> ## Documentation Index
> Fetch the complete documentation index at: https://docs.perceo.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Overview

> Self-hosted Obsidian-style second brain your agents read through MCP.

Archivum is a self-hosted, server-hosted Obsidian-style second brain. It runs a
private Markdown wiki in your browser, stores notes as local files, and exposes
the same knowledge base to AI agents through MCP.

Drop in a PDF or paste a URL and Archivum turns it into a searchable wiki page
with source metadata, linked to what you already have.

<CardGroup cols={2}>
  <Card title="Install" icon="download" href="/archivum/install">
    Guided installer, Docker Compose, configuration, and MCP setup.
  </Card>

  <Card title="The loop" icon="repeat" href="/concepts/the-loop">
    Where Archivum sits relative to Archductor and Archfleet.
  </Card>
</CardGroup>

## What it does

* **Markdown on disk is the source of truth.** Pages are files you can read,
  back up, and edit outside the app.
* **A browser wiki with the Obsidian shape** — folders, backlinks, `[[wikilinks]]`,
  autosave, search, and a graph view.
* **File and URL ingest** that turns sources into searchable wiki pages with
  source metadata.
* **Keyword and semantic search** with question answering and citations back to
  your own pages.
* **MCP server built in.** Connect Claude Desktop, Claude Code, Cursor, VS Code,
  or any MCP-compatible client and let your agent read and write your wiki.
* **Share links, read-only public pages, and HTML/PDF export** when something
  needs to leave the vault deliberately.

## Why it matters for agents

Repos carry code. They do not carry the reasoning: why a decision went the way it
did, which approach already failed, what a customer actually asked for.

Archivum is where that lives, and the MCP server is what makes it reachable. Your
agent reads the same notes you do, so you stop pasting the same background into
every session.

<Tip>
  Set `LLM_EXTRACTION_PROVIDER` and `EMBED_PROVIDER` to `ollama` to run extraction
  and embeddings against a fully local model. Nothing leaves the machine at all.
</Tip>

## Services

Docker Compose brings up the web app, the backend and its ingest pipeline, the
MCP server over stdio and HTTP/SSE, and Caddy in front. Storage is SQLite for
structured data, Kuzu for the knowledge graph, and Qdrant for vectors — all in
local Docker volumes. Ollama is available in the same stack when you want the
models local too.

## What it is not

Archivum is the human knowledge layer. It is not the agent execution surface — that
is [Archductor](/archductor/overview) — and it is not the verification layer, which
is [Archfleet](/archfleet/overview).