The vector database that works with no network and no accounts

Embedded ONNX models, GraphRAG, and eight storage backends. Nothing to sign up for, no API key, no service to run. pip install and search.

Get started Use the dashboard Command line SDKs MCP GitHub

A question goes in, and every result says how well it matched and what found it: meaning, keywords, or both.

pip install vectrixdb

from vectrixdb import Vectrix

db = Vectrix("my_docs")
db.add(["Python is great", "JavaScript powers the web", "Rust is fast"])

results = db.search("programming")
print(results.top.text)

That is the whole setup. The embedding model ships in the package, so the first search works offline on a machine that has never seen an API key.

Every way in

The same collections answer a Python program, the vectrixdb command, a client in TypeScript, Go or Rust, and an assistant over MCP. Each client keeps the same safety rules: no key over plain HTTP, no redirects, certificates always checked.

Command linePythonTypeScriptGoRustMCP

pip install "vectrixdb[client]"
vectrixdb login --url https://vectors.example.com --device
vectrixdb query "how long do refunds take?" --name handbook

Use the command line on a server

pip install "vectrixdb[client]"

import vectrixdb

db = vectrixdb.connect("https://vectors.example.com", key="...")

Use it from Python, TypeScript, Go or Rust

npm install vectrixdb

import { VectrixClient } from "vectrixdb";

const db = new VectrixClient({ url: "https://vectors.example.com", key: "..." });

Use it from TypeScript · npm

go get github.com/knowusuboaky/VectrixDB/sdk/go/v2

db := vectrixdb.New("https://vectors.example.com", vectrixdb.WithKey(key))

Use it from Go · pkg.go.dev

cargo add vectrixdb

let db = vectrixdb::Client::new("https://vectors.example.com").key(&key).build()?;

Use it from Rust · crates.io · docs.rs

pip install "vectrixdb[mcp]"
vectrixdb mcp --name handbook

Use your collections from an assistant over MCP · Coding agents

See it work

pip install "vectrixdb[api]" and vectrixdb serve give the same collections a REST API and a dashboard. These are the real pages, the real commands and the real containers, filmed by the scripts that keep the pictures current.

Command line

vectrixdb pointed at a server: a folder ingested, a search with citations, and a key it will not send over plain HTTP.

SDKs: Python, TypeScript, Go, Rust

One server, one document, and the same search from four languages, each answering with the same citation.

MCP

An assistant connects, lists the collections and searches them, with real calls and real answers. Every tool: Tools reference.

MCP keys for a team

A key made for one collection cannot tell the others exist, and every refusal comes back in words.

Doctor

vectrixdb doctor tries every part of an install and says what to do about anything missing.

Ingest

A PDF goes in. The server says what it read: pages, chunks, and a citation for each.

Collections

Every collection with its state, mode and model, and what it holds, tab by tab.

Evaluate

Every setup ranked on your own questions, and three picks: most found, best balance, fastest.

Access

People who sign in as themselves, roles, and a record of who read what.

Console

Build a request, send it, and copy the curl that does the same.

The whole walk

Overview, Collections, the evaluation run with its picks, and Audit.

Containers

Compose starts the server, the extraction service and Jaeger. A scan goes in and a search finds it.

Traces

One trace from the server into the extraction service: counts and timings, never the text.

Where to go next

• Getting started

  Learn the library by building something small, end to end.

• How-to guides

  Recipes for specific jobs: error handling, storage backends, the REST API.

• Reference

  Signatures and behaviour, generated from the source.

• Explanation

  Why the search modes differ, and where the pure-Python index runs out.

• Conversation memory

  Turns, pinned facts and a context block sized to a token budget.

• Command line

  The vectrixdb command on a server: sign in with a company account, ingest, search.

• SDKs

  Python, TypeScript, Go and Rust clients, one contract, the same safety rules.

• MCP server

  Give an assistant a collection to search and remember into.

• Ship your own client or command

  A company's own acme vectors command or SDK, built on these.

• Inside your company's registry

  JFrog Artifactory or Nexus: packages, images and models, and only your wrapper allowed in.

• Run it in containers

  The server, the extraction service and tracing, from one compose file.

• Use the dashboard

  Every page of the server's dashboard, with pictures, and who sees which.

• Deploy the server

  One settings file, checked before a start, read by every command.

• Measure retrieval

  Golden questions written from your own documents, and checked against them.

• Keys and roles

  Every role, keys scoped to collections and expiring, and who may do what.

• Size the server for load

  Model calls in line, /ready, threads and memory, so a busy server stays up.

• Chunking

  The strategies, parent sections and late chunking, and how to pick.

• What it does not do

  Every limit in one place, with what to do about each.

What is in the box

Search modes	Dense, Hybrid, Ultimate, Graph (GraphRAG)
Storage	Memory, SQLite, Lakebase, Delta Lake, Cosmos DB, Azure AI Search, OpenSearch, Aurora PostgreSQL
Models	Bundled ONNX, or bring your own from HuggingFace
Readers	PDF, Word, Excel, Markdown, scans by sight, speech, video
Memory	Conversation turns, pinned facts, recency and feedback weighting
Clients	Python, TypeScript, Go and Rust, and the vectrixdb command
Assistants	MCP server for Claude, VS Code and other MCP clients
Extras	Document index with chunking, dashboard, REST API

Install what you need

The core install is deliberately small: vector search, and nothing else.

pip install vectrixdb              # core
pip install vectrixdb[api]         # + REST API and dashboard
pip install vectrixdb[signin]      # + people signing in: single sign-on, passkeys, codes
pip install vectrixdb[mcp]         # + MCP server for assistants
pip install vectrixdb[documents]   # + PDF, Word and Excel readers
pip install vectrixdb[extract]     # + every local reader: documents, OCR, speech, video
pip install vectrixdb[aws]         # + OpenSearch and Aurora
pip install vectrixdb[azure]       # + Azure AI Search and Cosmos DB
pip install vectrixdb[databricks]  # + Lakebase and Delta Lake
pip install vectrixdb[nlp]         # + spaCy for GraphRAG entity extraction
pip install vectrixdb[all]         # every backend, model and server extra; not extract or nlp

The clients for a server somewhere else are small, and need nothing else:

pip install "vectrixdb[client]"                          # Python client and the vectrixdb command
npm install vectrixdb                                    # TypeScript and JavaScript
go get github.com/knowusuboaky/VectrixDB/sdk/go/v2       # Go
cargo add vectrixdb                                      # Rust