SELF-HOST

Your server, your domain,
your model keys.

StandMeet ships as prebuilt images and one compose file. There is no hosted service to sign up for: you run the whole thing, and your corpus never leaves your machine except to reach the model provider you chose.

Two ways to bring it up

01

On Coolify

New Resource → Docker Based → Docker Compose Empty, and paste the compose file in. Coolify generates every secret, assigns the domain and terminates TLS. This is the path the author’s own instance runs on.

02

Plain Docker Compose

Download the file, put the secrets and your public URL in a .env beside it, and start it. Put your own reverse proxy (Caddy, Traefik, nginx) in front for TLS.

curl -fsSLO https://raw.githubusercontent.com/atmaxmoj/standmeet/main/infra/deploy/docker-compose.yml
# set POSTGRES / SESSION / INSTANCE secrets and your public URL in .env
docker compose up -d

The plain-Compose route is newer and less travelled than Coolify. After it starts, run the four checks below rather than trusting a green container.

What you need

  • A Linux host with Docker and Docker Compose v2
  • One domain for your page
  • A reverse proxy for TLS that forwards X-Forwarded-For (Coolify’s does)
  • An API key from the model provider you want to use

Four checks that it really came up

  1. The database tables exist (a missed schema fails silently)
  2. The backend sees visitor IPs, so bans and rate limits work
  3. An image attached to a note renders on your public page
  4. A question on your page is answered from your corpus, not general knowledge
What it runs
The web app, the Go backend, Postgres, Redis, Meilisearch for search, Gotenberg for PDFs, S3-compatible object storage (Silo), a sandboxed microsite builder, the Telegram bridge and an updater. Images come from ghcr.io/atmaxmoj.
Models
Built-in presets for Anthropic, OpenAI, DeepSeek, Kimi, Groq, SiliconFlow, OpenRouter and Together AI, plus any OpenAI-compatible endpoint — Ollama, vLLM or LM Studio on your own hardware. Visitors without a code can bring their own key and read only the public slice.
Connect your AI
Your AI writes to the corpus through MCP. Generate a key in the admin, then add the client to Claude Desktop, Cursor or any MCP client: npx -y @standmeet/mcp-client@latest
Upgrades
The admin shows when a new release is out; one button pulls the new images and recreates the containers in place. Releases that change the database ship migrations to apply.
Backups
A backup script dumps Postgres and archives the data volumes; a matching restore script brings them back.
QUESTIONS

Before you ask
the corpus.

The questions people ask most before running their own.

All questions ↗
Is there a hosted version I can sign up for?

No. StandMeet is software you run on your own server, from prebuilt images. The author runs one at sijie.xyz, which is also the live demo.

What does it cost?

The software is free and open source. You pay for your server and for the model provider whose key you configure; visitors who bring their own key pay for their own inference.

Who can see what?

You decide per entry and per role. Public visitors see only what you published; an access code reads what its role allows; rough notes in raw are never readable by visitors. Answers point to the notes they used.

Does the AI make things up about me?

It answers from your corpus and cites what it read. When a topic is outside what a visitor may read, it says so instead of guessing. You can read every transcript afterwards.