Skip to main content

​
Deployment Overview

PipesHub runs anywhere Docker runs. The installer is the same on a laptop, a cloud virtual machine, and a server in your own rack — what changes is how people reach it and how you handle TLS.

​
Choose a path

​
Requirements

  • Docker with Compose v2.
  • RAM: 16 GB or more for the full deployment; the slim deployment needs less.
  • Disk: 50 GB or more, mostly for indexed content and model files.
  • A domain name and TLS certificate if anyone will reach PipesHub over anything other than localhost.
The installer checks Docker, memory and disk before it starts, and tells you if the machine is short.

​
Install

The same command works on any server:
It downloads the deployment files, asks a few questions, generates secrets, writes a .env file, pulls the images, and waits until the application answers its health check. For an unattended install, pass the public URL and skip the prompts:
See the Quickstart for the full list of installer flags.

​
Serving it over HTTPS

Browsers block features PipesHub needs when the page is served over plain HTTP from anything other than localhost. The usual symptom is a white screen after a successful deployment. Put a TLS-terminating reverse proxy in front of PipesHub for any deployment people reach by hostname.
The shape is the same whichever proxy you use:
  1. Point a DNS record at the server.
  2. Terminate TLS at Nginx, Caddy, Traefik, Cloudflare, or a cloud load balancer.
  3. Forward traffic to PipesHub on port 3000.
  4. Set the public HTTPS URL during installation, or afterwards with ./install.sh --reconfigure.
That URL is written to .env as FRONTEND_PUBLIC_URL, and it is what OAuth callbacks, email invitations and webhooks are built from. If it is wrong, sign-in and connector authorisation break in ways that are hard to trace back.

​
Choosing your backends

The installer asks which graph database, message broker and key-value store to use, and writes the answers to .env: The vector database is Qdrant, and the installer does not ask about it. The provider is chosen by VECTOR_DB_TYPE, which defaults to qdrant when unset, and the Docker Compose deployment does not set that variable. Kafka is worth choosing when you expect sustained high-volume indexing across many connectors. Redis Streams is simpler to run, and is what the slim deployment uses.

​
Day-two operations

Run these from the directory the installer prints when it finishes:
curl | bash does not change your shell’s directory, so the success banner prints the path to use.

​
Cloud guides

The GCP guide is the most detailed. The steps that are not GCP-specific — installing Docker, running the installer, setting up the reverse proxy, backing up volumes — apply anywhere.