Skip to content

README.md

Latest commit

 

History

History
122 lines (88 loc) · 3.46 KB

README.md

File metadata and controls

122 lines (88 loc) · 3.46 KB

OpenClaw Launcher Docker Images

Slim Docker images that install OpenClaw directly via npm, bypassing the bloated 4GB upstream image. Multi-stage build compiles native modules once, then copies only runtime artifacts to a minimal base.

Flavors

Flavor Target What's included Image size
slim slim OpenClaw + jq, ripgrep, fd, sqlite3 ~300 MB
base base slim + node-llama-cpp (local LLM support) ~500 MB
full full base + Python 3, Playwright + Chromium, ffmpeg ~1.2 GB

Recommended: Use base for future-proof local LLM support, or slim if you only use cloud APIs and want the smallest possible image.

How it works

Instead of extending the 4GB upstream image, we:

  1. Builder stage: Install build tools (cmake, clang) and compile native modules

    • openclaw (~56MB npm package)
    • node-llama-cpp (~30MB, requires compilation)
    • @napi-rs/canvas (prebuilt binaries)

  2. Runtime stage: Copy only compiled modules to node:22-bookworm-slim

    • No compilers, no dev headers, no build cache
    • Just the runtime libs needed by native modules

Result: 70% smaller than upstream (500MB vs 4GB).

Tool inventory

slim

  • jq — JSON processor (API responses, config manipulation)
  • ripgrep (rg) — fast recursive search
  • fd — fast file finder (aliased from fd-find)
  • sqlite3 — embedded SQL database
  • tini — proper PID 1 signal handling

base (includes slim)

  • node-llama-cpp — local LLM inference (CPU)
  • libgomp1 — OpenMP runtime for parallel processing

full (includes base)

  • Python 3 with pip
  • pandas — data analysis
  • matplotlib — charts and plots
  • Pillow — image processing
  • ffmpeg — video/audio/image conversion
  • Playwright + Chromium — headless browser automation

Building locally

All flavors are build targets in a single docker/Dockerfile. Use the Makefile:

cd docker

make help              # Show all targets

make build             # Build base (default)
make build FLAVOR=slim # Build slim
make build-all         # Build all flavors

make verify            # Verify base tools
make verify-all        # Verify all flavors

make shell             # Shell into base
make shell FLAVOR=full # Shell into full

make run               # Run base (gateway on localhost:18789)
make run FLAVOR=slim   # Run slim

make clean             # Remove all local images

Raw docker commands

docker build --target slim -f docker/Dockerfile -t openclaw-launcher:slim .
docker build --target base -f docker/Dockerfile -t openclaw-launcher:base .
docker build --target full -f docker/Dockerfile -t openclaw-launcher:full .

Check image sizes

docker images openclaw-launcher

Using a flavor

# Shell launcher
OPENCLAW_FLAVOR=base ./scripts/run.sh

# Or set in environment
export OPENCLAW_FLAVOR=slim
./scripts/run.sh

Verification

# Test gateway starts
docker run --rm -p 18789:18789 openclaw-launcher:base
# Visit http://localhost:18789

# Test local LLM module loads (base/full only)
docker run --rm openclaw-launcher:base node -e "require('node-llama-cpp')"

Multi-arch

make build-multiarch          # Validate multi-arch build (no load)
make build-multiarch-all      # Validate all flavors
make push FLAVOR=base         # Build + push to GHCR (requires login)
make push-all                 # Push all flavors