A second brain for you,
your agents, and your teams

Structured knowledge in markdown files. Schema-validated, versioned in git, searchable by any AI through MCP. One brain, every AI, persistent memory that compounds over time.

24
MCP tools
6
Extensions
1500+
Tests passing
MIT
Licensed

Why Pyrite instead of vectors-in-Postgres or Notion+AI?

{}

Typed entries with schema validation

Not flat vector blobs. Define person, decision, event, component types with validated fields. Query structurally, not just by vibes.

git

Git-native storage

Every change is a versioned commit. Branch, diff, review, rollback. Your knowledge has a full audit trail.

key

MCP server with three-tier access

Read/write/admin tiers. Give untrusted agents read-only. Give your own agents write. Keep admin for yourself.

search

Semantic + structured search

Find by meaning (vector embeddings) AND by type, tag, date range, or relationship. Both at once in hybrid mode.

plug

Plugin system

Custom entry types, MCP tools, CLI commands, validators, lifecycle hooks, relationship semantics. Six extensions ship out of the box.

disk

Zero running cost locally

Markdown files + SQLite on your disk. No cloud dependency, no subscription, no vendor lock-in. Your data is plain text.

How it works

Markdown files with YAML frontmatter in git are the source of truth. Pyrite builds a fast search index on top and exposes everything through multiple interfaces.

Architecture diagram: git-versioned markdown files flow into a SQLite index (FTS5 + vector embeddings), which serves MCP Server, CLI, REST API, and Web UI. Any AI connects here.
Source of truth
Plain markdown files in a git repo. Read them in any editor. Diff, branch, review.
Derived index
SQLite FTS5 + optional vector embeddings. Rebuild from files any time with pyrite index build.
Multiple interfaces
CLI for automation, MCP for AI agents, REST API for apps, Web UI for humans.

First-class AI integration

Pyrite speaks MCP natively. Connect Claude Desktop, Claude Code, Cursor, or any MCP-compatible AI — your knowledge base becomes persistent agent memory.

Three-tier access control

read Search, browse, explore — 14 tools. Safe for untrusted agents.
write Create, update, delete, link entries — +6 tools. For your own agents.
admin Index management, git operations — +4 tools. For you.

Connect in seconds

// claude_desktop_config.json

{

"mcpServers": {

"pyrite": {

"command": "pyrite",

"args": ["mcp"]

}

}

}

Built for

Deploy your own

Notion Team: $10/user/month. Pyrite on a $6 VPS: unlimited users, you own your data.

# Clone and deploy (Ubuntu VPS with Docker)

$ git clone https://github.com/markramm/pyrite && cd pyrite

$ bash deploy/selfhost/setup.sh kb.example.com

# Create your admin user

$ docker compose -f deploy/selfhost/docker-compose.yml exec pyrite \

python /app/deploy/selfhost/create-user.py admin yourpassword

~2 min
Clone to running instance
$6/mo
Hetzner CX22, unlimited users
Auto TLS
Caddy provisions certificates

Or install locally

pip install, create a KB, connect to Claude.

# Install

$ pip install "pyrite[all]"

# Create a knowledge base

$ mkdir my-kb && cd my-kb && pyrite init --name my-brain

# Add some knowledge

$ pyrite create --type person --title "Sarah Chen" \

--body "Engineering lead. Considering move to consulting."

# Search (keyword, semantic, or hybrid)

$ pyrite search "career transition"

# Connect to Claude Desktop or Claude Code

$ pyrite mcp   # starts the MCP server