Unbody

Everything the platform stores is reachable through one GraphQL endpoint that works as a knowledge base API, with filters, sorting, pagination and nested objects in the same query. You ask for the fields you need and get enriched records back, including generated summaries and keywords. Separate image and video endpoints cover media processing, and the Admin API manages projects and sources without the dashboard. The appeal is fewer moving parts: one schema, one auth token, one place to debug.

Google Drive, Discord and GitHub have ready-made integrations. Each one defines an entry point, such as a folder or a channel, then pulls the content on a schedule. For anything else, the Push API accepts records in a schema you define, which covers internal databases, exported archives and custom feeds. That combination handled the usual starting point for a knowledge product without a custom ETL job.

Before content is indexed, a configurable pipeline parses it and runs enhancers over the result. AutoSummary writes short abstracts, AutoKeywords pulls out key terms, and AutoVision generates image captions, extracts text with OCR and tags objects. Crawlers turn web pages into structured records. Pipelines are set up in the dashboard or through the Admin API, and custom plugins can be registered if the built-in set isn't enough.

The API supports semantic search, keyword search and hybrid search, plus two similarity modes: record similarity and visual similarity. Semantic search matches on meaning, which helps when a user's wording doesn't match the document. Keyword search stays predictable for exact terms. Hybrid blends the two. Visual similarity works on image vectors, so a product photo can find lookalikes in a catalog.

Two generation modes sit on top of retrieval. From One runs a prompt against each object and can inject field values into the prompt. From Many takes a set of objects and produces one result across them, which suits summaries and comparisons. A separate generative JSON mode returns structured data checked against a schema, useful for extracting fields instead of prose. You pick the model through plugins, so the backend isn't tied to a single provider.

The TypeScript SDK turns GraphQL calls into a chainable client: get a collection, filter, search, select fields, then generate. Starter projects exist for Next.js, Nuxt, NestJS and Express, and the frontend layer ships prebuilt components for React and Vue. None of it locks you into a hosted service, since the same SDK points at a local instance.