sg - the Sourcegraph developer tool

          _____                    _____
         /\    \                  /\    \
        /::\    \                /::\    \
       /::::\    \              /::::\    \
      /::::::\    \            /::::::\    \
     /:::/\:::\    \          /:::/\:::\    \
    /:::/__\:::\    \        /:::/  \:::\    \
    \:::\   \:::\    \      /:::/    \:::\    \
  ___\:::\   \:::\    \    /:::/    / \:::\    \
 /\   \:::\   \:::\    \  /:::/    /   \:::\ ___\
/::\   \:::\   \:::\____\/:::/____/  ___\:::|    |
\:::\   \:::\   \::/    /\:::\    \ /\  /:::|____|
 \:::\   \:::\   \/____/  \:::\    /::\ \::/    /
  \:::\   \:::\    \       \:::\   \:::\ \/____/
   \:::\   \:::\____\       \:::\   \:::\____\
    \:::\  /:::/    /        \:::\  /:::/    /
     \:::\/:::/    /          \:::\/:::/    /
      \::::::/    /            \::::::/    /
       \::::/    /              \::::/    /
        \::/    /                \::/____/

sg is the CLI tool that Sourcegraph developers can use to develop Sourcegraph. Learn more about the tool's overall vision in sg Vision, and how to use it in the usage section.


  1. Run the following to download and install sg:

    curl --proto '=https' --tlsv1.2 -sSLf | sh
  2. In your clone of sourcegraph/sourcegraph, start the default Sourcegraph environment:

    sg start
  3. Once the enterprise-web process has finished compilation, open https://sourcegraph.test:3443 in your browser.

A more detailed introduction is available in the development quickstart guide.


Run the following command in a terminal:

curl --proto '=https' --tlsv1.2 -sSLf | sh

That will download the latest release of sg from here, put it in a temporary location and run sg install to install it to a permanent location in your $PATH.

For other installation options, see Advanced installation.


Once set up, sg will automatically check for updates and update itself if a change is detected in your local copy of origin/main. To force a manual update of sg, run:

sg update

In order to temporarily turn off automatic updates, run your commands with the -skip-auto-update flag or SG_SKIP_AUTO_UPDATE environment variable:

sg -skip-auto-update [cmds ...]

On the next command run, if a new version is detected, sg will auto update before running.

To see what's changed, use sg version changelog.


You can get help about commands locally in a variety of ways:

sg help # show all available commands

# learn about a specific command or subcommand
sg <command> -h
sg <command> --help

sg help -full # full reference


If you have used sg setup, you should have autocompletions set up for sg. To enable it, type out a partial command and press the Tab key twice. For example:

sg start<tab><tab>

To get autocompletions for the available flags for a command, type out a command and - and press the Tab key twice. For example:

sg start -<tab><tab>

Both of the above work if you provide partial values as well to narrow down the suggestions. For example, the following will suggest run sets that start with web-:

sg start web-<tab><tab>


Default sg behaviour is configured through the sg.config.yaml file in the root of the sourcegraph/sourcegraph repository. Take a look at that file to see which commands are run in which environment, how these commands set setup, what environment variables they use, and more.

To modify your configuration locally, you can overwrite chunks of configuration by creating a sg.config.overwrite.yaml file in the root of the repository. It's .gitignored so you won't accidentally commit those changes.

If an sg.config.overwrite.yaml file exists, its contents will be merged with the content of sg.config.yaml, overwriting where there are conflicts. This is useful for running custom command sets or adding environment variables specific to your work.

You can run sg run debug-env to see the environment variables passed sg's child processes.

Configuration examples

Changing database configuration

In order to change the default database configuration, the username and the database, for example, create an sg.config.overwrite.yaml file that looks like this:

  PGUSER: 'mrnugget'
  PGDATABASE: 'my-database'

That works for all the other env variables in sg.config.yaml too.

Defining a custom environment by setting a commandset

You can customize what boots up in your development environment by defining a commandSet in your sg.config.overwrite.yaml.

For example, the following defines a commandset called minimal-batches that boots up a minimal environment to work on Batch Changes:

      - docker
      - redis
      - postgres
      - enterprise-frontend
      - enterprise-worker
      - enterprise-repo-updater
      - enterprise-web
      - gitserver
      - searcher
      - symbols
      - caddy
      - zoekt-indexserver-0
      - zoekt-indexserver-1
      - zoekt-webserver-0
      - zoekt-webserver-1
      - batches-executor-firecracker

With that in sg.config.overwrite.yaml you can now run sg start minimal-batches.

Run gitserver in a Docker container

sg start runs many of the services (defined in the commands section of sg.config.yaml) as binaries that it compiles and runs according to the settings in their cmd and install sections. Sometimes while developing, you need to run some of the services isolated from your local environment. This example shows what to add to sg.config.overwrite.yaml so that gitserver will run in a Docker container. The gitserver service already has a build script that generates a Docker image; this configuration will use that script in the install section, and use the env defined in sg.config.yaml to pass environment variables to docker in the run section.

A few things to note about this configuration

  • PGHOST is set to host.docker.internal so that gitserver running in the container can connect to the database that's running on your local machine. See the Docker documentation for more information about host.docker.internal. In order to use host.docker.internal here, you will need to add it to /etc/hosts so that the services not running in Docker containers will be able to use it also. If you are using a different database, you will need to adjust PGHOST to suit.
  • ${SRC_FRONTEND_INTERNAL##*:}, ${HOSTNAME##*:}, and ${SRC_PROF_HTTP##*:} use shell parameter expansion to pull the port number from the environment variables defined in sg.config.yaml. This parameter expansion works in at least the ksh, bash and zsh shells; might not work in others.
  • The gitserver Docker containers will be left running after sg has terminated. You will need to manually stop them.
  • The Prometheus agent gets metrics about the /data/repos mount. Because the Docker operating system is Linux, the metric function uses the sysfs pseudo filesystem and assumes that /data/repos is on a block device. However, Docker bind mounts (which is what -v ${SRC_REPOS_DIR}:/data/repos creates) create virtual filesystems, not block filesystems. This will result in error messages in the container about "skipping metric registration" with a reason containing "failed to evaluate sysfs symlink". You can ignore those errors unless your development work involves the Prometheus metrics, in which case you will need to create Docker volumes and mount those instead of the bind mounts (Docker volumes are mounted as block devices). Using a Docker volume means that the repo dirs will not be on your local filesystem in the same location as SRC_REPOS_DIR.
  # MUST ADD ENTRY TO /etc/hosts: host.docker.internal
  PGHOST: host.docker.internal
    install: |
      [[ $(uname) == "Darwin" ]] && config="--config darwin-docker"
      bazel build //cmd/gitserver:image_tarball ${config} && \
      docker load --input $(bazel cquery //cmd/gitserver:image_tarball ${config} --output=files)      
    cmd: |
      docker inspect gitserver-${GITSERVER_INDEX} >/dev/null 2>&1 && docker stop gitserver-${GITSERVER_INDEX}
      docker run \
      --rm \
      -e "GITSERVER_ADDR=${HOSTNAME##*:}" \
      -e "SRC_FRONTEND_INTERNAL=host.docker.internal:${SRC_FRONTEND_INTERNAL##*:}" \
      -e "SRC_PROF_HTTP=${SRC_PROF_HTTP##*:}" \
      -e "HOSTNAME=${HOSTNAME}" \
      -p ${GITSERVER_ADDR}:${HOSTNAME##*:} \
      -p ${SRC_PROF_HTTP}:${SRC_PROF_HTTP##*:} \
      -v ${SRC_REPOS_DIR}:/data/repos \
      --detach \
      --name gitserver-${GITSERVER_INDEX} \
    cmd: |
      docker inspect gitserver-${GITSERVER_INDEX} >/dev/null 2>&1 && docker stop gitserver-${GITSERVER_INDEX}
      docker run \
      --rm \
      -e "GITSERVER_ADDR=${HOSTNAME##*:}" \
      -e "SRC_FRONTEND_INTERNAL=host.docker.internal:${SRC_FRONTEND_INTERNAL##*:}" \
      -e "SRC_PROF_HTTP=${SRC_PROF_HTTP##*:}" \
      -e "HOSTNAME=${HOSTNAME}" \
      -p ${GITSERVER_ADDR}:${HOSTNAME##*:} \
      -p ${SRC_PROF_HTTP}:${SRC_PROF_HTTP##*:} \
      -v ${SRC_REPOS_DIR}:/data/repos \
      --detach \
      --name gitserver-${GITSERVER_INDEX} \

Attach a debugger

To attach the Delve debugger, pass the environment variable DELVE=true into sg. Read more here

Offline development

Sometimes you will want to develop Sourcegraph but it just so happens you will be on a plane or a train or perhaps a beach, and you will have no WiFi. And you may raise your fist toward heaven and say something like, "Why, we can put a man on the moon, so why can't we develop high-quality code search without an Internet connection?" But lower your hand back to your keyboard and fret no further, you can develop Sourcegraph with no connectivity by setting the OFFLINE environment variable:

OFFLINE=true sg start

Ensure that the sourcegraph/syntax-highlighter:insiders image is already available locally. If not, pull it with the following command before going offline to ensure that offline mode works seamlessly:

docker pull -q sourcegraph/syntax-highlighter:insiders

sg and pre-commit hooks

When sg setup is run, it will automatically install pre-commit hooks (using, with a provided configuration that will perform a series of fast checks before each commit you create locally.

Amongst that list of checks, is a script that tries to detect the presence of tokens that would have been accidentally committed. While it's implementation is rather simple and won't catch all tokens (this is covered by automated scans in CI), it's enough to catch common mistakes and save you from having to rotate secrets, as they never left your computer. Due to the importance of such a measure, it's an opt-out process instead of opt-in.

Therefore, it's strongly recommended to keep the pre-commit git hook. In the eventuality of the pre-commit detecting a false positive, you can disable it through sg setup disable-pre-commit and prevent sg setup from installing it by passing a flag sg setup --skip-pre-commit.


There are legitimate cases where code contains what appears to be a Sourcegraph token but isn't usable on any existing deployments. Testing code for generating tokens is good example.

You can tell pre-commit to simply skip these files by adding a // pre-commit:ignore_sourcegraph_token top-level comment, as shown in the example below:

package accesstoken

// pre-commit:ignore_sourcegraph_token

import (

func TestGenerateDotcomUserGatewayAccessToken(t *testing.T) {
	type args struct {
		apiToken string
	tests := []struct {
		name    string
		args    args
		want    string
		wantErr bool
			name:    "valid token 1",
			args:    args{apiToken: "0123456789abcdef0123456789abcdef01234567"},
			want:    "LOOKS_LIKE_A_REAL_TOKEN",
			wantErr: false,
  // (...)

Contributing to sg

Want to hack on sg? Great! Here's how:

  1. Read through the sg Vision to get an idea of what sg should be in the long term.
  2. Explore the sg source code.
  3. Look at the open sg issues.

When you want to hack on sg it's best to be in the dev/sg directory and run it from there:

cd dev/sg
go run . -config ../../sg.config.yaml start

The -config can be anything you want, of course.

Have questions or need help? Feel free to open a discussion! Sourcegraph teammates can also leave a message in #dev-experience.

Advanced installation

Dockerized sg

A sourcegraph/sg Docker image is available:

# ...
COPY --from /usr/local/bin/sg ./sg
# ...

Manually building the binary

If you want full control over where the sg binary ends up, use this option.

In the root of sourcegraph/sourcegraph, run:

go build -o ~/my/path/sg ./dev/sg

Then make sure that ~/my/path is in your $PATH.