Recipes

A few patterns show up again and again. Each one below is a complete, real use of Splendor — what you get, and the handful of calls that get you there. They all build on the Quickstart, so do that first if you haven’t.

Search your logs and events by meaning

You have a stream of events — application logs, support tickets, audit trails — and you want to find things by what they mean, not the exact words. A user writes “I can’t log in”; you want the auth-timeout errors that never say “log in.” Push events to a source, then search semantically:

# Push a batch of events (ingest token; no tenant header)
curl https://api.withsplendor.com/v1/ingest/sources/app-events/logs \
  -H "Authorization: Bearer $SPLENDOR_INGEST_TOKEN" \
  -H "Content-Type: application/json" \
  -d '[{"level":"error","message":"auth token expired during checkout","service":"auth"}]'

# Find it by meaning, not keywords
curl https://api.withsplendor.com/v1/search \
  -H "Authorization: Bearer $SPLENDOR_TOKEN" \
  -H "X-Splendor-Tenant-Id: $SPLENDOR_TENANT_ID" \
  -H "Content-Type: application/json" \
  -d '{"text": "users can'\''t log in", "datasets": ["app-events"], "semantic": true, "limit": 10}'

The same dataset also answers keyword and SQL queries — so you can filter by service, range over time, and search by meaning, all against one set of records. → Full walkthrough: Stream logs · Semantic search

Find records by what their image looks like

You have records with images — products, listings, documents with photos — and you want to find a record by the content of its image, then filter it like any other record. Search “a brown wooden chair” and get back the product, which you can still filter by price or in_stock. Tell the source which field holds the image, then search the image modality:

# When you create the source, declare the image field
#   "semantic": { "image_embeddings": { "fields": ["photo_key"] } }

curl https://api.withsplendor.com/v1/search \
  -H "Authorization: Bearer $SPLENDOR_TOKEN" \
  -H "X-Splendor-Tenant-Id: $SPLENDOR_TENANT_ID" \
  -H "Content-Type: application/json" \
  -d '{"text": "a brown wooden chair", "datasets": ["products"],
       "semantic": true, "content_filter": "images", "limit": 10}'

Splendor embeds each referenced image onto its record, so a visual match returns the record — metadata and all — not a bare image. One entry, found by what it says and what it looks like. → Full walkthrough: Sources & images

Load a large image corpus

Your images live somewhere else — a bucket, or a dataset of image URLs — and you want to search them. Splendor embeds images from your own storage (not arbitrary web URLs), so bring the pixels into a source first. For many images, presign a batch in one call, PUT them in parallel, then point each record at the returned image_reference.

# 1. Presign up to 256 image objects in one call (one PUT URL each). No ingest
#    run is created per image — the objects are just stored for records to reference.
curl https://api.withsplendor.com/v1/ingest/sources/$SOURCE_KEY/asset-uploads \
  -H "Authorization: Bearer $SPLENDOR_INGEST_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"assets": [
        {"content_type": "image/jpeg", "object_size_bytes": 84213, "filename": "a12.jpg"},
        {"content_type": "image/png",  "object_size_bytes": 12044, "filename": "b7.png"}
      ]}'
# → { "assets": [ { "upload_url": "https://…", "required_headers": { … },
#       "image_reference": "s3://…/sources/…/a12.jpg" }, … ] }

# 2. PUT each image to its upload_url in parallel, sending required_headers.

# 3. Ingest records whose declared image field holds the returned image_reference
#    (source config: "semantic": { "image_embeddings": { "fields": ["image"] } }):
#    { "id": "sku-a12", "image": "s3://…/sources/…/a12.jpg", "price": 249 }

Each record embeds its referenced image when it ingests; then search with content_filter: "images" as above. Use image_reference verbatim — a bare object key is read relative to the source prefix and would resolve to the wrong object. → Sources & images

Give every one of your customers their own search backend

You’re building a product on Splendor and each of your customers needs isolated search over their own data. You don’t want to run a search cluster per customer, and their data must never mix. Provision a tenant per customer with one platform key, and act inside any of them by naming it in the header:

# Create a tenant for a customer
curl https://api.withsplendor.com/v1/platform/tenants \
  -H "Authorization: Bearer $SPLENDOR_PLATFORM_KEY" \
  -H "Content-Type: application/json" \
  -d '{"name": "Acme Corp", "external_id": "cus_8210"}'

# Then do anything inside that tenant with the SAME key + its id
curl https://api.withsplendor.com/v1/search \
  -H "Authorization: Bearer $SPLENDOR_PLATFORM_KEY" \
  -H "X-Splendor-Tenant-Id: t-3f9c2a1b..." \
  -H "Content-Type: application/json" \
  -d '{"text": "invoice", "datasets": ["documents"], "limit": 10}'

One credential creates and operates an unlimited fleet of tenants — each fully isolated — all over the API. → Full walkthrough: Build a platform · Provision tenants

Get started

Core concepts

Guides

Build a platform

Roles, permissions & plans

Platform reference

Search your logs and events by meaning

Find records by what their image looks like

Load a large image corpus

Give every one of your customers their own search backend

Keep going

The search model

Connect a source

​Search your logs and events by meaning

​Find records by what their image looks like

​Load a large image corpus

​Give every one of your customers their own search backend

​Keep going

The search model

Connect a source

Search your logs and events by meaning

Find records by what their image looks like

Load a large image corpus

Give every one of your customers their own search backend

Keep going