Building the Code Interpreter Image (mypy)#

Sherpa AIServer uses an isolated Docker container (mypy) as a sandbox to execute Python code. The image contains Python 3.11 and a set of scientific libraries (pandas, numpy, matplotlib, openpyxl, pillow, etc.) and runs under an unprivileged user.

All required files are located in the build/mypy/ directory of the repository.

Requirements#

• Docker (20.10+)
• Internet access to download dependencies (or prebuilt wheels — see below)
• Permissions to run Docker (or sudo)

Prebuilt image (download)#

If you do not need to build the image locally, a prebuilt mypy image can be downloaded together with other images — the download links are listed in the release documents.

Then load the image:

docker load -i mypy_<version>.tar.gz

Building the image#

1. Change to the build directory#

cd build/mypy

2. Build the image#

docker build -t mypy:latest .

3. Verify the image was created#

docker images | grep mypy

Expected output:

mypy   latest   <IMAGE_ID>   ...   ~500MB

Saving the image to an archive (for transfer)#

If you need to transfer the image to a target server without internet access:

# In the build/mypy/ directory
docker save mypy:latest -o mypy.tar

The resulting mypy.tar file can be copied to the server via SCP/SFTP and loaded there:

# On the target server — in build/mypy/ or next to mypy.tar
docker load -i mypy.tar

You can also use helper scripts:

# Build and save (developer machine)
bash build/mypy/mypy_build.sh

# Load the image (on the target server)
bash build/mypy/mypy_load.sh

Offline build (no PyPI access)#

If the target server has no internet access, prepare wheels on a machine with internet access:

# On a machine with internet
pip download -r build/mypy/requirements.txt -d build/mypy/wheels/

Then enable offline mode in the build/mypy/dockerfile by copying the wheels and using --find-links:

# Uncomment:
COPY wheels /wheels
RUN pip install --no-index --find-links=/wheels -r requirements.txt

# Comment out:
# RUN pip install -r requirements.txt

And rebuild:

docker build -t mypy:latest build/mypy/

Updating dependencies#

The package list is stored in build/mypy/requirements.txt. To add a new library:

1. Add the package and version to requirements.txt
2. Rebuild the image: docker build -t mypy:latest build/mypy/
3. Restart the code interpreter service in compose: docker compose restart aiserver-code-interpreter

Service integration#

After building the image it is used by the aiserver-code-interpreter service in docker-compose.yml. Ensure the image name in compose matches the tag (mypy:latest).