Introduction

Docspell Documentation

Introduction🔗

Docspell aims to be a simple yet effective document organizer that makes stowing documents away very quick and finding them later reliable (and also fast). It is a bit opinionated and more targeted for home use and small/medium organizations.

In contrast to many DMS, the main focus is not so much to provide all kinds of features to manually create organizational structures, like folder hierarchies, where you place the documents yourself. The approach is to leave it as a big pile of documents, but extract and attach metadata from each document. These are mainly properties that emerge from the document itself. The reason is that this is possible to automate. This makes it very simple to add documents, because there is no time spent to think about where to put it. And it is possible to apply different structures on top later, like show first all documents of a specific correspondent, then all with tag 'invoice', etc. If these properties are attached to all documents, it is really easy to find a document. It even can be combined with fulltext search for the, hopefully rare, desperate cases.

Of course, it is also possible to add custom properties and arbitrary tags.

Docspell analyzes the text to find metadata automatically. It can learn from existing data and can apply NLP techniques to support this. This metadata must be maintained manually in the application. Docspell looks for candidates for:

  • Correspondents
  • Concerned person or things
  • A date and due date
  • Tags

For tags, it sets all that it thinks do apply. For the others, it will propose a few candidates and sets the most likely one to your item.

This might be wrong, so it is recommended to curate the results. However, very often the correct one is either set or within the proposals where you fix it by a single click.

Besides these properties, there are more metadata you can use to organize your files, for example custom fields, folders and notes.

Docspell is also for programmers. Everything is available via a REST or HTTP api and can be easily used within your own scripts and tools, for example using curl. There are also features for "advanced use" and many configuration options.

Components🔗

Docspell consists of multiple components that run in separate processes:

  • REST server
  • JOEX, short for job executor
  • Fulltext Search Index (optional, currently Apache SOLR)

The REST server provides the Api and the web application. The web application is a SPA written in Elm and is a client to the REST api. All features are available via a http/rest api.

The joex is the component that does the “heavy work”, executing long-running tasks, like processing files or importing your mails periodically. While the joex component also exposes a small REST api for controlling it, the main user interface is all inside the rest server api.

The rest server and the job executor can be started multiple times in order to scale out. It must be ensured, that all connect to the same database. And it is also recommended (though not strictly required), that all components can reach each other.

The fulltext search index is another separate component, where currently only SOLR is supported. Fulltext search is optional, so the SOLR component is not required if docspell is run without fulltext search support.

Terms🔗

In order to better understand the following pages, some terms are explained.

Item🔗

An item is roughly your document, only that an item may span multiple files, which are called attachments. An item has meta data associated:

  • a correspondent: the other side of the communication. It can be an organization or a person.
  • a concerning person or equipment: a person or thing that this item is about. Maybe it is an insurance contract about your car.
  • tag: an item can be tagged with one or more tags (or labels). A tag can have a category. This is intended for grouping tags, for example a category doctype could be used to group tags like bill, contract, receipt etc. Usually an item is not tagged with more than one tag of a category.
  • a folder: a folder is similiar to a tag, but an item can only be in exactly one folder (or none). Furthermore folders allow to associate users, so that items are only visible to the users who are members of a folder.
  • an item date: this is the date of the document – if this is not set, the created date of the item is used.
  • a due date: an optional date indicating that something has to be done (e.g. paying a bill, submitting it) about this item until this date
  • a direction: one of "incoming" or "outgoing"
  • a name: some item name, defaults to the file name of the attachments
  • some notes: arbitrary descriptive text. You can use markdown here, which is properly formatted in the web application.

Collective🔗

The users of the application are part of a collective. A collective is a group of users that share access to the same items. The account name is therefore comprised of a collective name and a user name.

All users of a collective are equal; they have same permissions to access all items. The items don't belong to a user, but to the collective.

That means, to identify yourself when signing in, you have to give the collective name and your user name. By default it is separated by a slash /, for example smith/john. If your user name is the same as the collective name, you can omit one; so smith/smith can be abbreviated to just smith.

By default, all users can see all items of their collective. A folder can be used to implement other visibilities: Every user can create a folder and associate members. It is possible to put items in these folders and docspell shows only items that are either in no specific folder or in a folder where the current user is owner or member.

Rationale🔗

In 2019, I started to think about creating a dms-like tool that is now Docspell. It started at the end of that year with the initial version, including the very basic idea around which I want to create some kind of document management system.

The following anecdote summarizes why I thought yet another dms-like tool might be useful.

I tried some DMS at that time, to see whether they could help me with the ever growing pile of documents. It's not just postal mail, now it gets mixed with invoices via e-mail, bank statements I need to download at some "portal" etc. It's all getting a huge mess. When looking for a specific document, it's hard to find.

I found all the enterprisy DMS are way above of what I need. They are rather difficult to setup and very hard to explain to non-technical people. They offer a lot of features and there is quite some time required to extract what's needed. I then discovered tools, that seem to better suite my needs. Their design were simple and very close to what I was looking for, making it a good fit for single user. There were only a few things to nag:

  1. Often it was not possible to track multiple files as one "unit". For example: reports with accompanying pictures that I would like to treat as a single unit. It also more naturally fits to the common e-mail.
  2. Missing good multi-user support; and/or a simple enough interface so that non-technical users can also make sense of it.
  3. Missing some features important to me, like "send this by mail", a full REST api, and some more
  4. still a lot of "manually" organizing documents

These are not big complaints, they are solvable somehow. I want to focus on the last point: most systems didn't offer help with organizing the documents. I didn't find any, that included basic machine learning features. On most systems it was possible to organize documents into a custom folder structure. But it was all manually. You would need to move incoming documents into some subfolder. Some systems offered rules that get applied to documents in order to put them into the right place. Many offered tags, too, which relieves some of weight of this text. But they were also all manual. So the idea came to let the computer do a little more to help organize documents.

Let's start with the rules approach: A rule may look like this:

when the document contains a text 'invoice' and 'repair company x', then put it in subfolder B".

This rule can be applied to all the new documents to get automatically placed into this subfolder. I think there are some drawbacks to this approach:

  • rules may change over time. Then you either must re-apply them all to all documents or leave older ones where they are. If re-applying them, some documents may not be in places as before which can easily confuse coworkers.
  • these rules may interfere with each other, then it might get more difficult to know where a document is
  • rules can become complex, be comprised of regular expressions, which are really only suited to technical people and need to be maintained.

I decided to try out a different approach: a "search-only" one¹. Instead of using a manual created folder structure, I simply search every time using this rule. In essence such a rule is a search query. But searching with rules like the one above is not very efficient. One would need to do fulltext searches, even extracting dates "on the fly" etc. It wouldn't be very reliable either. That's why documents have properties (called metadata). In my case most of them have a correspondent, a date and so on. If these properties were defined on documents, the queries become quite efficient. The idea is now, not to use rules for moving documents to some place, but for attaching properties, information, to each document. This solves a few issues: they can't get easily out of sync, and they can't interfere. Then docspell can help with finding some of these properties automatically. For example: it can propose properties by looking at the text. It can also take existing documents into account when suggesting tags. In docspell, it is not possible to define custom rules, instead it tries to find these rules for you by looking at the text and your previous documents.

That said, there is still a manual process involved, but I found it much lighter. Once in a while, looking at new documents and confirming or fixing the metadata is necessary. This doesn't involve deciding for a place, though. What properties you are interested to track can be configured; should you only need a correspondent and a date, everything else can be hidden.

So in docspell, all documents are just in one big pile… but every document has metadata attached that can be used to quickly find what you need. There is no folder structure, but it is possible to later apply certain hierarchical structures. It would be possible to create a "folder structure", like the one mentioned above: click on correspondent repair company x; then on tag invoice, then concerning=car and year=2019. A UI could be created to present exactly this hierarchy. Since I can't know your preferred structure (not even my own…!), the docspell ui allows every combination, regardless any hierarchies. You can first select a correspondent, then a tag or the other way around. Usually it's not necessary to go very deep.

That's all about it! I thought why not try this approach and at the same time learn about some technologies around. In the last year, docspell evolved to a quite usable tool, imho. This was only possible, because very nice people gave valueable feedback and ideas!

¹This is inspired by tools like mu and GMail.