Getting Started
Overview
What you need to run Vectorless and ask your first question.
Updated 2026
The hosted service is closing, so getting started means running an engine yourself. If you just want the commands, jump to the Quickstart.
The mental model
Ingest a document
You send a PDF (or Markdown, HTML, DOCX, text). For a PDF, the engine reads its layout, finds its contents page, places each entry on its page and splits long sections. Nothing is chunked or embedded. See How it works.
Ask a question
Jev ranks the sections, then the pages, and the engine returns the evidence pages verbatim with their page ranges.
Get an answer with pincites
On POST /v1/answer, a generative model writes the answer over the evidence
pages and each claim is pinned to a line on its page.
What you'll need
- Go 1.25+ and Postgres, or Docker.
- A generative model key (Anthropic, OpenAI, Gemini, or an Anthropic-compatible gateway such as GLM on z.ai) for the answer.
- A TypeSafe key for Jev, if you want the pipeline the benchmarks measured. Without one the engine falls back to generative navigation; see what needs a TypeSafe key.
- One of the SDKs, or plain HTTP.