Install Prism

Start with Docker, protect the installation, then connect the AI provider from the browser. The standard stack includes the execution workspace, PostgreSQL/pgvector, SearXNG and Traefik.

1. Prepare the host

Install Git, Docker Engine and the Docker Compose plugin. Check that ports 48080 (Prism) and 80 (Traefik) are free. You do not need a host GPU when the model runs on another server or API. If you host a model, size RAM/VRAM and disk for that model separately.

docker --version
docker compose version
docker info

2. Download the project

git clone https://github.com/blazux/prism.git
cd prism
cp .env.example .env
chmod 600 .env

3. Set your access token

Run the command below and put its output in PRISM_TOKEN in .env, without the leading comment marker. Generate your own value; do not reuse a documentation placeholder. Keep the bundled database and path defaults for this first installation.

openssl rand -hex 32

For shared mode, set MULTI_USER=1 before starting and use the shared-mode guide. The first account becomes the global administrator. Choose this mode deliberately: converting an existing personal installation migrates its configuration and secrets.

4. Start the services

docker compose up -d
docker compose ps

Open http://localhost:48080 and sign in with your token. The first launch downloads the images; allow the database and workspace to initialize. Use the optional protonmail profile only if you need Proton Mail Bridge.

5. Connect the first model

Open Settings → AI provider, add a source, enter its URL/key and click Connect. Choose a tool-capable conversation model. Model selection saves automatically; use Use as default when choosing among sources, and Save other changes for remaining edits. Wait for All changes saved.

6. Verify the installation

The UI should open, the execution workspace should be running, and a simple chat should produce a response. Then follow Get started to verify a widget and persistence. A model returning plain text instead of structured tool calls needs a server/model configuration check.

docker compose logs --tail=100 prism-server
docker compose logs --tail=50 workspace