Mayan EDMS

Last reviewed Sep 13, 2026 · 882 words

Most people who install Mayan EDMS wanted Paperless-ngx and did not know it yet. Mayan is a real electronic document management system in the enterprise sense: document types with metadata schemas, multi-stage workflows with role-based transitions, cabinets, indexes, retention, an audit trail, and granular permissions on every one of those. That is a different product from "scan my mail and let me search it," it wants 2 GB of RAM and four containers, and the catalogue is right to rate it Hard. If you have a compliance requirement or a process with approvals, keep reading. If you have a shoebox of receipts, go install Paperless-ngx.

A workflow engine that also stores files

The mental shift is that documents are not the top-level object; document types are. A document type (invoice, lease, HR record) defines which metadata fields apply, which OCR and parsing settings run, which indexes the document appears in, and which workflows attach on upload. A workflow is a state machine: states, transitions, and the roles allowed to trigger each transition, with actions fired on entry (add a tag, send an email, move to a cabinet, call a webhook). An invoice can move from Received to Approved to Paid with the accounting role gated on the middle step and an audit event on every hop.

Versioning is built in and every file upload against an existing document is a new version with a diff of pages. Retention and scheduled deletion, access control down to per-document ACLs, and an LDAP integration round it out. None of this exists in the lighter tools, and none of it is needed by a household.

Four containers before the first page

Mayan is a Django app with Celery workers, and the supported deployment is the project's own compose file with PostgreSQL, Redis, RabbitMQ and the app, plus separate worker containers for fast, medium and slow queues. The skeleton looks like this, with the project's full file the thing you should actually run:

services:
  app:
    image: mayanedms/mayanedms:latest
    ports:
      - "80:8000"
    environment:
      MAYAN_DATABASES: "{'default':{'ENGINE':'django.db.backends.postgresql','NAME':'mayan','USER':'mayan','PASSWORD':'mayanpass','HOST':'postgresql'}}"
      MAYAN_CELERY_BROKER_URL: amqp://mayan:mayanpass@rabbitmq:5672/mayan
      MAYAN_CELERY_RESULT_BACKEND: redis://:mayanpass@redis:6379/1
    volumes:
      - mayan-data:/var/lib/mayan
    depends_on: [postgresql, redis, rabbitmq]

On first boot the app creates an admin user and prints a random password to the container log; grab it with docker compose logs app | grep -i password and change it. OCR runs through Tesseract inside the image and is queued to the slow workers, so a 200-page scan takes minutes, not seconds, and the UI does not block. The /var/lib/mayan volume holds every file; PostgreSQL holds everything else. Back up both together or restore nothing.

Budget 2 GB of RAM for an idle instance and more under OCR load. On a 4 GB VPS it will run; on a Raspberry Pi it will make you unhappy.

The model you have to learn before it is useful

Freshly installed, Mayan does nothing helpful. You must create a document type, attach metadata types to it, define at least one index or cabinet, and set up a source (a watch folder, an email inbox, a staging folder for the scanner) before a single document lands where you expect. This is the "complex to configure" con in the catalogue and it is accurate; I would allow a full weekend to model a small office's document flow, and I would write the model down before clicking anything, because reorganising types after 5,000 documents are in is painful.

The reward is that once modelled, the system enforces the process. Uploads to the invoice inbox cannot skip metadata; the approval step cannot be taken by the person who entered the invoice; the audit log shows who did what to which version. That enforcement is the product, and the document management category has nothing else open source that does it this thoroughly.

Who actually runs it

Small legal and accounting practices, municipal offices, NGOs with donor compliance obligations, and engineering teams replacing DocuWare or M-Files without the licence. They have someone whose job includes administering it. A 1,100-star Apache-2.0 project that has been shipping since 2011 with a company offering support is a reasonable thing to bet a small organisation's records on, and it is the direct self-hosted answer to DocuWare for people leaving that world.

Home users and solo freelancers are better served by Paperless-ngx, which does OCR, tags, correspondents and full-text search in 512 MB with a tenth of the concepts. Docspell is the middle option if you want more structure than Paperless without workflows.

What I'd do

If you can name the workflow (who receives, who approves, who archives, and what the retention rule is), run Mayan on a 4 GB box with the official compose, spend a weekend on the document type model, and put PostgreSQL plus the data volume in a nightly backup. If you cannot name the workflow, you do not need it: install Paperless-ngx this afternoon and revisit Mayan only when someone asks for an approval step or an audit trail.