We recommend using the Nix package manager as it makes any pain related to maintaining dependency versions just disappear. Once nix has been installed the development environment can be manually instantiated through the nix-shell command:

$ cd ~/code/coder

$ nix-shell

copying path '/nix/store/3ms6cs5210n8vfb5a7jkdvzrzdagqzbp-iana-etc-20210225' from ''...
copying path '/nix/store/dxg5aijpyy36clz05wjsyk90gqcdzbam-iana-etc-20220520' from ''...
copying path '/nix/store/v2gvj8whv241nj4lzha3flq8pnllcmvv-ignore-5.2.0.tgz' from ''...

If direnv is installed and the hooks are configured then the development environment can be automatically instantiated by creating the following .envrc, thus removing the need to run nix-shell by hand!

$ cd ~/code/coder
$ echo "use nix" >.envrc
$ direnv allow

Now, whenever you enter the project folder, direnv will prepare the environment for you:

$ cd ~/code/coder

direnv: loading ~/code/coder/.envrc
direnv: using nix
direnv: export +AR +AS +CC +CONFIG_SHELL +CXX +HOST_PATH +IN_NIX_SHELL +LD +NIX_BINTOOLS +NIX_BINTOOLS_WRAPPER_TARGET_HOST_x86_64_unknown_linux_gnu +NIX_BUILD_CORES +NIX_BUILD_TOP +NIX_CC +NIX_CC_WRAPPER_TARGET_HOST_x86_64_unknown_linux_gnu +NIX_CFLAGS_COMPILE +NIX_ENFORCE_NO_NATIVE +NIX_HARDENING_ENABLE +NIX_INDENT_MAKE +NIX_LDFLAGS +NIX_STORE +NM +NODE_PATH +OBJCOPY +OBJDUMP +RANLIB +READELF +SIZE +SOURCE_DATE_EPOCH +STRINGS +STRIP +TEMP +TEMPDIR +TMP +TMPDIR +XDG_DATA_DIRS +buildInputs +buildPhase +builder +cmakeFlags +configureFlags +depsBuildBuild +depsBuildBuildPropagated +depsBuildTarget +depsBuildTargetPropagated +depsHostHost +depsHostHostPropagated +depsTargetTarget +depsTargetTargetPropagated +doCheck +doInstallCheck +mesonFlags +name +nativeBuildInputs +out +outputs +patches +phases +propagatedBuildInputs +propagatedNativeBuildInputs +shell +shellHook +stdenv +strictDeps +system ~PATH


Alternatively if you do not want to use nix then you'll need to install the need the following tools by hand:

  • Go 1.18+
    • on macOS, run brew install go
  • Node 14+
    • on macOS, run brew install node
  • GNU Make 4.0+
    • on macOS, run brew install make
  • shfmt
    • on macOS, run brew install shfmt
  • nfpm
    • on macOS, run brew install goreleaser/tap/nfpm && brew install nfpm
  • pg_dump
    • on macOS, run brew install libpq zstd
    • on Linux, install zstd
  • pkg-config
    • on macOS, run brew install pkg-config
  • pixman
    • on macOS, run brew install pixman
  • cairo
    • on macOS, run brew install cairo
  • pango
    • on macOS, run brew install pango
  • pandoc
    • on macOS, run brew install pandocomatic

Development workflow

Use the following make commands and scripts in development:

  • ./scripts/ runs the frontend and backend development server
  • make build compiles binaries and release packages
  • make install installs binaries to $GOPATH/bin
  • make test

Adding database migrations and fixtures

Database migrations

Database migrations are managed with migrate.

To add new migrations, use the following command:

$ ./coderd/database/migrations/ my name
Run "make gen" to generate models.

Then write queries into the generated .up.sql and .down.sql files and commit them into the repository. The down script should make a best-effort to retain as much data as possible.

Database fixtures (for testing migrations)

There are two types of fixtures that are used to test that migrations don't break existing Coder deployments:

Both types behave like database migrations (they also migrate). Their behavior mirrors Coder migrations such that when migration number 000022 is applied, fixture 000022 is applied afterwards.

Partial fixtures are used to conveniently add data to newly created tables so that we can ensure that this data is migrated without issue.

Full database dumps are for testing the migration of fully-fledged Coder deployments. These are usually done for a specific version of Coder and are often fixed in time. A full database dump may be necessary when testing the migration of multiple features or complex configurations.

To add a new partial fixture, run the following command:

$ ./coderd/database/migrations/ my fixture

Then add some queries to insert data and commit the file to the repo. See 000024_example.up.sql for an example.

To create a full dump, run a fully fledged Coder deployment and use it to generate data in the database. Then shut down the deployment and take a snapshot of the database.

$ mkdir -p coderd/database/migrations/testdata/full_dumps/v0.12.2 && cd $_
$ pg_dump "postgres://[email protected]:..." -a --inserts >000069_dump_v0.12.2.up.sql

Make sure sensitive data in the dump is desensitized, for instance names, emails, OAuth tokens and other secrets. Then commit the dump to the project.

To find out what the latest migration for a version of Coder is, use the following command:

$ git ls-files v0.12.2 -- coderd/database/migrations/*.up.sql

This helps in naming the dump (e.g. 000069 above).



Our style guide for authoring documentation can be found here.


Use Go style

Contributions must adhere to the guidelines outlined in Effective Go. We prefer linting rules over documenting styles (run ours with make lint); humans are error-prone!

Read Go's Code Review Comments Wiki for information on common comments made during reviews of Go code.

Avoid unused packages

Coder writes packages that are used during implementation. It isn't easy to validate whether an abstraction is valid until it's checked against an implementation. This results in a larger changeset, but it provides reviewers with a holistic perspective regarding the contribution.


Our frontend guide can be found here.


The following information has been borrowed from Go's review philosophy.

Coder values thorough reviews. For each review comment that you receive, please "close" it by implementing the suggestion or providing an explanation on why the suggestion isn't the best option. Be sure to do this for each comment; you can click Done to indicate that you've implemented the suggestion, or you can add a comment explaining why you aren't implementing the suggestion (or what you chose to implement instead).

It is perfectly normal for changes to go through several rounds of reviews, with one or more reviewers making new comments every time, then waiting for an updated change before reviewing again. All contributors, including those from maintainers, are subject to the same review cycle; this process is not meant to be applied selectively or to discourage anyone from contributing.


Coder releases are initiated via ./scripts/ and automated via GitHub Actions. Specifically, the release.yaml workflow. They are created based on the current main branch.

The release notes for a release are automatically generated from commit titles and metadata from PRs that are merged into main.

Creating a release

The creation of a release is initiated via ./scripts/ This script will show a preview of the release that will be created, and if you choose to continue, create and push the tag which will trigger the creation of the release via GitHub Actions.

See ./scripts/ --help for more information.

Creating a release (via workflow dispatch)

Typically the workflow dispatch is only used to test (dry-run) a release, meaning no actual release will take place. The workflow can be dispatched manually from Actions: Release. Simply press "Run workflow" and choose dry-run.

If a release has failed after the tag has been created and pushed, it can be retried by again, pressing "Run workflow", changing "Use workflow from" from "Branch: main" to "Tag: vX.X.X" and not selecting dry-run.

Commit messages

Commit messages should follow the Conventional Commits 1.0.0 specification.

Allowed commit types (feat, fix, etc.) are listed in conventional-commit-types. Note that these types are also used to automatically sort and organize the release notes.

A good commit message title uses the imperative, present tense and is ~50 characters long (no more than 72).


  • Good: feat(api): Add feature X
  • Bad: feat(api): Added feature X (past tense)

A good rule of thumb for writing good commit messages is to recite: If applied, this commit will ....

Note: We lint PR titles to ensure they follow the Conventional Commits specification, however, it's still possible to merge PRs on GitHub with a badly formatted title. Take care when merging single-commit PRs as GitHub may prefer to use the original commit title instead of the PR title.

Breaking changes

Breaking changes can be triggered in two ways:

  • Add ! to the commit message title, e.g. feat(api)!: Remove deprecated endpoint /test
  • Add the release/breaking label to a PR that has, or will be, merged into main.


The security label can be added to PRs that have, or will be, merged into main. Doing so will make sure the change stands out in the release notes.

Was this page helpful? Share your experience with us.


See an opportunity to improve our docs? Make an edit.

A tool that provisions remote development environments via Terraform
On this page