Sync your catalog

Where to go depends on your stack.

• Shopify merchants: see Shopify catalog sync. The Humind app handles everything via webhooks; nothing to build on your side.
• Everyone else (WooCommerce, Magento, headless commerce, custom backends, ERPs that own product data): push your catalog from any stack with the Humind API. Authenticate with one bearer token, send JSON over HTTPS, get JSON back.

If you can't run code against the API (a PrestaShop or CMS module, a PIM or ERP export, a feed built by your agency), expose your catalog as a NDJSON file at an HTTPS URL and register that URL in the dashboard under Settings > Catalog > Catalog import > Product feed. Humind pulls it on the schedule you pick (every 1 to 24 hours) and runs it through the same import pipeline. See Product feed URL.

Recommended integration path

A complete custom integration is four steps. Each links to the full reference.

1. Push your products

Use POST /products for a handful of products, or the NDJSON bulk import for a full catalog (one product per line, asynchronous, per-line error reporting).

Push as much structured product data as possible: the assistant is only as good as the data it can ground on. In practice:

• Standard fields first: title, description, images, variants with prices and stock, online_store_url, categories, brand.
• Everything else as metafields: composition, usage advice, scent notes, materials, dimensions, care instructions, internal rankings. The naming is yours; a stable namespace per data domain works best (e.g. olfactif.notes_tete, conseil.utilisation).
• Lists (a fragrance pyramid, a list of themes) are JSON-encoded strings with a list.* type: {"namespace": "olfactif", "key": "notes_tete", "type": "list.single_line_text_field", "value": "[\"Bergamote\",\"Mandarine\"]"}.
• Push AI-relevant metafields at product level, not variant level: variant metafields are stored but not read by the assistant.
• Selling in several languages? Use translations for titles, descriptions, and per-language metafield overrides.
• Curated recommendations ("pairs well with", "you may also like") go in related_products / complementary_products; the assistant uses them for cross-sell.

2. Activate your metafields

Stored metafields don't feed the assistant until they are activated. Call GET /settings/metafields to see what Humind detected, then PUT /settings/metafields to opt fields into the AI (include_in_ai), search filters (use_as_filter), and display. This can run as the last step of your import pipeline; the reference documents the re-indexing jobs the call triggers.

3. Add non-catalog knowledge

FAQ, shipping and return policies, size guides, brand content: push them through the Knowledge API so the assistant answers support questions, not just product ones.

4. Keep it fresh

Re-run your import on catalog changes (a daily or hourly NDJSON run is the common pattern; imports are idempotent by external_id), or let Humind do it for you with a product feed URL. Use webhooks if you need to react to Humind-side events.

Grouping and merchandising

• Product groups: declare that several products are the same model in different colors or materials so the assistant proposes alternatives inside the family.
• Collections: merchandising groupings ("summer essentials") the assistant can browse and filter on.