Mari

Mari is the product knowledge platform. It reads the places your team already writes, keeps one searchable copy of what it knows, and shows which parts have an owner, a source, and a recent check behind them.

Where a search tool stops at ten links, Mari carries on:

How it fits together

Sources are ingested on a schedule or by webhook, chunked, hashed, and embedded. Unchanged content is never re-embedded, so re-syncing a repository that has not moved costs almost nothing.

Retrieval is hybrid: Postgres full-text scoring and pgvector cosine similarity over the same chunks, weighted by the tags you apply. Answers stream with numbered citations.

An agent sits on top with the same permissions you have. It can search, read and edit documents, tag, manage facts and answers, trigger a sync, and move the console around you. Every action it takes lands in the audit trail.

How you can run it

Mari is open source under Apache 2.0, and the open-source build is free to run yourself: a Docker Compose stack with Postgres (pgvector included), plus ollama if you want local models. Every model-backed feature has a deterministic fallback.

The enterprise plan runs it for you, two ways:

The enterprise plan also carries features that are not in the open-source build. Data handling for either option is set out in the MSA and DPA you sign. Talk to us if you want it run for you.

Where to start

  1. Quick start
  2. Set up a workspace
  3. Connect a source

Then one guide per thing:

And the reference pages:

About these pages

This documentation is new and does not yet cover every screen. Pages are written against the code they describe, so what is here is accurate, but the set is not complete. The console reference lists every screen, including ones without a guide yet.