version: "1.0.0" name: build-images description: Building Capsem VM images with capsem-builder. Use when working with guest image configuration, Dockerfiles, kernel builds, rootfs builds, the builder CLI, or guest config TOML files. Covers the config-driven build system, guest config layout, Dockerfile templates, multi-arch support, the builder CLI commands, AND the internal architecture for modifying the builder itself (models, context flow, template variables, adding install managers).

Building VM Images

Overview

capsem-builder is a config-driven build system. It reads TOML configs from guest/config/, renders Jinja2 Dockerfile templates, and builds kernel + rootfs via Docker. Assets output to assets/{arch}/.

Guest config layout

guest/config/
  build.toml              Architectures, compression, base images
  manifest.toml           Image name, version, changelog
  ai/*.toml               AI provider configs (Claude, Gemini, Codex)
  packages/*.toml         Package sets (apt, python)
  mcp/*.toml              MCP server configs
  security/web.toml       Web security (allow/block domains)
  vm/resources.toml       CPU, RAM, disk
  vm/environment.toml     Shell, TLS, env vars
  kernel/*.defconfig      Kernel defconfigs per architecture

All configs use Pydantic models for validation. Run uv run capsem-builder validate guest/ to lint.

CLI commands

uv run capsem-builder doctor guest/          # Check build prerequisites
uv run capsem-builder validate guest/        # Lint all configs (E001-E302, W001-W012)
uv run capsem-builder build guest/ --dry-run # Preview rendered Dockerfiles
uv run capsem-builder build guest/ --arch arm64 --template rootfs  # Build rootfs
uv run capsem-builder build guest/ --arch arm64 --template kernel  # Build kernel
uv run capsem-builder inspect guest/         # Show config summary
uv run capsem-builder new my-image/ --from guest/  # Scaffold new image from base
uv run capsem-builder audit                  # Parse trivy/grype vulnerability output

Building assets

Full rebuild (kernel + rootfs):

just build-assets    # Runs doctor + validate + build for host arch

Individual templates:

just build-kernel arm64
just build-rootfs arm64

Per-arch asset layout

assets/
  manifest.json          Version, checksums, asset list
  B3SUMS                 BLAKE3 checksums
  arm64/
    vmlinuz              Kernel
    rootfs.squashfs      Root filesystem
    initrd.img           Initial ramdisk (repacked by just run)

Rootfs squashfs settings live under [build] in guest/config/build.toml. The current default is compression = "zstd", compression_level = 15, and squashfs_block_size = "128K", balancing sequential rootfs reads, CLI startup, image size, and small-file reads.

Adding packages to the VM

1. Edit the appropriate config in guest/config/packages/ (apt or python TOML)
2. Run uv run capsem-builder validate guest/ to check
3. Run just build-assets to rebuild the rootfs
4. Verify: just run "capsem-doctor"

Do not edit Dockerfiles directly -- they are rendered from Jinja2 templates in src/capsem/builder/templates/.

Adding a new AI provider

1. Create guest/config/ai/<provider>.toml with provider config
2. Add domain entries to guest/config/security/web.toml if needed
3. Validate: uv run capsem-builder validate guest/
4. Rebuild: just build-assets

Dockerfile templates

Templates live in src/capsem/builder/templates/:

• Dockerfile.rootfs.j2 -- rootfs image (apt packages, Python packages, AI CLIs, diagnostics)
• Dockerfile.kernel.j2 -- kernel build (defconfig, modules, vmlinuz extraction)

Templates use Jinja2 with variables from the merged guest config. Preview with --dry-run.

Builder Internals (for modifying the builder itself)

Architecture: TOML -> Pydantic -> context dict -> Jinja2 -> Dockerfile

The data flows through four layers:

1. TOML configs (guest/config/) -- user-facing, declarative
2. Pydantic models (src/capsem/builder/models.py) -- validation + types
3. Context dict (src/capsem/builder/docker.py) -- template variables
4. Jinja2 templates (src/capsem/builder/templates/) -- Dockerfile output

Key files

Context dict (rootfs template variables)

_rootfs_context() in docker.py builds the dict passed to Dockerfile.rootfs.j2:

{
    "arch": ArchConfig,           # Per-arch settings (docker_platform, rust_target, etc.)
    "arch_name": str,             # "arm64" or "x86_64"
    "apt_packages": list[str],    # From packages/apt.toml
    "python_packages": list[str], # From packages/python.toml
    "python_install_cmd": str,    # e.g. "uv pip install --system --break-system-packages"
    "npm_packages": list[str],    # From ai/*.toml where install.manager == "npm"
    "npm_prefix": str,            # e.g. "/opt/ai-clis"
    "curl_installs": list[str],   # From ai/*.toml where install.manager == "curl"
    "guest_binaries": list[str],  # ["capsem-pty-agent", "capsem-net-proxy", "capsem-mcp-server"]
}

Kernel context dict

{
    "arch": ArchConfig,
    "arch_name": str,
    "kernel_version": str,  # e.g. "6.6.130"
}

How to: Add a new install manager

Example: adding a curl manager so a CLI can be installed via curl | bash instead of npm.

Step 1: Add enum value to PackageManager

In src/capsem/builder/models.py:

class PackageManager(str, Enum):
    APT = "apt"
    UV = "uv"
    PIP = "pip"
    NPM = "npm"
    CURL = "curl"  # <-- new

Step 2: Collect packages in _rootfs_context()

In src/capsem/builder/docker.py, add a new list and populate it from providers:

curl_installs: list[str] = []
for provider in config.ai_providers.values():
    if provider.enabled and provider.install:
        if provider.install.manager == PackageManager.CURL:
            curl_installs.extend(provider.install.packages)

Add "curl_installs": curl_installs to the returned dict.

Step 3: Add template block

In src/capsem/builder/templates/Dockerfile.rootfs.j2:

{% for url in curl_installs %}
# CLI installed via installer script
RUN curl -fsSL {{ url }} | bash
{% endfor %}

Step 4: Add to scaffold

In src/capsem/builder/scaffold.py, add to _INSTALL_CMDS:

"curl": "curl -fsSL",

Step 5: Update the TOML config

In guest/config/ai/<provider>.toml:

[provider.install]
manager = "curl"
packages = ["https://example.com/install.sh"]

Step 6: Update tests

• tests/test_docker.py -- context dict assertions (what's in npm_packages vs curl_installs)
• tests/test_cli.py -- Dockerfile rendering assertions (corporate config tests)

How to: Change how an AI CLI is installed

1. Edit guest/config/ai/<provider>.toml -- change [provider.install] section
2. If changing install manager type, may need to update _rootfs_context() in docker.py
3. Check extract_tool_versions() in docker.py -- it hardcodes version-check paths
4. Update tests in test_docker.py and test_cli.py
5. Rebuild: just build-assets && just run "capsem-doctor"

How to: Add a new package to an existing set

1. Edit guest/config/packages/apt.toml or guest/config/packages/python.toml
2. Add the package name to the packages list
3. Validate: uv run capsem-builder validate guest/
4. Rebuild: just build-assets

How to: Add a new guest binary

Guest binaries are compiled from crates/capsem-agent/. On macOS, cross_compile_agent() delegates to container_compile_agent() which builds inside a Linux container (docker). On Linux (CI), cargo builds natively.

1. Add the binary target in crates/capsem-agent/Cargo.toml
2. Add the binary name to GUEST_BINARIES list in docker.py
3. The template already loops {% for binary in guest_binaries %} to COPY + chmod 555

Verifying Linux builds locally

just cross-compile [arch] builds everything in a container: agent binaries, frontend, and the full Tauri app (deb + AppImage). Useful for catching linuxdeploy and system dep issues before CI.

just cross-compile           # Build for host arch (arm64 on Apple Silicon)
just cross-compile x86_64    # Build x86_64 deb + AppImage

AI provider TOML schema

[provider_key]
name = "Provider Name"
description = "What this provider does"
enabled = true  # false to exclude from build
 
[provider_key.cli]
key = "cli-binary-name"      # e.g. "claude", "gemini", "codex"
name = "CLI Display Name"
 
[provider_key.api_key]
name = "API Key Name"
env_vars = ["ENV_VAR_NAME"]   # At least one required
prefix = "sk-"                # Key prefix for validation
docs_url = "https://..."
 
[provider_key.network]
domains = ["*.example.com"]   # At least one required
allow_get = true
allow_post = true
 
[provider_key.install]
manager = "npm"               # "npm", "curl", "apt", "uv", "pip"
prefix = "/opt/ai-clis"       # Install prefix (npm only)
packages = ["@scope/package"] # Package names or URLs
 
[provider_key.files.some_config]
path = "/root/.config/file.json"
content = '{"key": "value"}'

Build pipeline (what build_image() does)

For rootfs:

1. Build guest agent binaries (cross_compile_agent -- on macOS delegates to container_compile_agent which builds inside a Linux container; on Linux compiles natively)
2. Assemble build context (prepare_build_context) -- copies CA cert, shell configs, diagnostics, agent binaries
3. Render Dockerfile from template
4. docker build
5. Export container filesystem as tar
6. Create squashfs from tar (create_squashfs -- runs mksquashfs in a container)
7. Extract tool versions (extract_tool_versions)
8. Clean up container image

For kernel:

1. Resolve latest kernel version from kernel.org
2. Assemble build context (defconfig, capsem-init)
3. Render Dockerfile from template
4. docker build
5. Extract vmlinuz + initrd.img from image
6. Clean up

Container runtime requirements

On macOS, Docker runs inside a Colima VM with limited resources. The rootfs build runs apt, npm, and curl-based CLI installers concurrently -- the default RAM allocation may cause OOM kills (exit code 137).

Minimum: 12GB RAM. Recommended: 16GB RAM, 8 CPUs.

# Colima (macOS)
colima stop && colima start --vm-type vz --vz-rosetta --memory 16 --cpu 8
 
# Linux: Docker runs natively, no memory tuning needed
# sudo apt install docker.io

just doctor and capsem-builder doctor both check these resources automatically.

The resource check lives in src/capsem/builder/doctor.py:

• check_container_resources() -- checks docker info
• Thresholds: DOCKER_MIN_MEMORY_MB = 4096, DOCKER_RECOMMENDED_MEMORY_MB = 8192

Container image compatibility

The container builds use rust:slim-bookworm -- a minimal Debian image. Many common utilities (file, less, vim, etc.) are NOT available. Any shell commands run inside the container must use only coreutils (ls, cp, cat, test, etc.) or tools explicitly installed via apt-get in the same RUN step.

Lesson learned: using file /output/binary to verify compiled binaries failed because file is not in slim images. Replaced with ls -l which is always available and still confirms the copy succeeded. The real validation (existence + non-zero size) is done in Python after the container exits.

Rule: never assume a command exists in a slim container image. Stick to coreutils or install what you need explicitly.

Clock skew workaround

All apt-get update calls use -o Acquire::Check-Valid-Until=false to handle container VM clock drift. Without this, apt rejects Release files whose timestamp is in the future relative to the VM's clock. This can occur with any container VM backend on macOS.

Files affected:

• Dockerfile.kernel.j2 (line 11)
• Dockerfile.rootfs.j2 (line 11)
• docker.py create_squashfs() function