Skip to content


Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time

SAFE - A web stack designed for developer happiness

The following document describes the SAFE-Stack sample project. SAFE is a technology stack that brings together several technologies into a single, coherent stack for typesafe, flexible end-to-end web-enabled applications that are written entirely in F#.


You can see it running on Microsoft Azure at

Build status Build Status


If you are looking to create your own SAFE application, we recommend that you use the official SAFE template which provides a clean, flexible and regularly updated template dotnet designed for starting brand-new applications quickly and easily.


Development mode

This development stack is designed to be used with minimal tooling. An instance of Visual Studio Code together with the excellent Ionide plugin should be enough.

Start the development mode with:

> ./build.cmd run // on windows
$ ./ run // on unix

This command will call the target "Run" in build.fsx. This will start in parallel:

  • dotnet fable webpack-dev-server in src/Client (note: the Webpack development server will serve files on http://localhost:8089)
  • dotnet watch msbuild /t:TestAndRun in test/serverTests to run unit tests and then server (note: Giraffe is launched on port 8085)

You can now edit files in src/Server or src/Client and recompile + browser refresh will be triggered automatically. For the case of the client "Hot Module Replacement" is supported, which means your app state is kept over recompile.

Development mode

Usually you can just keep this mode running and running. Just edit files, see the browser refreshing and commit + push with git.

Getting started

Create a new page

This topic will guide you through creating a new page. After every section you should check whether you can see the changes in your browser.

Minimal setup

Let's say we want to call our new page Tomato

  1. Adjust the src/Client/Pages.fs and register our Tomato page as a page and define the corresponding path.

    type Page =
        | Home
        | Login
        | WishList
        | Tomato // <- our page
    let toPath =
        | Page.Home -> "/"
        | Page.Login -> "/login"
        | Page.WishList -> "/wishlist"
        | Page.Tomato -> "/tomato" // <- our page
    let pageParser : Parser<Page->_,_> =
            [ map Page.Home (s "home")
              map Page.Login (s "login")
              map Page.WishList (s "wishlist")
              map Page.Tomato (s "tomato") ] // <- our page
  2. Adjust the model, update and view:

    • PageModel type inside src/Client/Shared.fs
      type PageModel =
          | TomatoModel
    • view function inside src/Client/Shared.fs
      let view model dispatch =
          div [ Key "Application" ] [
              div [ centerStyle "column" ] [
                  match model.PageModel with
                  | TomatoModel ->
                      yield div [] [ words 60 "Tomatos taste good"]
  3. Adjust the urlUpdate function inside src/Client/App.fs:

    • urlUpdate function
      let urlUpdate (result:Page option) model =
          match result with
          | Some Page.Tomato ->
              { model with PageModel = TomatoModel }, Cmd.none
  4. Fix up incomplete pattern matches in the hydrateModel function inside src/Client/App.fs:

    • hydrateModel function
      let hydrateModel (json:string) (page: Page option) : Model * Cmd<_> =
          match page, model.PageModel with
          | Some Page.Tomato, TomatoModel -> model, Cmd.none
          | _, HomePageModel
          | _, TomatoModel ->
              // unknown page or page does not match model -> go to home page
              { User = None; PageModel = HomePageModel }, Cmd.none
  5. Try it out by navigating to http://localhost:8089/tomato

You should see Tomatoes taste good!

Adding the page to the menu

Inside src/Client/views/Menu.fs:

let inline private clientView onLogout (model:Model) =
    div [ centerStyle "row" ] [
        yield viewLink Page.Tomato "Tomato"

Move code to separate Tomato.fs files

  1. Add a new .fs file to the pages folder: src/Client/pages/Tomato.fs. Add the src/Client/pages/Tomato.fs to your Client.fsproj and Server.fsproj files and move it above Shared.fs.

    • Client.fsproj
    <Compile Include="pages/Login.fs" />
    <Compile Include="pages/Tomato.fs" /> <!-- <- our page -->
    <Compile Include="Shared.fs" />
    • Server.fsproj
    <Compile Include="../Client/pages/Login.fs" />
    <Compile Include="../Client/pages/Tomato.fs" />  <!-- <- our page -->
    <Compile Include="../Client/Shared.fs" />
  2. Place following code in the src/Client/pages/Tomato.fs:

    module Client.Tomato
    open Client.Styles
    let view() =
        words 60 "Tomatoes taste VERY good!"
  3. Remove old 'view' code from the view function in src/Client/Shared.fs and replace it with:

    | TomatoModel ->
        Tomato.view ()

Define a model for the page that holds the state

  1. Replace the code in src/Client/pages/Tomato.fs with

    module Client.Tomato
    open Client.Styles
    open Fable.React
    type Model = {
    let init() =
        { Color = "red" }
    let view model =
        div []
                div [] [words 60 "Tomatoes taste VERY good!"]
                div [] [words 20 (sprintf "The color of a tomato is %s" model.Color)]
  2. Adjust the PageModel discriminated union in Shared.fs

    type PageModel =
        | HomePageModel
        | LoginModel of Login.Model
        | WishListModel of WishList.Model
        | TomatoModel of Tomato.Model
  3. urlUpdate should now call the init function and set Tomato.Model as the new PageModel

    | Some Page.Tomato ->
        let m = Tomato.init()
        { model with PageModel = TomatoModel m }, Cmd.none
  4. hydrateModel needs to explicitly ignore the payload of the TomatoModel case in its pattern match

        | _, TomatoModel _ ->
  5. The view function in Shared.fs should call the view function of the the Tomato module and pass in the page model if it is a TomatoModel

    | TomatoModel m ->
        yield Tomato.view m

Make it interactive (update the state)

  1. Add a new message DU in src/Client/pages/Tomato.fs

    type Msg =
        | ChangeColor of string
  2. Add a message to the Msg DU in src/Client/Shared.fs

    type Msg =
        | TomatoMsg of Tomato.Msg
  3. Adjust the match pattern in the update function of src/Client/App.fs

    | TomatoMsg msg, TomatoModel tm ->
        let color = match msg with Tomato.Msg.ChangeColor c -> c
        let tm = { tm with Color = color }
        { model with PageModel = TomatoModel tm }, Cmd.none
    | TomatoMsg msg, _ -> model, Cmd.none // in case we receive a delayed message originating from the previous page
  4. Change the Tomato.view function (and open another namespace):

    open Fable.React.Props
    let view model dispatch =
        div []
                words 60 "Tomatoes taste VERY good!"
                words 20 (sprintf "The color of a tomato is %s" model.Color)
                br []
                button [
                    ClassName ("btn btn-primary")
                    OnClick (fun _ -> dispatch (ChangeColor "green"))]
                    [ str "No, my tomatoes are green!" ]
  5. Edit Shared.view and pass the dispatch function to Tomato.view, remapping Tomato.Msg onto App.Msg

    | TomatoModel m ->
        yield Tomato.view m (TomatoMsg >> dispatch)


The server side of the application can be debugged using Ionide.

  1. Run build.cmd \ to restore everything properly
  2. Open repo in VSCode
  3. Open debug panel, choose Debug from combobox, and press green arrow (or F5). This will build server and start it with debugger attached. It will also start Fable watch mode for the Client side and open the browser.

Client side debugging is supported by any modern browser with any developer tools. Fable even provides source maps which will let you put breakpoints in F# source code (in browser dev tools). Also, we additionally suggest installing React-devtools (for better UI debugging) and Redux-devtools (time travel debugger).

Technology stack

Saturn on .NET Core

The webserver backend is using Saturn - a web framework building on top of Giraffe and ASP.NET Core.

In development mode the server is automatically restarted whenever a file in src/Server is saved.

Freya on .NET Core

The SAFE stack does not force you to use Giraffe/Saturn. Check out the freya branch for an alternative implementation of the backend.

If you are new to Freya, you can find an introduction at:

React/Elmish client

The client is React single page application that uses fable-elmish.

The communication to the server is done via HTTPS calls to /api/*. If a user is logged in then a JSON Web Token is sent to the server with every request.


The Fable compiler is used to compile the F# client code to JavaScript so that it can run in the browser.

Shared code between server and client

"Isomorphic F#" started a bit as a joke about Isomorphic JavaScript. The naming is really bad, but the idea to have the same code running on client and server is really interesting. If you look at src/Server/Shared/Domain.fs then you will see code that is shared between client and server. On the server it is compiled to .NET core and for the client the Fable compiler is translating it into JavaScript. This is a really convenient technique for a shared domain model.

Server-Side Rendering

This sample uses Server-Side Rendering (SSR) with fable-react. This means the starting page is rendered on the ASP.NET Core server and sent as HTML to the client. This allows for better Search Engine Optimization and gives faster initial response, especially on mobile devices. Everything else is then rendered via React on the client.

More info can be found in the SSR tutorial.


Start the full build (incl. UI tests) with:

> build.cmd // on windows
$ ./ // on unix


Expecto is a test framework like NUnit or xUnit, but much more developer friendly. With Expecto you write tests as values in normal code. Tests can be composed, reduced, filtered, repeated and passed as values, because they are values. This gives the programmer a lot of leverage when writing tests.

If you are in development mode then you can use Expecto's focused test feature to run a selected test against the running server.


canopy is a F# web automation and testing library, built on top of Selenium. In our expecto suite it looks like the following:

testCase "login with test user" <| fun () ->
    url serverUrl
    waitForElement ".elmish-app"

    click "Login"

    "#username" << "test"
    "#password" << "test"

    click "Log In"

    waitForElement "Isaac Abraham"

Canopy in action

Additional tools


FAKE is a build automation system with capabilities which are similar to make and rake. It's used to automate the build, test and deployment process. Look into build.fsx for details.


Paket is a dependency manager and allows easier management of the NuGet packages.


The deployment for this repo works via docker with Linux containers and therefore you need docker installed on your machine.

Microsoft Azure

The following part shows how to set up automatic deployment to Microsoft Azure.

Auto Deployment to Azure

Docker Hub

Create a new Docker Hub account and a new public repository on Docker Hub.

Release script

Create a file called release.cmd with the following content and configure your DockerHub credentials:

@echo off

.paket\paket.exe restore
if errorlevel 1 (
  exit /b %errorlevel%

packages\build\FAKE\tools\FAKE.exe build.fsx Deploy "" "DockerImageName=****" "DockerUser=****" "DockerPassword=***" %*

Don't worry the file is already in .gitignore so your password will not be commited.

Initial docker push

In order to release a container you need to create a new entry in [] and run release.cmd. This will build the server and client, run all test, put the app into a docker container and push it to your docker hub repo.

Azure Portal

Go to the Azure Portal and create a new "Web App for Containers". Configure the Web App to point to the docker repo and select latest channel of the container.

Docker setup

Also look for the "WebHook Url" on the portal, copy that url and set it as new trigger in your Docker Hub repo.

Note that entering a Startup File is not necessary.

The Dockerfile used to create the docker image exposes port 8085 for the Giraffe server application. This port needs to be mapped to port 80 within the Azure App Service for the application to receive http traffic.

Presently this can only be done using the Azure CLI. You can do this easily in Azure Cloud Shell (accessible from the Azure Portal in the top menu bar) using the following command:

az webapp config appsettings set --resource-group <resource group name> --name <web app name> --settings WEBSITES_PORT=8085

The above command is effectively the same as running docker run -p 80:8085 <image name>.

Now you should be able to reach the website on your url.

Further releases

Now everything is set up. By creating new entries in [] and a new run of release.cmd the website should update automatically.

Azure Storage

With the steps above the website is only using local file storage. If you want to use it together with Azure Storage, then go back to the Azure Portal and create a new "Storage account". Copy the Connection String from "Access keys" tab and move over to your Azure app service.

Storage Account

Google Cloud AppEngine

The repository comes with a sample app.yaml file which is used to deploy to Google Cloud AppEngine using the custom flex environment. At the moment it seems like the application must run on port 8089 and that is set as a environment variable in the app.yaml file. When you run the deploy command it will first look for the app.yaml file and then look for the Dockerfile for what should deploy. The container that is deploy is exactly that same as the one that should have been deployed to Azure, but it is only set up to deploy from local to Google Cloud at the moment, and not from CI server to Google Cloud.

Before you can execute the deploy command you also need to build the solution with the Publish target, that is so the container image will get the compiled binaries when the container build is executed via the deploy command.

To execute the deploy you need a Google Cloud account and project configured as well as the tooling installed, The command you need to run is:

gcloud app deploy -q <--version=main> <--verbosity=debug>

The version and verbosity flag isn't need, but it is recommended to use the version flag so you don't end up with too many versions of your application without removing previous ones. Use verbosity=debug if you are having some problems.

Deploy to the flex environment with a custom runtime like this is might take some time, but the instructions here should work.

Workaround for Split Health Checks

Newly created projects have Split Health Checks enabled by default, which causes the deployment to fail. This can be resolved by disabling them on the project.

First install the beta components if you don't already have them:

gcloud components install beta

then run the following command on your project:

gcloud beta app update --no-split-health-checks --project <YOUR PROJECT ID>

After that, deploying as described above should work just fine.

Known Issues

Getting rid of errors in chrome

  • Either comment out the lines in App.fs:
|> Program.withDebugger
  • Or install the Redux DevTools as a Chrome Extensions (recommended) Only one error remains, when visiting the WebApp the first time.