Skip to content
Getting Started
Introduction

Introduction

What Vectora is

Vectora is a self-hosted AI workspace for development teams. It runs entirely on your server — your VPS, your local machine, your company’s server — and solves two problems most AI assistants ignore: giving you and the agent the same working surfaces (filesystem, terminal, git, browser — not a chat narrating actions you never see), and making the agent truly know your project, not just generate generic code.

This happens through two complementary paths:

  • Hybrid RAG — indexing code, docs, and past decisions with keyword search (BM25) + dense vector search + reranking, so the agent answers based on what was indexed, not on what it “thinks” is true.
  • Native Context Graph — a workspace knowledge graph (functions, classes, concepts, and how they relate), built via AST parsing (tree-sitter) and LLM semantic extraction. Complements RAG with structural context embeddings alone can’t capture.

Architecture in one picture 

You (CLI / Web Chat / MCP client)
        │
        ▼
   Orchestrator (native engine — imperative loop)
        │
   ┌────┴────┐
   ▼         ▼
 coder     search      ← specialized subagents
   │         │
   └────┬────┘
        ▼
  160+ native tools (files, git, terminal, browser, RAG, web, integrations)
        │
        ▼
  SQLite + LanceDB (lite, default)  or  Postgres + Qdrant + Redis (complete)

The orchestrator is the supervisor: it answers directly for simple questions, or delegates to a specialized subagent (coder for file/git/terminal operations, search for web search and RAG) through human-in-the-loop (HITL) middleware that pauses before destructive actions for your approval.

Three ways to use it

The same backend serves three different surfaces, at the same time:

  1. Web chat — multi-user React interface, with a workbench (files, git, terminal, browser, RAG, Context Graph, Kanban) shared between you and the agent — see Using the chat and Using the workbench.
  2. CLI — vectora start, vectora config, vectora storage — see the CLI reference.
  3. MCP client — Vectora connects to MCP servers you install (a connector marketplace, plus manual registration) so the agent can use their tools. Vectora does not expose itself as an MCP server to other harnesses — see MCP client.

Free vs. Pro

Vectora is commercial, closed-source software — you run it on your own infrastructure, but the code belongs to Vectora Company (same model as Cursor, Linear, Notion).

  • Free — 100% local, no account, no dependency on Vectora Company whatsoever. You bring your own API keys (LLM, Cohere/Voyage for embeddings, Tavily for web search). Lite storage (SQLite + LanceDB).
  • Pro — optional, covers trial/billing/licensing through services.vectora.company (a small Cloudflare Worker, not a “Vectora Cloud” hosting your instance). Unlocks multi-user web chat, complete storage (Postgres + Qdrant + Redis), and webhook-triggered automations.

See current pricing.

Next step

→ Installation

Installation