work on searches with words and without, and with every sort.

    • A pinned product shows when it passes the search's filters, even if the words didn't find it, so a category page never shows another category's product. It leaves its own place, and the result count includes it.
    • Find products to pin in the rule editor, under Pin products. Hits a rule pinned carry pinned: true.
    • Import from Algolia carries promoted records over as pins.
    • Browser SDK 0.19.0 and server SDK 0.20.0 have the types for this week's changes: context, rulesAt, rule conditions, times and pins, pinned hits, and the server SDK's analytics.revenue().
    What rules do
  55. Revenue from search

    The Analytics page now shows what shoppers bought of products they found through search: the value of those orders beside all orders, per query and per product. A product counts when the shopper clicked it in search results in the 24 hours before buying it.

    • Each query shows its searches, clicks, add-to-carts, orders and revenue, and each product its revenue from search and the query that found it most.
    • It needs visitor IDs on, search clicks, and purchases sent as events with prices. Each order counts once, even when your pages and your server both send it.
    • The weekly report sums it up, and the API's /analytics/revenue and get_search_revenue for AI agents give the same numbers.
    Revenue from search
  56. Rules for category pages, contexts and campaigns

    Rules can now merchandise a category page, not only searches with words. A rule fires on the search's filters, so { category: "Skór" } without words boosts, buries or hides products on the shoe page. It can also fire on a context your page sends, such as black-friday, and only between two times, so a campaign starts and ends by itself.

    • On pages without words, boosted products come first within rank first's groups, as in searches with words, so a sold-out product stays below what's in stock. Named sorts keep the order the shopper chose.
    • A rule with times changes cached results within about a minute of each, with no settings change. The settings page shows each rule as scheduled, active or ended, in your time zone.
    • Searches take context, a name or up to 10, and Import from Algolia carries over rule contexts, conditions on filters, rules for an empty query and time ranges. The Algolia-compatible API turns ruleContexts into contexts.
    Category pages and campaigns
  57. Variants: each product once, as the variant that matches

    A new Variants setting shows each product once, as the variant a search matched, whichever way your catalog keeps them. When colours or sizes are records of their own, records sharing a field such as productId show as one result: searching "blue trail shoe" shows each shoe as its blue record. When variants are listed inside each product, such as a supplement's flavours, each result says which one matched, so the page can show its image and price.

    • Result counts and paging count products, and totalRecords the matching records. Facet counts still count records.
    • Searches can send distinct: false to show every record, or a number from 1 to 4 to show that many of each product's records.
    • For variants listed inside each product, a hit's variant says which one matched, by the search's words in the variants' own fields or by filters on them. Facets then count products.
    • Import from Algolia carries attributeForDistinct over, and the Algolia-compatible API follows a search's distinct.
    • Browser SDK 0.18.0 and server SDK 0.19.0 have the types for distinct, totalRecords and each hit's group and variant. The server SDK also has the two settings.
    Variants
  58. Keep your InstantSearch pages

    A site whose search runs on InstantSearch or Autocomplete can now search Findlane without changing its pages. Findlane answers Algolia's search API: point the Algolia client your site already uses at api.findlane.dev, with your workspace ID and a search key.

    • Create the key on API keys, under Sites built on InstantSearch, and choose the indexes it opens. Each stays held to its public search settings and the sites it allows.
    • Indexes answer to their Algolia names: Import from Algolia keeps the index's and each replica's, as the sort it became, and you can add more, such as a Query Suggestions index.
    • Filters, facets, searchable facet lists, highlights, snippets, sorts and paging work as InstantSearch expects. Queries InstantSearch adds only to count a refined facet's values aren't counted as searches.
    Keep your InstantSearch pages
  59. Import from Algolia copies the records too

    Import from Algolia can now copy an index's records along with its settings. Your browser reads them from Algolia and sends them to the index as one replacement, so searches keep the current records until all have arrived.

    • It's chosen for a new index. On an index with records it's left unchosen, since the copied records replace them, and it can't be chosen when the records wouldn't fit your plan.
    • A record the index refuses, such as one with text in a number field, is left out and listed with the reason. Past 50 refusals the copy stops and nothing changes.
    • A staged batch the API refuses now names the record, as an upsert does: Record 3: ….
    Import from Algolia
  60. Search settings at a glance

    An index's Search settings page now groups its settings into Matching, Ranking and sorting, Filters and facets, Synonyms and rules, and Suggestions. Each setting shows what it's set to on one line, such as title ×8, brand ×4 or 3 rules, and opens to change it. The settings themselves, and saving them, work as before.

    • A setting with unsaved changes keeps its dot. When a change needs fixing before it can be saved, Review opens that setting.
    • In the Playground, a search that finds nothing lists its related results under their own heading, with their scores, when the index has them on.
    • Agents on the MCP server can follow related results with get_related_results, calibrate them again with calibrate_related_results, and see them in search.
    Search settings
  61. Related results when nothing matches

    When a search's words find nothing, it can now show the products closest to it in meaning, under a heading of their own. They catch what typo tolerance and synonyms can't: run-together words such as peanutjellycat, other words for the same thing, and searches in English for products described in Icelandic. Searches that find something don't change.

    • Turn it on in an index's settings, on paid plans, for indexes of up to 200,000 records. Every record is embedded within minutes, and new and changed ones within minutes of their write.
    • The floor, how close a product must be to show, is calibrated from the index's own searches that found nothing, so at most about 1 in 20 of them shows a poor first result. The settings page shows examples, and you can move it.
    • Related results come in the response's related, beside an empty answer, so analytics still count the search as one that found nothing. Browser SDK 0.17.0's React Hits shows them, and their clicks are counted apart. Server SDK 0.18.0 has the types.
    • What it costs is capped at 3% of your plan's price a month.
    Related results
  62. A spending limit on every paid plan

    Every paid workspace starts with a spending limit on extra searches: the difference to the next plan's price, which covers up to that plan's searches. Extra searches and records never cost more than a larger plan would. Before, each was capped at the difference to the next plan, and usage past the cap wasn't billed.

    • Starter's default, $30, covers 750,000 searches a month, Team's. Team's $100 covers 3 million, Growth's $150 covers 10 million, and Scale's $299 covers 11,495,000.
    • Owners change the limit under Settings, from $0 to no limit. Past what it covers, searches the cache can't answer pause until the next month, as before.
    • With the limit raised, a busy month costs what a larger plan would: Starter with a million searches costs $99, the same as Team.
    Pricing
  63. Words with apostrophes and periods

    Searches for words with an apostrophe, period, colon, or comma inside, such as don't, levi's, o'brien, 3.5mm, 1,000, or J.R.R., find them. Before, such a search could find nothing.

    • Either apostrophe works: levi's finds Levi’s, and levi’s, as phones type it, finds Levi's.
    • A word with an apostrophe also finds it written without one, ranked slightly below. Typed without its apostrophe, dont finds Don't through typo tolerance.
    • Highlights mark the whole word, and search analytics and query suggestions keep the apostrophe.
    Text normalization
  64. Faster results as you type

    Searches answered from the cache near your shoppers stay there longer and keep answering while a changed index catches up, and the React components no longer wait after each keystroke.

    • Popular searches and category pages keep coming from the cache after a catalog update: for a minute after a change, the previous results answer once while fresh ones load.
    • In browser SDK 0.16.0, the first keystroke searches at once, and earlier searches keep going, so their answers show while the next one loads. Only keys typed faster than 50 ms apart wait.
    • Focusing a search box opens the connection to the API, so the first search doesn't wait for it. client.warm() does the same for your own box.
    Search as you type
  65. Lists

    Rows of products for your pages, such as Popular, New arrivals or a holiday's products, and menus of a field's values with their counts. Set them up once, and pages ask for them by name. List requests aren't searches, so they never count toward your plan.

    • Each list is a filter and a sort, with products pinned first or hidden. The editor shows what it holds as you change it.
    • Lists are built ahead, within a minute of a change to the index, and served from a cache near your shoppers. Refresh now builds them at once.
    • A page asks for all its lists in one GET: client.lists(["popular", "themes"]) in browser SDK 0.15.0, or index.lists.getMany() in server SDK 0.17.0.
    • 3 lists per index on Free, 25 on paid plans. Agents set them up with save_list.
    Lists
  66. Where clicks land

    Analytics shows where shoppers click on searches that fit on one page apart from broader ones, and lists the searches where they click further down than the top 3.

    • On a search with one page of results, shoppers look for something specific, so the share of their clicks in the top 3 shows how well ranking works. Clicks on broader searches are counted by page.
    • Clicked further down lists those searches, each with the result shoppers click most below the top 3, and Try, which opens the search in the playground.
    • Top queries say where their clicks usually land: the middle click's place, which a few clicks far down don't pull up.
    • The analytics API's top queries add middlePosition, and agents get both lists from get_search_analytics.
    • The weekly email gives the share of clicks on searches with one page of results that were in the top 3.
  67. A million books in the playground

    A new playground searches the million works in Open Library with the most editions and readers, from your browser, with the time each search took beside the count.

    • Search titles, authors and subjects, typos welcome, and narrow by subject, language and when a book first came out.
    • Among equally good matches, the books more people read come first, from custom ranking. Sort by readers, editions or year instead.
    • Open Code to see the request, or Response to see what came back.
    Search a million books
  68. Events in Analytics

    The Analytics page has an Events tab beside Searches and Traffic: views, clicks, add-to-carts and orders across all your indexes, or one, with the same range, index and source filters.

    • A tile for each kind of event, with its change from the previous period. Click one to chart it, with a line per index.
    • Choose an index to see its products: the most viewed, added to carts and ordered, and those viewed but rarely bought.
    • Recent events from every index, and the events that weren't recorded and why.
    • Each index's Events page keeps its last 7 days, how to send events, and their settings.
    • Agents get the counts, and the latest events with what each carried, from get_event_analytics.
  69. Icons and colors for indexes

    Give each index an icon and a color, so you can tell them apart at a glance in the dashboard: in the sidebar, the index switcher, ⌘K, API keys and Analytics.

    • Click the icon next to an index's name to pick one of 40 icons and 8 colors. Indexes start with an icon for their kind of data, and each gets a color no other index has.
    • On Analytics, an index's line on the searches chart is its color, and its icon marks it in the legend, the recent searches and the tables.
    • The index's name at the top of the page now switches to another index, on the same tab.
    • Agents set them with set_index_appearance, or when they create_index.
  70. What searches with no results filtered on

    On the Analytics page, each search with no results opens to show what its searches filtered on and how they were sorted. That tells a product you don't sell apart from a filter that leaves nothing, such as an empty category.

    • Searches with filters and no query, such as a category page with no products, are listed too, as “No query, only filters”.
    • From 12 October 2026, searches from shoppers' browsers show their filter values, like categorySlugs: halloween. **Try** runs the search in the index's Playground, filters included.
    • Agents get the same details from get_search_analytics.
  71. A real estate playground

    A new playground searches 12,000 homes for sale in Los Angeles, Austin, Denver and Washington, DC, the way people search a real estate site: an address, a street, a neighborhood or a zip code, with price, bedroom and feature filters, sorting, and a map.

    • Type what you want, like “3 bed 2 bath austin under 600000”: rules with placeholders turn the words into filters, and the page shows what they read.
    • Street and neighborhood synonyms: “st” finds Street, “cap hill” finds Capitol Hill, “weho” West Hollywood.
    • Query suggestions as you type: “cap” suggests Capitol Hill and Cape Cod. With the box empty, it shows your recent searches.
    • The homes are made up on real streets, from the cities' open address and property data. Photos and descriptions are AI-generated.
    Search the homes
  72. Query suggestions

    As shoppers type, the search box suggests whole searches above the products, the way search engines do: “hrek” suggests “hrekkjavöku búningar”. With the box empty, it shows your store's popular searches and the shopper's recent ones.

    • Suggestions come from what your shoppers search, from your catalog's category, theme and brand names and words, and from searches you pin, add or hide. Typing on the way to a longer search, typos and long numbers don't become suggestions.
    • They're answered from memory near the shopper and never reach the database, so they're free on every plan and don't count as searches.
    • On by default. The index's Settings page shows what shoppers see, and the Searches tab which suggestions they pick.
    • Browser SDK 0.14.0 has querySuggestions, popularSearches and recentSearches, and React's SearchBox the querySuggestions, popularSearches and recentSearches options. Server SDK 0.16.0 has the same methods. The Shopify app shows them with its new theme settings.
    Query suggestions
  73. Preview settings, and search tests

    Before saving search settings, see what they'd change: your search tests that would start failing, and the results that change for your most searched queries.

    • A search test is a query and what its results must hold: a product in the top 3, a product kept out of the top 10, or at least some results. Save them in Search settings, or from a result in the Playground.
    • Preview changes runs your tests and your 100 most searched queries of the last 30 days with your settings and the new ones, side by side. Nothing is saved, and previews don't count as searches.
    • Agents compare with compare_settings before update_settings, and save tests with set_search_tests. The server SDK has compareSettings (0.15.0).
    Preview changes
  74. Placeholders in rules

    One rule can now cover every number or every value of a facet: "kerti {number} ára" filters number candles to the number in the query, and "{Thema_name}" boosts whichever theme a query names.

    • {number} takes a number from the query; a facet field's placeholder, such as {brand}, takes words that are one of its values, as the field writes it.
    • Filters, boosts, buries, replaced words and userData can use the value. A number field gets it as a number.
    • Preview changes shows what a placeholder rule changes before you save it.
    Placeholders
  75. Analytics API

    Show an index's analytics in your own admin: searches, searches without results, top queries with their clicks, events, and each product's views, add-to-carts and orders, through the API or the server SDK.

    • Search and event reports cover the last 24 hours to 90 days, or any dates in that time, in UTC, by hour or day. Product counts are exact, from every event kept for 13 months.
    • One product's report is fast enough for a product page: its views per day, the searches it was clicked from, and its orders.
    • A new analytics scope reads them, and no other scope does. A key limited to an index reads only that index's, and only from a server.
    • Server SDK 0.15.0 has index.analytics.searches(), queries(), filters(), events(), products(), productCounts() and product(objectID). Free on every plan, and not counted as searches.
    Analytics API
  76. Compound words

    With compoundWords on, searches find compound words whether they're typed together or apart: "hár kolla" finds wigs ("Hárkolla"), and "face paint" finds "facepaint".

    • Neighbouring query words also match the word they make, and a query word that's a compound in your records also matches its parts written apart.
    • The compounds come from the words in your records, in every language. Exact matches still rank first.
    • Off by default; turn it on in an index's Search settings, or with compoundWords: true (server SDK 0.14.0 has the type). The Playground's Why this result? marks compound matches.
    Compound words
  77. Rules

    A rule changes the results of searches whose query has one of its phrases: boost or bury records, filter them, hide some, change the query's words, or return data for a banner.

    • "blöðrur" can put the balloons category first, and "kerti 5 ára" can search for candles among the number 5s. Phrases match as synonyms do, with plurals in the index's languages.
    • Boost is the safe default: a filter leaves out every record without the field. The rule editor says how many results a filter would keep before you save it.
    • Rules apply within a second, without a rebuild. The Search settings page shows how many searches each rule fires on, and the Playground can search with rules off. Agents edit them with update_settings and check them with get_rule_hits.
    • Searches return appliedRules, with each rule's userData, and rewrittenQuery when a rule changed the words. Server SDK 0.13.0 and browser SDK 0.13.0 have the types.
    Rules
  78. Why each result matched

    explain: true says, for each result, how each query word matched in each field, its relevance band, and the values that ordered it.

    • Each match says whether it was exact, a prefix, a typo (and how many), a synonym or a plural, with the weight it had in the search.
    • For API keys, AI agents, and the Playground's new Why this result? tab. Browser search doesn't take it.
    Explain results
  79. Rank first, typo controls, one-way synonyms, and more filters

    Settings for what shops kept working around: in stock first everywhere, finer typo tolerance, synonyms that work one way, and filters with not, all, and or.

    • Rank first orders every result by up to two fields before relevance, such as in stock before sold out, in text searches, browsing and sorts.
    • Equally relevant within a percent you choose (10 by default), so custom ranking such as popularity can outweigh small differences.
    • Typo word lengths, typos off for a field such as SKUs, and exact words such as brand names. Prefixes still match.
    • One-way synonyms: "öskudagur" finds costumes, but "costume" doesn't find Ash Wednesday.
    • Filters take not, notIn, all for lists, several conditions on one field, and $or groups.
    • The Playground suggests a filter field's values, and when a filtered search finds nothing, names the field that has the value.
    Rank first
  80. Recommendations follow product views

    For a catalog without a popularity field, the products that get recommendations first, and the ones the add-on covers, now follow views as well as orders.

    • Orders in the last year still come first. Views in the last 30 days order the products with as many orders, which is most of a catalog.
    • Views are read once a day from the events your pages send.
    Turn them on
  81. Products on the Events page

    Each index's Events page lists its products with exact views, viewers, clicks from search, add-to-carts and orders, and the products people look at but rarely buy.

    • The most viewed, most added to carts and most ordered products in the page's range, sortable by any column, with add-to-carts and orders per 100 views.
    • Viewed, not bought: products with 50 or more views and the fewest orders per view.
    • Counted from every event kept, about two minutes behind. Agents get the same numbers from get_product_analytics, and the weekly report names the three most viewed products.
    Products
  82. Visitor IDs, and 13 months of events

    Events can link one shopper's views, carts and orders with random IDs, when a store turns them on, and every event is now kept 13 months, so a season can be compared with the last.

    • Turn on Link each shopper's events on an index's Events page, and visitorIds: true in the browser SDK (0.12.0). A server can send its own anonymous IDs (server SDK 0.12.0).
    • Findlane keeps a keyed hash of each ID, different for each index, never the ID itself. Getting consent is up to the store; the Shopify app sends IDs only while the shopper allows analytics.
    • client.events.forget(), or index.events.forget(ids) from a server, deletes the events sent with a shopper's IDs within a week.
    • Every accepted event is kept, unsampled, for 13 months, including a server's past orders. The Events page still shows the last three months.
    Visitor IDs
  83. Product views in one line

    Product pages can send their views with one line, or none, so each index's Events page sees which products people look at.

    • React: useProductView(objectID) on a product page, or <ProductView objectID={…} /> in a server component (browser SDK 0.11.0).
    • A repeat view of the same product within 5 seconds isn't sent, so re-renders and React's double mount in development count once.
    • With its next update, the Shopify app sends product page views from the search app embed, when the shopper allows analytics. Its clicks and views go by fetch, which ad blockers let through, instead of beacons.
    Product views
  84. Synonyms work from phrases

    A synonym group with a phrase now works both ways. With ["face paint", "andlitsmálning"], searching for face paint finds andlitsmálning, not only the other way round.

    • The phrase's words count when typed together and in order, among other words too, such as gold cake topper.
    • With languages set, plurals count as well, so face paints finds it.
    Synonyms
  85. Events through your own domain

    Most ad blockers block beacons sent to other sites, so their users' events were lost. Send events through a route on your own site instead, set up in one line.

    • Browser SDK 0.10.0 has an eventsUrl option, such as "/findlane/events".
    • Server SDK 0.11.0 has index.events.proxy(request) for that route in Next.js, Cloudflare Workers, Hono, Bun and Deno, and events.relay() for Express.
    • They're still the visitor's events: the rules for browsers apply, and limits count each visitor.
    Through your own domain
  86. Clicks

    See which results shoppers click: each top query's click-through, the place of clicked results, and searches whose results nobody clicked.

    • React's SearchBox, SearchDialog, Hits and Recommendations send clicks for you (browser SDK 0.9.0). Recommendations gains getHref and hrefTemplate for product links.
    • The Shopify app sends them from the search box, the results view and the Recommendations block, when the shopper allows analytics.
    • Each index's Recommendations page shows the clicks on each list, and the weekly report counts the week's clicks.
    • A catalog without a popularity field takes it from orders, so recommendations are built for what sells first.
    Clicks
  87. Bought together

    Recommendations gain a third list: products bought in the same orders, from the orders your store sends.

    • Each product's list holds the products in the most orders with it over the last year, once at least 2 orders had both.
    • Your few best sellers can't fill every list: past 2 of them, a list shows its other picks first.
    • React's <Recommendations list="together" fill="goesWith" /> tops up short lists from Goes with, and says "Frequently bought together" when most of the list comes from orders (browser SDK 0.9.0).
    • The Recommendations page shows how many products have a list, each pick's orders, and lets you hide a pick.
    • On Shopify, choose Use my orders in the Findlane app: it reads your orders from the last 60 days, then new ones as they're placed, and the Recommendations block gains a Frequently bought together list. See Shopify.
    • Try it in the store playground: open any product. The demo store has no shoppers, so its orders are simulated.
    Bought together
  88. Feedback from the dashboard

    Report a bug, ask a question or suggest an idea from any page of the dashboard, and get a reply by email.

    • Feedback is at the bottom of the sidebar, next to a link to the docs.
    • Each message includes the page you sent it from, your workspace and your browser, so we can look into a bug without asking first.
  89. Events

    Send views, clicks, add-to-carts and orders from your pages and your server, and watch them arrive on each index's new Events page.

    • Browser SDK 0.8.0 has client.events, sent in the background, with events: "afterConsent" to hold them until a shopper agrees.
    • Server SDK 0.9.0 has index.events.send(), importOrders() for up to 400 days of past orders, and cancelOrder(). API keys have a new Events scope.
    • Orders are kept for Bought together, a list of products bought in the same orders, coming next to Recommendations.
    • Events carry no cookie, IP address or customer details, and don't count as searches.
    Events
  90. Limits and bots, per index and per key

    Choose how fast each index answers visitors, crawlers and scripts, and see who searched on the Analytics page.

    • Each index's Public access page has Limits and bots: searches a minute per visitor and for the whole index, and whether crawlers, scripts and data-center networks are limited, blocked, or treated as visitors.
    • Crawlers get cached results first, and a limit on the rest. Searches from verified crawlers, such as Googlebot, stay free.
    • Each API key has a searches-a-minute limit on the API keys page, in case it leaks.
    • Analytics has a Traffic section: what counted toward your plan, what was free, what was refused, and why. The weekly report says who searched.
    • Server SDK 0.8.0 can search for a visitor, { visitor: { ip, userAgent } }, so server-rendered search pages get the same limits, and Googlebot's searches are free there too.
    Searching for a visitor
  91. A spending limit, and free crawler searches

    Choose the most a paid workspace spends on extra searches in a month, and stop paying for search engines' crawlers.

    • Owners set the limit under Settings, from $0 up to the plan's overage cap, which stays the default. It applies at once.
    • Searches our cache can answer keep working after a workspace's searches pause, on every plan.
    • Browser searches from verified crawlers, such as Googlebot, are free and never paused, so your pages stay indexed with their results.
    • Owners get an email when extra searches begin, at 80% of a spending limit, and when searches suddenly jump.
    Limits and billing
  92. Recommendations

    A new add-on gives every product page two lists: products that go with it, and similar products to choose instead. They're built from your catalog alone.

    • Lists are built in the background, the most popular products first, with no orders or tracking needed.
    • Turn them on from an index's Recommendations tab, after a free preview of 20 products. The add-on is priced per 1,000 products, from $6 a month.
    • Show them with client.recommendations(), index.recommendations(), or React's <Recommendations> (SDK 0.7.0), or with the Recommendations block on Shopify product pages.
    • The price limit, variants, and the category map are yours to change, and settings apply at once.
    • Try them in the store playground: open any product to see what goes with it.
    Recommendations
  93. Fewer requests from the browser

    With browser SDK 0.6.0, a visitor's first search starts a round trip sooner, and pages send fewer requests.

    • Searches go out without a CORS preflight, so the first one no longer waits for it. Older SDK versions keep working, and browsers now remember their preflight for 2 hours instead of 10 minutes.
    • Components that ask for the same search at once share one request.
    • The Shopify app's results view gets its filter counts in one request, instead of one more for each filtered field.
    Browser SDK
  94. Agents across your workspaces

    When you connect an AI agent, choose all your workspaces or only some, instead of just one.

    • All workspaces includes the ones you create or join later.
    • Agents call list_workspaces and name the workspace they work in. In each one, they have your role.
    • Disconnecting an agent on a workspace's API keys page removes only that workspace.
    Connect an agent
  95. A pricing page, and this changelog

    Every plan's limits, the overage caps, and billing answers on one page, and a changelog with an RSS feed.

    See pricing
  96. Use cases and data recipes

    Pages for each kind of data Findlane searches, and a Your data section in the docs with the records and settings to start from.

    • Use cases for online stores, docs and help centers, articles and research, and listings and places, each with a live demo.
    • Plan your records, then a recipe for products, docs, articles, or listings.
    • The MCP server's set-up prompt reads the recipe for your kind of data before it chooses settings.
    See the use cases
  97. Location search

    Give records a location, then search near a point, within a radius, or inside a map's view.

    • A locationField setting, and near, within, and sort: "distance" in searches. Each hit says how far away it is, in meters.
    • near: { visitor: true } starts near the visitor's city, with no permission prompt.
    • In React, useGeoSearch keeps a map, its results, and the URL in step (@findlane/browser 0.5.0).
    • Try it on 35,000 restaurants.
    Read about location search
  98. An MCP server for AI agents

    Connect Claude Code, Cursor, VS Code, or Claude to Findlane, and an agent can set up and look after your search.

    • Sign in once in your browser, and choose the workspace the agent may manage.
    • 21 tools: indexes, records, settings, test searches, allowed origins, analytics, API keys, and the docs.
    • Searches agents run to check their work are free.
    Connect an agent
  99. Did you mean

    Searches say when typo tolerance changed the query, and a search that finds nothing can ask for a correction.

    • Turn on didYouMean in an index's settings.
    • suggest() returns a query that finds results, and React's DidYouMean shows it.
    • Suggestions don't count toward your plan's searches.
    Read about did you mean
  100. React components

    @findlane/browser/react has a search box, a ⌘K search dialog, a results page with filters, and the hooks behind them.

    • Facet checkboxes keep the other values listed after one is ticked, with disjunctiveFacets.
    • Sorts can settle ties with up to two more fields, with thenBy.
    • Findlane's own docs search with SearchDialog.
    Read the React docs
  101. Price sliders and more facets

    facetStats returns the smallest and largest values of number fields, such as the ends of a price slider.

    • An index can have up to 128 filter, facet, and sort fields, and a search can ask for 32 facets.
    Read about number ranges
  102. Where people search

    Analytics shows where searches come from on a map, with top cities, devices, browser languages, and a feed of recent searches.

  103. Languages

    A languages setting for Danish, English, German, Icelandic, Norwegian, and Swedish: searches leave out common words, and match plurals and other spellings.

    • New indexes also match letters such as ø, ß, æ, ł, and þ to their plain spellings, so kobenhavn finds København.
    • Icelandic doesn't match plurals yet.
    Read about languages
  104. Findlane for Shopify

    An app that keeps a store's products in sync and answers the theme's own search box, without code. It's on its way to the Shopify App Store.

    Read about the app
  105. Findlane is live

    Hosted search with an HTTP API, server and browser SDKs, a dashboard, and a playground with a live store catalog.

    • Typo tolerance, facets, filters, sorts, and synonyms.
    • Search straight from the browser with an index's public ID and the origins you allow.
    • The browser SDK on a CDN, for a script tag.
    • Search analytics, and invitations for teammates.
    Start with the quick start