Quickstart

You need Docker and about five minutes. No clone, no Node toolchain, no account anywhere: the image is public at ghcr.io/beany-vu/michi-chat.

1. Get the three files

Download the contents of examples/quickstart/ into an empty directory:

• docker-compose.yml - the app image, LiteLLM, and Postgres (pgvector)
• .env.example - the settings template
• litellm.config.yaml - which real models serve the app's model aliases

2. Configure

cp .env.example .env

Open .env and set the three required values (the stack refuses to start without them):

Variable	What it is
MICHI_ADMIN_PASSWORD	Login for the operator UI at /admin
MICHI_CHAT_LITELLM_KEY	Any long random string; shared between app and proxy
MICHI_DB_PASSWORD	Any random string; the database is not exposed outside Docker

Then pick a model backend - local Ollama works out of the box if you pull three models; a hosted provider is a small yaml edit. See Model backends.

3. Run

docker compose up -d

Migrations run automatically on first boot. Then:

• http://localhost:3001 - the seeded demo tenant (Mugshot Artisan Cafe, sample data), so you can try the chat immediately
• http://localhost:3001/admin - the operator UI: create your own tenant, write its persona, enable tools, add knowledge

The demo tenant is just that - a demo. Your real bot is a new tenant you create in the admin UI.

Where things run

Service	Image	Notes
app	ghcr.io/beany-vu/michi-chat	Next.js UI + API, migrations on boot
litellm	pinned by digest	One OpenAI-compatible door to any provider
db	pgvector/pgvector:pg17	Required: a migration enables the vector extension

Everything binds to 127.0.0.1 only. For anything public-facing, put a reverse proxy in front - see Security model.