Building Docker Images¶

Build and distribute custom Docker images for Agentomics-ML.

Quick Build (Multi-Arch + Push)¶

Use the provided build script to build and push multi-architecture images:

./build.sh <dockerhub-username> [version]

This script always pushes to the registry and requires docker login first.

Build Script Usage¶

# Build and push to Docker Hub (latest)
./build.sh myusername

# Build and push a specific version
./build.sh myusername v1.0

Images¶

Two Docker images are built:

Image	Purpose	Dockerfile
`agentomics`	Main agent execution	`Dockerfile`
`agentomics-prepare`	Dataset preparation	`Dockerfile.prepare`

Manual Build (Local Only)¶

Main Image¶

docker build -t agentomics_img .

Preparation Image¶

docker build -f Dockerfile.prepare -t agentomics_prepare_img .

Multi-Architecture Builds¶

For production deployments supporting both Intel/AMD and ARM processors.

Prerequisites¶

Enable Docker's multi-platform builder:

# Create and use a multi-platform builder
docker buildx create --name multiplatform --use
docker buildx inspect --bootstrap

Build Both Architectures¶

Main agent image:

docker buildx build \
  --platform linux/amd64,linux/arm64 \
  -t myusername/agentomics:latest \
  --push .

Preparation image:

docker buildx build \
  --platform linux/amd64,linux/arm64 \
  -f Dockerfile.prepare \
  -t myusername/agentomics-prepare:latest \
  --push .

With Version Tags¶

docker buildx build \
  --platform linux/amd64,linux/arm64 \
  -t myusername/agentomics:v1.0 \
  -t myusername/agentomics:latest \
  --push .

Architecture-Specific Builds¶

Build for specific architectures only:

# ARM64 only (Apple Silicon, ARM servers)
docker buildx build \
  --platform linux/arm64 \
  -t myusername/agentomics:arm64 \
  --push .

# AMD64 only (Intel/AMD processors)
docker buildx build \
  --platform linux/amd64 \
  -t myusername/agentomics:amd64 \
  --push .

Verification¶

Check that images support multiple architectures:

docker buildx imagetools inspect myusername/agentomics:latest

Expected output shows both: - linux/amd64 - linux/arm64

Build with Proxy¶

If behind a corporate proxy:

docker build \
  --build-arg HTTP_PROXY=$HTTP_PROXY \
  --build-arg HTTPS_PROXY=$HTTPS_PROXY \
  --build-arg http_proxy=$http_proxy \
  --build-arg https_proxy=$https_proxy \
  -t agentomics_img .

If you need proxy settings for multi-arch builds, configure the Docker daemon proxy or use manual docker buildx commands with --build-arg.

Image Contents¶

Main Image (Dockerfile)¶

The multi-stage build includes:

Stage 1: Create conda environment (agentomics-env)
Stage 2: Pre-download foundation models
Stage 3: Pack environment with conda-pack
Final: Runtime image with entry point

Key contents: - Python 3.12 with ML libraries - PyTorch, scikit-learn, transformers - pydantic-ai for agent framework - Weave and W&B for logging

Preparation Image (Dockerfile.prepare)¶

Lightweight image for dataset preparation: - Python 3.10 - pandas, rich - Minimal dependencies

Customization¶

Adding Dependencies¶

Edit environment.yaml and rebuild:

dependencies:
  - your-new-package==1.0.0

Modifying Entry Point¶

The entry point is /repository/src/run_agent_interactive.py.

To change, edit the ENTRYPOINT in the Dockerfile.

Storage Requirements¶

Image	Size (approx)
agentomics	8-12 GB
agentomics-prepare	1-2 GB

Includes conda environments and pre-downloaded models.

Pushing to Registry¶

Docker Hub¶

docker login
docker push myusername/agentomics:latest

Other Registries¶

# AWS ECR
docker push 123456789.dkr.ecr.us-east-1.amazonaws.com/agentomics:latest

# Google Container Registry
docker push gcr.io/my-project/agentomics:latest

Troubleshooting¶

Build fails with memory error¶

Increase Docker memory limit or use --memory flag:

docker build --memory=8g -t agentomics .

Build takes too long¶

Use BuildKit for faster builds:

DOCKER_BUILDKIT=1 docker build -t agentomics .

Platform mismatch¶

Ensure you're building for the correct platform:

docker build --platform linux/amd64 -t agentomics .