Skip to main content
This page lists what your infrastructure must provide before you install Collabase. It is written for the team that runs the servers or the cluster, so you can send it to them as it is. It covers Docker on a single server, Kubernetes and Docker Swarm.

Choose a deployment

All three options run the same software. They differ in how it is started, updated and scaled. If you have no reason to choose otherwise, pick Docker on one server. If your operations team already runs a Kubernetes platform and wants to operate Collabase with the tools it uses for everything else, pick Kubernetes. Both are fully supported.

Operating system and platform

macOS and Windows with Docker Desktop (WSL2) work for testing and local trials. They are not a production platform.

CPU and memory

Size for the people who work at the same moment, not for the number of accounts. Sizing & scaling maps concurrent users to resources. These are the starting points: On Docker the installer stops below 2 GB of RAM or 10 GB of free disk space. Below 4 GB, turn off live collaboration. On Kubernetes, reserve the same capacity in the namespace. What each part of Collabase may use is listed in What runs in your cluster. The memory limits of all parts add up to a little more than 11 GB, and the memory the scheduler reserves for them is about 1.75 GB. Normal operation uses far less than the limits. The limits protect the cluster from a runaway process. Run a test system and a production system as separate installations, each with its own database. A test system can use the minimum.

Storage

The total is about 45 GB on Kubernetes and 20 GB on a single Docker server, plus backups. On Kubernetes you need a default storage class that provides ReadWriteOnce volumes, or you set the class per volume. Watch the disk and grow it before it fills. Hosting & Architecture lists what takes space.

Database

Collabase ships with its own PostgreSQL. If your organization runs a managed or central database service instead, it must provide:
  • PostgreSQL 16 or later
  • The pgvector extension, which Collabase enables with CREATE EXTENSION vector during installation. The database user that installs needs the right to do this, or your database team enables the extension once beforehand.
  • A database and a user that owns it
Without pgvector, Collabase does not start. Check this with your database team before you decide to use an external database.

Network

Inbound

Nothing else is reachable from outside. The parts of Collabase talk to each other on an internal network. Live collaboration uses a WebSocket connection on the same address under /collab. A load balancer or proxy in front of Collabase must pass WebSocket upgrades through, and must allow uploads of at least 32 MB.

Outbound

Collabase makes only the connections you set up. Allow the ones you use:
Collabase verifies its licence key on the server and needs no connection to Collabase for it. Usage telemetry is off until an administrator agrees to it, and an administrator can switch it off again.

Certificates

An internal address such as collabase.company.local needs a certificate from your own certificate authority, and every browser must trust that authority.

Container images

The Collabase images share one version number and are public, so pulling them needs no sign-in. Many organizations copy the images into an internal registry first. On Kubernetes, set the repository of each image in your values file, see Use your own image registry.

Optional integrations

These are set up in Collabase after installation. Plan them with your identity and mail teams in advance.

Information for your operations team

If a team other than yours installs and maintains Collabase, agree on these points first.
Collabase stores integration and mail credentials encrypted with keys that are created at installation. Include these keys in your backup. On Kubernetes they live in the collabase secret. On Docker they are in the .env file. Without them, a restored database cannot decrypt those credentials, and you enter them again.

Installation with Docker

Install on one Linux server.

Installation on Kubernetes

Install with the Helm chart.

Sizing & scaling

Match resources to your team.
Last modified on November 5, 2026