Turn anything into
something queryable.
Vectorless is the retrieval platform that structures your documents, PDFs, notes, and data — and makes them instantly queryable by any LLM. No chunking. No embeddings. Just reasoning.
RAG is broken. Chunks lose context. Embeddings miss meaning. You deserved a better primitive.
Structure first. Query second.
Most retrieval systems start by destroying your documents — slicing them into chunks that lose coherence at every cut. Vectorless takes the opposite approach.
We read your document. We understand its structure. We build a navigable map of it. And when a question comes in, an LLM reasons its way to the right section — not the nearest vector.
The result: retrieval that actually understands what it's looking for.
Structure-aware ingest
Every document is parsed into its natural sections — chapters, headings, clauses, entries. The structure becomes the retrieval unit.
Reasoning-based navigation
At query time, an LLM reads your document's map and decides what to fetch. No similarity math. No threshold tuning.
Parallel resolution
When multiple sections are needed, they're fetched simultaneously. Complex questions cost the same as simple ones.
Three steps. Clean API. Done.
Add your document
Point Vectorless at anything: a PDF, a Word doc, a URL, a plain text file, a knowledge base. We parse it, extract or generate a table of contents, and store every section as an addressable unit.
Get a navigable map
You receive a structured document map — a JSON manifest of every section, with titles, summaries, and direct retrieval links. This is your document's new interface.
Query with reasoning
Pass the map and a user question to any LLM. It reads the map, picks the relevant sections, and we fetch them — in parallel, in milliseconds. You get complete, structured context back.
// 01 - Add your document
const { doc_id, toc } = await vectorless.addDocument(file);
// 02 - Get a navigable map
const sectionIds = await llm.reason(toc, userQuestion);
// 03 - Query with reasoning
const context = await vectorless.fetchSections(doc_id, sectionIds);
const answer = await llm.answer(context, userQuestion);Everything retrieval should be.
Structured Ingest
Turn PDFs, DOCX, TXT, URLs, and raw text into structured, addressable section maps automatically.
LLM-Navigable Maps
Every document becomes an llms.txt-style manifest — a map any language model can reason over natively.
Parallel Section Fetching
Fetch one section or twenty. The latency is the same. Fan-out retrieval without the fan-out cost.
Hybrid Retrieval
Not all documents have structure. Enable embedding-based fallback alongside reasoning — the best of both approaches.
Deterministic Links
Every section gets a stable, addressable URL. Retrieval is auditable, debuggable, and reproducible.
Any LLM, Any Framework
Vectorless is a retrieval primitive. Bring your own LLM. Drop it into LangGraph, CrewAI, LlamaIndex, or your own stack.
ToC Generation
No headings in your document? We generate a semantic table of contents using an LLM — structure from the unstructured.
Postgres Native
One database. JSONB for maps, pgvector for embeddings, full-text search built in. No separate vector infrastructure.
Built for documents that matter.
Research & Academia
Query across papers, systematic reviews, and literature libraries with precision that embeddings can't match.
Legal & Compliance
Contracts, policies, and regulatory documents — structured retrieval that respects clause boundaries, not chunk limits.
Clinical & Medical
Guidelines, protocols, drug references. Retrieve complete clinical sections, not fragments of dosing tables.
Technical Documentation
API references, system manuals, codebases-as-docs. Navigate like an expert, not a search engine.
Enterprise Knowledge Bases
Internal wikis, SOPs, policy documents. Turn your company's knowledge into a queryable, structured layer.
Education & e-Learning
Textbooks, syllabi, course materials. Students and AI tutors get the right chapter — not a chunk of three different ones.
A retrieval primitive that fits anywhere you build.
Vectorless ships as a TypeScript SDK first, with Python bindings coming. It's intentionally narrow — ingest and retrieve. Your reasoning, your LLM, your orchestration. We handle the hard parts of structured storage.
MCP server integration coming — use Vectorless natively inside Claude, Cursor, and any MCP-compatible agent runtime.
vectorless.addDocument(file, options)
// → { doc_id, toc }
vectorless.getToC(doc_id)
// → ToC manifest
vectorless.fetchSection(doc_id, section_id)
// → { title, content, page_range }
vectorless.fetchSections(doc_id, section_ids)
// → Section[]
vectorless.listDocuments()
// → DocumentSummary[]Simple pricing.
Coming soon.
Vectorless is in early access. If you're building something interesting with RAG, documents, or LLM retrieval — we want to hear from you.
Join the Early Access List