Skip to main content
Splendor turns your operational data into a single searchable API. Point it at your logs, documents, and structured records — even images — and query all of it three ways: by keyword, by SQL, or by meaning. There’s no search cluster to run, no embedding pipeline to build, and no per-format ingestion glue to write. You add a file (or stream events), and within minutes it’s searchable — with every result traceable to the exact source it came from.

​
What you get

  • Three kinds of search, one interface. Keyword (full-text), SQL over your records’ fields, and semantic (vector) search — including images — all return the same response shape.
  • Just add data. Upload JSONL, CSV, or images, or stream events; Splendor detects each file’s type and makes it searchable. No format declarations, no separate pipelines.
  • Traceable by default. Every result carries provenance back to the source object and line it came from.
  • Isolated by default. Every dataset belongs to one tenant; data never crosses tenant boundaries.

​
How it fits together

A few ideas recur across every endpoint — learn them once and the whole API is predictable:
  • Sources & datasets — a source is where your data lands (a file upload, or your own bucket); it feeds a dataset you query.
  • Search — one request runs text, sql, or semantic mode over one or more datasets and returns a single shared envelope.
  • Handles — a search_id, view_id, or cursor lets you page, export, and re-run a query without resending it.
  • Provenance — every result points back to the source object and line it came from.

​
One human step, then it’s automatable

Exactly one thing needs a person in a browser: signing into the console and minting an API key. Key creation accepts only a signed-in admin — not an API key — so your first credential always starts with a human.After that, everything in these docs is plain HTTP with that key: ingesting data, searching, provisioning tenants, managing sources. Hand the key to your backend — or to a coding agent like Claude Code — and it can run the rest end to end. (Think of it like cloud setup: a person does the one login, then automation takes it from there.)

​
Get started

Quickstart

From an empty workspace to your first result in a few minutes — create a source, add records, search them.

See what you can build

Search by meaning, search by image, one isolated search backend per customer.

Authentication

Get a token, select your tenant, and learn the two kinds of API key.

Core concepts

Datasets, the search model, handles, views, and provenance.

​
How a request is shaped

Every authenticated request carries two things: a bearer token that identifies you, and a tenant header that selects which workspace you are acting in.
From there the same envelope and the same handles apply whether you are running a keyword search, a SQL projection, or a semantic lookup. The next pages walk through each piece.