A neutral schema parsing tool that outputs diagrams and code.
Clone or download
nmoya fix(lacinia): Remove empty :implements key in lacinia EDN
This doesnt pass Lacinia's spec in the latest version.
Latest commit a84f427 Oct 8, 2018



Umlaut is a schema language used to describe - in a neutral fashion - any data relationship domain.

Umlaut's unique design allows it to have several pluggable generators that can be used to generate target schema definitions in the language you want. The benefit here is that you write your schema once and consume it in as many places as needed having one single source of truth.

Table of Contents

Clojars Project

Getting Started

Option 1: As a binary

Simply run:

$ curl -sSL https://raw.githubusercontent.com/workco/umlaut/v0.6.0/scripts/install.sh | bash

Then you should be able to run umlaut normally as a binary:

$ umlaut

Once you get Umlaut installed, head to the usage section for more details. Do check recommended dependencies as well.

Option 2: As a Clojure dev util

If you are working on a Clojure project, Umlaut can be added to your deps.edn file:

   {umlaut {:mvn/version "x.y.z"}}}}}

Attention: replace x.y.z with the latest version.

Then use it as:

$ clojure -A:dev umlaut.main

Once you get Umlaut installed, head to the usage section for more details. Do check recommended dependencies as well.

Option 3: As a Clojure project library

There are situations where you would need to generate code as part of your Clojure project - i.e. reading an Umlaut file as part of your system's bootstrap and injecting a generated schema as part of initialization of a DB or any other system.

In these cases you need to use Umlaut as a normal dependency of your project. Add it to your project.clj or deps.edn file as usual. A deps.edn example:

 {umlaut {:mvn/version "x.y.z"}}

Attention: replace x.y.z with the latest version.

Recommended Dependencies

Umlaut can generate nice visual renderings of your schema by using GraphViz. We recommend installing Graphviz if you intend to use this feature of Umlaut.


Basic concept as a CLI

Depending on how you installed (see Getting Started above) you can either run Umlaut as an installed binary (umlaut) or as a Clojure app (clojure -A:dev umlaut.main). For brevity, all examples use the binary version.

Umlaut's CLI accepts parameters in tradition Unix fashion but also as an EDN payload. Each parameter can be specified as a keyword in a map.

For instance:

$ umlaut --gen dot --in path/to/in/umlaut/files --out path/to/out/diagrams

Is equivalent to:

$ umlaut --edn '{:gen dot :in "path/to/in/umlaut/files" :out "path/to/out/diagrams"}'

For more information, $ umlaut --help should do the trick.

Usage as a CLI

All files ending with .umlaut inside the input folder will be considered.

To create .dot (Graphviz) versions of your schema:

$ umlaut --gen dot --in [umlaut-files-folder] --out [output-folder]

If you have dot installed, Umlaut will also create .png renderings.

To create a GraphQL schema:

$ umlaut --gen graphql --in [umlaut-files-folder] --out [output-file]

To create a lacinia EDN schema:

$ umlaut --gen lacinia --in [umlaut-files-folder] --out [output-file]

To create clojure.spec files:

$ umlaut --gen spec --in [umlaut-files-folder] --out [output-folder] \
         --spec-package [spec-package] \
         --custom-validators-filepath [custom-validators-filepath] \
         --id-namespace [id-namespace]

For further details run $ umlaut --help

Usage as a Library

Every generator implements a gen method. You can either dig into the source code at umlaut/src/generators or check the documentation below for:

Schema Language

Umlaut expects to receive an umlaut schema file. An umlaut schema is general-purpose schema definition language that has been designed to provide support for generating models in several different languages in a flexible way.

Let's go over umlaut's syntax and characteristics.


The primitive types are: ["String" "Float" "Integer" "Boolean" "DateTime", "ID"]. If an attribute has a type that is not primitive, it must be properly declared.

If you try to declare an attribute that is not primitive or declared, an error will be thrown.

By default, all declared fields are non-null. Nullable fields must have the symbol ? after the attribute type.

Identifiers are validated by this regex: [_A-Za-z][_0-9A-Za-z]* and they can't be any of the reserved words listed on "List of reserved words" below.

Attributes and methods

Inside of types and interfaces you can declare fields.

A "field" can either be a method or an attribute:

A field without parameters is considered a simple attribute and is declared like this:

attributeIdentifier: type


name: String

To represent a method, simply declare a field with parameters, like this:

methodIdentifier(arg1Identifier: type, ..., arg2Identifier: type): type


weight(unit: UnitType): Float

All fields must have a type (primitive or not). You can add arity after the type to indicate a collection of items of that same type, like this:

attributeIdentifier: type[min..max]` or `methodIdentifier(...): type[min..max]

And n can be used to indicate inifinity.


employees: Employee[1..n]

Umlaut also accepts exact arity. Examples:

exactlyFourEmployees: Employee[4]
oneToFourRmployees: Employee[1..4]

You can use the optional modifier ? after anyone of the types to indicate a nullable field.


reportTo: Employee?

A few more examples compiled together:

  firstName: String
  middleName: String?
  lastName: String  // This is a comment in the schema source code
  dob: DateTime {
    @doc "This is a documentation comment in the dob field"
  age: Float
  friends: String[0..n]
  setFirstName(name: String): String
  computeAge(dateOfBirth: DateTime, today: DateTime): Integer

The type keyword

You can define new types by using the type reserved word.

type Person {
  id: ID
  name: String
  friends: String[0..n]
  isEmployed: Boolean
  email: String?


  • friends is an example of a collection.
  • email is a nullable attribute.
  • All attributes are primitives in this example.
  • All fields are mandatory except for email
  • The list of reserved words should be observed before naming a type.

The interface keyword

You can declare an interface, usually to create composition of models.

interface Person {
  id: ID
  name: String
  email: String?

interface Human {
  temperature: Float

type Employee : Person Human {
  salary: Float
  changeSalary(newSalary: Float): Float
  • umlaut supports single/multiple inheritance, in this example Employee inherits from Person and Human. So Employee will have all the fields of its parent types, plus its own.
  • An interface cannot inherit fields from other interfaces.
  • A type can only inherit fields from an interface.
  • Inheriting from multiple interfaces (multiple inheritance) is allowed.
  • The list of reserved words should be observed before naming an interface.

The enum keyword

You can create enums like this:

enum RelationshipStatus {

type Person {
  name: String
  status: RelationshipStatus
  • Enums can also be used to create GraphQL unions, in order to do that, you need to add an annotation: @lang/lacinia identifier union above the enum definition. In that case, the contents of the enum need to be other custom types.
  • The list of reserved words should be observed before naming an enum.


Annotations add extra metadata to your model and are either used for documentation purposes or can be used by the generators. Generators tend to use these annotations to know how you want your schema to be interpreted in that target environment.

Annotations have one of these two forms:

  • @<space> <key> [value]
  • @<space> [value]

The second format is used only for documentation purposes.

You can provide more than one value to the [value] in this syntax. However, this is only useful given a generator requirement.

In the example below we are telling the Lacinia generator (@lang/lacinia) that its identifier key for the type QueryRoot is query:

@lang/lacinia identifier query
type QueryRoot {
  friend: Person[0..n]

You can document your schema using the @doc space. This can be done above the type/interface/enum definition, like this...

@doc "This is my new type"
type MyType {

... or as a part of the field annotations like this...

@doc "This is my new type"
type MyType {
  myField: String {
    @doc "This is the comment of field a inside type MyType"

Diagram Generator

Umlaut uses Graphviz to create diagram images of your schema. Whenever the diagram generator is activated, at least a file all.png will be created containing all your schema models.

In order to build a specific diagram, you can use the diagram keyword in the umlaut file:

enum RelationshipStatus {

type Person {
  name: String
  status: RelationshipStatus

diagram filename {

The above example generates this diagram:


  • A file named all.png is always created with all the types defined in the umlaut document.
  • The file created is filename.png because of the identifier in the diagram keyword.
  • The diagram includes the RelationshipStatus box because of the ! used in the (Person) group. This tells umlaut to recursively draw all entities that are dependencies for the entity Person. If the ! was omitted, the diagram would have a single Person.
  • You can have as many diagrams definitions/combinations as you want, just give them different names.

You can specify the colors of a box this:

@lang/dot color <color>

The available colors are defined here: http://www.graphviz.org/doc/info/colors.html This annotation needs to be above the definition of a type/enum/interface.

To use this generator you must have graphviz installed. This package is available for Mac via Homebrew and Windows via direct download.

Lacinia Generator

Lacinia is a GraphQL framework for Clojure. It uses an EDN file with all the types, queries, inputs, mutations, subscriptions, and enums of your schema.

Umlaut will generate the EDN expected by Lacinia.

enum RelationshipStatus {

type Person {
  name: String?
  status: RelationshipStatus
  salary: Float {
    @doc "The annotation below assigns the resolve-salary resolver to the salary field"
    @lang/lacinia resolver resolve-salary

@lang/lacinia identifier query
type QueryRoot {
  friends: Person[0..n]

Will generate:

   {:name {:type String},
    :status {:type (non-null :RelationshipStatus)},
    :salary {:type (non-null Float), :resolver :resolve-salary}},
   :implements []},
  {:fields {:name {:type (non-null String)}}, :implements []}},
 :enums {:RelationshipStatus {:values ["SINGLE" "MARRIED"]}},
 :interfaces {},
 {:friends {:type (non-null (list (non-null :Person)))}}}

Lacinia-specific Annotations

The Lacinia generator uses annotations to describe certain expected behaviours that are specific to the use of schemas in Lacinia itself.

In the above example, we used @lang/lacinia identifier query to indicate that QueryRoot should be placed under the queries key. In practice this means that all fields within QueryRoot will be added to the root query of your GraphQL schema.

The following list describes the Lacinia-specific annotations that can be attached to types:

@lang/lacinia identifier query        // Qualifies the type as a GraphQL query root
@lang/lacinia identifier subscription // Qualifies the type as a GraphQL subscription root
@lang/lacinia identifier mutation     // Qualifies the type as a GraphQL mutations root
@lang/lacinia identifier input        // Qualifies the type as a GraphQL input objects

The following list describes the Lacinia-specific annotations that can be attached to fields:

@lang/lacinia resolver resolve-name // Tells Lacinia the field requires a `resolve-name` resolver
@lang/lacinia stream stream-name    // Tells Lacinia the field requires a stream called `stream-name`

Node that streams are used for subscriptions and resolvers used for queries, mutations, and fields. After generating lacinia's EDN, the resolvers and streams need to be attached using: com.walmartlabs.lacinia.util/attach-resolvers and com.walmartlabs.lacinia.util/attach-streamers respectively.


type Person {
  name: String {
    @doc "Name should hold the entire name, first name + last name"
    @lang/lacinia resolver resolve-name

@lang/lacinia identifier subscription
type SubscriptionRoot {
    getMessages(roomId: String): String[0..n] {
      @lang/lacinia stream stream-messages

@lang/lacinia identifier query
type QueryRoot {
    messages(roomId: String): String[0..n] {
      @lang/lacinia resolver resolve-messages

You can also add annotations at a global level like this, which for now are only used to create GraphQL unions:

annotations {
  @lang/lacinia union SearchResult Type1 Type2

Spec Generator

Umlaut can generate spec code based on your schema. This is a more complex generator that expects several parameters.

You can have a @lang/spec validator <name> above a type to add a custom validator function for that type. This custom validator function needs to be implemented in a common validator file.

Interfaces do not generate spec code, they are replaced by s/or of the types that implement that interface.

TBD Examples

Datomic Generator

Umlaut can generate Datomic schemas. In order to use it first require the generator in your code:

(require '[umlaut.generators.datomic :as datomic])

Then call function datomic/gen passing it the paths of all the umlaut files you want generated.

A simple example, the following umlaut file...

type Person {
  name: String 
  relationshipStatus: RelationshipStatus {
    @doc "Whether the person is married or single."

enum RelatiionshipStatus {

... will generate the following Datomic schema:

[{:db/ident :person/name
  :db/valueType :db.type/string
  :db/cardinality :db.cardinality/one}
 {:db/ident :person/relationship-status
  :db/valueType :db.type/ref
  :db/cardinality :db.cardinality/one
  :db/doc "Whether the person is married or single."}
 {:db/ident :relationship-status/single}
 {:db/ident :relationship-status/married}]

Uniqueness, identity and indexes:

In order to annotate any field for uniqueness or identity use the @lang/datomic value <...> form. Example email as an unique value:

type Person {
  email: String {
    @lang/datomic unique value

Example name as an unique identity:

type Person {
  email: String {
    @lang/datomic unique identity

Indexes can be achieved with the annotations @lang/datomic index true and @lang/datomic fulltext true.

Example name is indexed:

type Person {
  name: String {
    @lang/datomic index true

Example description is fulltext indexed:

type Person {
  name: String {
    @lang/datomic fulltext true

Excluding types from datomic generation

Types can be excluded from datomic generation with the annotation @lang/datomic skip true

@lang/datomic skip true
type Person {
  name: String {
    @lang/datomic fulltext true

Datomic Components

Non-primitive Umlaut types are translated to Datomic ref attributes. These can be tagged as components to facilitate transacting in nested maps, recursive retracts, etc.

type Person {
  address: Address {
    @lang/datomic isComponent true

Fine-grained types

Datomic has a much finer-grained set of scalar types than Umlaut so the annotation @lang/datomic precision <...> is used to give access to these types. The table below summarizes how to use it:

Datomic type Annotation Resulting Datomic type
String None :db.type/string
String @lang/datomic precision keyword :db.type/keyword
String @lang/datomic precision uri :db.type/uri
Float None :db.type/float
Float @lang/datomic precision double :db.type/double
Float @lang/datomic precision bigdec :db.type/bigdec
Integer None :db.type/long
Integer @lang/datomic precision bigint :db.type/bigint
Boolean None :db.type/boolean
DateTime None :db.type/instant
ID None :db.type/uuid

Limitations and caveats of the Datomic generator:

  1. :db.cardinality will be automatically set to :db.cardinality/many for any cardinality different than 1.
  2. Bidirecational relationships therefore are discouraged or you'll end up not knowing which entity owns what. The recommendation is to use Datomic's reverse lookup in queries (i.e. :person/_department from the department entity).
  3. :db/ident is created by converting everything to kebab casing (an opinionated decision).
  4. interfaces are not created as there's no such concept in Datomic. All the inherited fields though are properly created. Feel free use interfaces if you have too many repeated fields and want to avoid repetition.
  5. Fields that have parameters (potential methods in some languages?) are totally ignored (as in, not generated at all). We wouldn't know what to do with them from a Datomic standpoint.
  6. Optional and non-optional markers don't mean anything as Datomic doesn't capture this concept.

List of Reserved Words

These words are reserved for any of the supported generators and should not be used when defining a new type/interface/enum or field:

  • node
  • edge
  • graph
  • digraph
  • subgraph
  • strict

Umlaut Developers


If you notice something wrong and/or missing:

  • First check the issues. Chances are someone else has also been working on it
  • If you don't find a related issue, do create one.
  • Then, fork the repo. See notes below for more details.

Development notes

  • Clojure 1.9.0 or higher is a development dependency
  • GraalVM 1.0.0 RC 4 or higher is a dependency for building the native code
  • Maven 3.5.0 or higher is a dependency for building and deploying
  • Tests are run with $ clojure -A:test
  • deps.edn is your source of truth for project dependencies


  1. Update the pom.xml with the deps.edn dependencies by running $ clojure -Spom
  2. If major of minor version, change it on VERSION_TEMPLATE
  3. Otherwise, bump the patch with $ scripts/bump_version.sh
  4. Apply version with $ scrips/apply_version.sh
  5. Create a deployment tag $ git tag vX.Y.Z where X.Y.Z are what you have in VERSION_TEMPLATE
  6. Push the tag to Github $ git push --tags
  7. Deploy to Clojars with $ mvn deploy
  8. Create native binary with $ clojure -A:native-image --graalvm-home [graalvm-home-path]
  9. Upload binary target/umlaut as an asset to the release X.Y.Z on Github
  10. Publish X.Y.Z as a release on Github


If you find a bug, submit a Github issue.


This project is looking for team members who can help this project succeed! If you are interested in becoming a team member please open an issue.


Copyright © 2017 Work & Co

Distributed under the MIT License.