Self-Hosting Pyrite — guide

Pyrite runs anywhere — your laptop, a $4/month VPS, or a Docker container. You own your data completely.

Local install (simplest)

```bash pip install "pyrite[all]" pyrite init --template research --path my-kb pyrite-server # web UI at http://localhost:8088 pyrite mcp # MCP server for AI agents ```

No server needed. Everything runs locally with SQLite.

Docker (recommended for servers)

```bash git clone https://github.com/markramm/pyrite.git && cd pyrite bash deploy/selfhost/setup.sh kb.example.com ```

This installs Docker, starts Pyrite + Caddy (auto TLS), and creates your first KB. Total setup time: ~10 minutes.

Create users

```bash docker compose -f deploy/selfhost/docker-compose.yml exec pyrite \ python /app/deploy/selfhost/create-user.py admin yourpassword ```

One-click deploy

Deploy buttons available for:

Railway — simple container hosting

Render — auto-deploy from GitHub

Fly.io — edge deployment with persistent volumes

See the Deploy documentation for details.

Privacy and security

Pyrite is designed for privacy-conscious users:

No telemetry — zero outbound network calls unless you initiate them

No cloud dependency — everything runs on your hardware

Git-native storage — your data is plain markdown files, portable and auditable

AI is BYOK — bring your own API key, or use local models. No AI runs without your explicit configuration

Per-KB access control — read/write/admin tiers, per-KB permissions

For journalism deployments with specific threat models, see the hosting security requirements in the Pyrite project KB.

Connecting AI agents to your server

See connecting-via-mcp for MCP configuration. AI agents connect locally to your Pyrite install — they don't need access to the server directly.

Self-Hosting Pyritenote