Set up a workspace

A workspace is one team's knowledge base: its sources, its facts, its people.

First run

Open the console and sign in. A brand new workspace lands on Setup, which is the only screen reachable before you have a session, alongside the sign-in page. Once a workspace exists, Setup redirects to the overview.

What Mari needs

This page describes the self-hosted build. On the enterprise plan, bring your own cloud or hosted, Mari runs this part for you and you start at sign-in.

• Postgres with pgvector. Storage, full-text search, and vector similarity all live in the same database, so there is no separate search cluster to run.
• A model provider. Embeddings and answer generation. Local models through ollama are supported, and the console degrades honestly when a provider is offline rather than failing silently.

Model choice is per workspace under Settings, Models.

For Kubernetes, a Helm chart ships with each release as mari-<version>.tgz on the releases page, pinned to that release's public images. Upgrades apply in place with helm upgrade --install and leave volumes and data untouched.

Invite the rest of the team

Settings, Members manages who is in the workspace. Members inherit the workspace's sources; they do not each connect their own.

Next

Connect a source, which is where a workspace stops being empty.