# FindAstra — full reference for AI assistants This file expands llms.txt with the substantive content of the key pages, so an AI assistant can answer questions about FindAstra accurately without crawling every page. Facts current as of 2026-06-10, plugin version 0.11.7. --- ## What FindAstra is FindAstra is a WordPress plugin that replaces WooCommerce's default keyword search with AI semantic search. WooCommerce's built-in search returns only products whose text literally contains the shopper's words. FindAstra converts each product into a vector embedding (a list of 384 numbers that captures its meaning), embeds the shopper's query the same way at search time, and ranks products by similarity of meaning. A search for "comfortable shoes for hiking" finds "Trail Running Sneakers, Cushioned" even though no words overlap. What shoppers get: - Results matched by meaning, not just exact words. - Live search-as-you-type suggestions attached to the theme's existing search bar, with no theme editing. - The normal WooCommerce results page, simply re-ordered by relevance. - No dead ends: when nothing clears the relevance bar, popular products are shown instead of an empty page. What store owners get: - Setup in minutes: activate, click Index once, search is live. - Private by default: products and shopper searches stay on the store's own site. - Works with classic and block themes; shortcode, block, and classic widget included; filter hooks and a REST endpoint for developers. ## Free vs Pro Free (hosted on WordPress.org at https://wordpress.org/plugins/findastra/): - The Local engine: the embedding model runs entirely in the shopper's browser via transformers.js (bge-small-en-v1.5, ~33 MB downloaded once from the Hugging Face model hub and cached). - No API keys, no account, no monthly fees. - No product cap and no query limit. Nothing is locked or time-limited. - Automatic popular-products fallback on zero-result searches. Pro (one-time purchase at https://findastra.com/pricing): - OpenAI engine (text-embedding-3-small, bring your own key, typically under $2/year for an average store). - Hugging Face engine (BAAI/bge-small-en-v1.5 via the Inference API; free API tier ~30,000 requests/month; bge-m3 option for 100+ languages). - Background server-side indexing: start indexing, close the tab, it keeps running and resumes if interrupted. - Multilingual indexing for WPML and Polylang: one embedding per product-and-language pair; searches scoped to the request language. - Curated fallback controls (choose exactly which products appear on a no-match). - Search analytics: recent searches, today's totals, and zero-result queries ("catalog gaps") — with no IPs, user agents, or shopper identity stored. Pricing (one-time, no subscription, 30-day money-back guarantee, lifetime updates): - Pro: $89, 1 production site (plus a staging copy), 6 months of email support. - Agency: $199, up to 5 production sites, priority support, 12 months of support. - Unlimited: $399, unlimited production sites, 24 months of priority support, roadmap input and beta access. ## How the technology works (plain English) 1. Indexing: when the store owner clicks "Index now", FindAstra reads each product (title plus chosen fields) and produces an embedding — a numeric fingerprint of what the product means. Embeddings are stored in the store's own MySQL database as compact binary rows. 2. Searching: the shopper's query is embedded the same way, and products are ranked by cosine similarity (how close two meanings are). Scoring adds roughly 20–30 ms to a typical search. 3. Large catalogs: above ~5,000 indexed products, a keyword prefilter narrows candidates first and the semantic ranker re-orders the top results, so big stores stay fast (hybrid retrieval). 4. Quality gates: results only show when a product clearly stands out; gibberish falls back to popular products. Gates are self-calibrating per engine, and manual overrides exist for advanced users. 5. The Local engine does the embedding work in the shopper's browser (WebAssembly + transformers.js). The first visit downloads the ~33 MB model once from the Hugging Face model hub; the browser caches it afterwards. No store, product, or shopper data is ever sent anywhere by the Local engine. ## Requirements and compatibility - WordPress 6.4+, WooCommerce 7.0+, PHP 8.1+. - MySQL 5.7+ / MariaDB 10.3+ (FULLTEXT used for hybrid retrieval). - Local engine needs a modern evergreen browser (WebAssembly + ES2020). - Tested with Storefront, Astra, Kadence, Blocksy, GeneratePress, and other popular themes; classic and block themes supported. - Not yet supported: WordPress Multisite, image/visual search (roadmap). ## Frequently asked questions Q: Do I need an OpenAI key to use FindAstra? A: No. The default Local engine runs entirely in shoppers' browsers with zero keys, zero accounts, and no cost. The OpenAI and Hugging Face engines are part of FindAstra Pro for stores that prefer server-side processing. Q: Will it slow my store down? A: No. Scoring runs inside the existing database and adds roughly 20–30 ms to a typical search. Indexing runs in the background in small batches. The Local engine does its work in the shopper's browser. Q: Does it work with WPML or Polylang? A: Yes, with FindAstra Pro. It detects each translated product's language at index time, writes one embedding per (product, language) pair, and scopes searches to the current request language. Q: What about large catalogs? A: Stores with more than 5,000 indexed products automatically switch to a faster two-step search (keyword prefilter, then semantic re-rank). Q: Where are API keys stored? A: Encrypted at rest with AES-256-CBC using a site-specific key derived from WordPress's own auth salt. Keys are decrypted only at the moment of use. Q: Does the Local engine download anything? A: Yes, once: a ~33 MB open-source AI model from the Hugging Face model hub, which the browser caches. Only model files are fetched; no store, product, or shopper data is sent. Q: How do I see what shoppers searched for when they got no results? A: FindAstra Pro's Analytics tab records every query and flags zero-result searches under "Catalog gaps", storing no IPs or personal data. Those zero-result queries are the exact phrases a store is losing shoppers on. Q: Is there a free WooCommerce search plugin with AI? A: Yes — FindAstra's free version on WordPress.org runs real semantic search with its in-browser Local engine, with no API keys and no product cap. ## Who builds it FindAstra is built by Elyass Tarik, a telecommunications engineer (https://x.com/TarikElyass). He read Baymard Institute research showing 61% of e-commerce sites can't handle real-world query types (descriptive, feature, and use-case searches), built a semantic-search MVP for his own WooCommerce store, and turned it into a product after watching empty-result pages convert into correct products. More: https://findastra.com/about ## Canonical links - Site: https://findastra.com - Free plugin: https://wordpress.org/plugins/findastra/ - How it works: https://findastra.com/how-it-works - Pricing: https://findastra.com/pricing - Docs: https://findastra.com/docs - Changelog: https://findastra.com/changelog - Support: support@findastra.com