Skip to main content
Each tutorial walks through one endpoint end to end, with copy-pasteable examples. New to retrieval-augmented generation? Start with From documents to answers to learn the concepts and pick the right endpoint. Otherwise, jump straight to Uploading & managing files, then Searching documents.

Build a searchable knowledge base

Ingest documents once, then query them. A persistent, indexed corpus you can search or ask questions over.

Uploading & managing files

Upload documents into LightOn so they become searchable in seconds, plus tagging, metadata, listing, and filtering.

Searching documents

Find the most relevant passages with a natural-language query, with scoping, reranking, and vision-mode search.

Asking questions

Get a grounded, LLM-generated answer with the sources it used, returned synchronously or streamed token by token.

Classify and organise documents

Three layers that compose: partition files into workspaces, group them with tags, and enrich them with structured facets. Pick the simplest one that solves your problem.

Workspaces

Hard containers that isolate a team’s, customer’s, or tenant’s documents. The only layer that’s also a permission boundary, via workspace-scoped API keys.

Partitioning documents into workspaces

Isolate documents per team, customer, or tenant, and scope API keys for permission-level segmentation.

Tags

Flat, reusable labels that group documents into collections, even across workspaces.

Grouping documents with tags

Build reusable collections that cut across workspaces, then scope search and ask to a collection.

Facets

Typed, hierarchical metadata for precise structured queries: classify by content type and filter by attribute values.

Organizing documents with metadata

Understand content types, attributes, and how facets enrich search.

Defining content types

Build classification trees with custom attributes, from starter templates or from scratch.

Classifying files

Assign content types to files, set attribute values, and read back structured metadata.

Filtering by facets

Scope search and ask queries by content type and attribute values.

Process documents on the fly

Convert or extract from a document in a single call. Nothing is stored, useful for feeding your own pipeline.

Parsing documents

Convert any document to clean Markdown, synchronously for quick jobs or async for large documents.

Extracting structured data

Pull typed fields out of documents using a JSON Schema you provide, in sync or async mode.