Dockerfile.slim

# ═══════════════════════════════════════════════════════════════════════════════
# Milady Agent — Slim Dockerfile
# ═══════════════════════════════════════════════════════════════════════════════
# Baseline: 4.02 GB (milady/agent:v2.0.0-alpha.89)
# Target:   < 2.0 GB (ideal)  |  < 1.5 GB (stretch)
#
# Strategy: 3-stage build
#   Stage 1 (builder) — identical to the current build process
#   Stage 2 (pruner)  — surgically deletes dev / build / GPU / browser packages
#   Stage 3 (runtime) — clean node:22-slim + pruned /app only
#
# ──────────────────────────────────────────────────────────────────────────────
# SAVINGS SUMMARY
# ──────────────────────────────────────────────────────────────────────────────
# TIER 1  (always applied, zero functionality lost)
#   node-llama-cpp GPU binaries      −684 MB  (CUDA, Vulkan, ARM; VPS has no GPU)
#   onnxruntime-web                  − 91 MB  (browser WASM build; server uses node)
#   @biomejs/biome CLI bins          −101 MB  (linter; devDep)
#   TypeScript compiler ×2           − 81 MB  (tsx uses esbuild, not tsc)
#   rolldown build binaries          − 41 MB  (bundler; only needed at build time)
#   Three.js ×3 + pixiv-VRM          −112 MB  (browser 3D; bundled in dist/ already)
#   lucide-react ×2                  − 87 MB  (browser icons; bundled in dist/)
#   @babylonjs/core                  − 67 MB  (browser 3D engine)
#   @mediapipe/pose                  − 51 MB  (browser CV; plugin-vision browser side)
#   musl native bins on glibc host   − 47 MB  (wrong-platform variants)
#   Storybook packages               − 38 MB  (UI devDep)
#   apps/app/public/ (duplicate)     −141 MB  (Vite copies to dist/ on build)
#   vitest + testing infra           − 12 MB  (devDeps)
#   ESLint + Babel compiler stack    − 20 MB  (devDeps)
#   source files + build configs     −  3 MB
#   .map + .d.ts + test/ in pkgs     − 60 MB  (estimate)
#   ─────────────────────────────────────────
#   TIER 1 TOTAL:                 ~ −1,636 MB
#   Estimated image after TIER 1: ~   2.38 GB
#
# TIER 2  (optional; uncomment sections to enable; disables ML features)
#   onnxruntime-node                 −208 MB  (disables local ONNX inference)
#   @tensorflow/* + tf-models        −200 MB  (disables plugin-vision TF path)
#   @huggingface/transformers        − 48 MB  (disables HF local models)
#   ─────────────────────────────────────────
#   TIER 2 ADDITIONAL:            ~ −  456 MB
#   Estimated image after T1+T2:  ~   1.92 GB  ← under 2 GB target ✓
#
# TIER 3  (aggressive; see IMAGE-OPTIMIZATION.md for stretch-goal options)
# ═══════════════════════════════════════════════════════════════════════════════

# ─── Stage 1: Builder ─────────────────────────────────────────────────────────
# Identical to the current alpha.89 build — do not change this stage.
FROM node:22-slim AS builder

RUN apt-get update && apt-get install -y --no-install-recommends \
    python3 make g++ git ca-certificates curl unzip \
    && rm -rf /var/lib/apt/lists/*

RUN curl -fsSL https://bun.sh/install | bash -s "bun-v1.3.9"
ENV PATH="/root/.bun/bin:$PATH"

WORKDIR /app
COPY . .

RUN bun install --frozen-lockfile --ignore-scripts 2>&1
RUN node ./scripts/link-browser-server.mjs 2>/dev/null || true && node ./scripts/patch-deps.mjs
RUN npx tsdown 2>&1 || echo "tsdown done"

# Patch @elizaos/agent version to match the milady wrapper version
RUN node -e "const fs=require('fs'); const v=require('./package.json').version; \
    try { const p=JSON.parse(fs.readFileSync('node_modules/@elizaos/agent/package.json','utf8')); \
    p.version=v; fs.writeFileSync('node_modules/@elizaos/agent/package.json',JSON.stringify(p,null,2)); \
    console.log('Patched autonomous version to', v); } catch(e) { console.log('skip:', e.message); }"

# ─── Stage 2: Pruner ──────────────────────────────────────────────────────────
# Removes dev / build / GPU / browser-only packages from the builder filesystem.
# The runtime stage does COPY --from=pruner which picks up the pruned state.
# Note: bun stores actual package files in node_modules/.bun/{pkg@version}/ and
# uses symlinks in node_modules/ → we delete both.
FROM builder AS pruner

# ╔═════════════════════════════════════════════════════════════════════════════╗
# ║  TIER 1 — Safe removals (no runtime functionality lost)                   ║
# ╚═════════════════════════════════════════════════════════════════════════════╝

# ── GPU platform binaries for node-llama-cpp ─────────────────────────────────
# CUDA ext / CUDA / Vulkan: require NVIDIA GPU or Vulkan-capable hardware.
# ARM64 / ARMv7l: wrong CPU arch (container is x86-64).
# node-llama-cpp auto-detects available backends and falls back to the plain
# linux-x64 CPU build when GPU variants are absent. Zero behaviour change.
# Savings: 444 + 153 + 76 + 5 + 6 = 684 MB
RUN rm -rf \
    node_modules/.bun/@node-llama-cpp+linux-x64-cuda-ext@* \
    node_modules/.bun/@node-llama-cpp+linux-x64-cuda@* \
    node_modules/.bun/@node-llama-cpp+linux-x64-vulkan@* \
    node_modules/.bun/@node-llama-cpp+linux-arm64@* \
    node_modules/.bun/@node-llama-cpp+linux-armv7l@*

# ── onnxruntime-web: browser-only WASM build ─────────────────────────────────
# The server runtime uses onnxruntime-node (native bindings).
# The web/WASM variant is loaded only in browser environments.
# Savings: 91 MB
RUN rm -rf node_modules/.bun/onnxruntime-web@*

# ── @biomejs/biome: linter (devDependency) ───────────────────────────────────
# Ships two heavy platform CLIs (gnu + musl). Never needed at runtime.
# Savings: 50 + 50 + ~1 = ~101 MB
RUN rm -rf \
    node_modules/.bun/@biomejs+cli-linux-x64@* \
    node_modules/.bun/@biomejs+cli-linux-x64-musl@* \
    node_modules/.bun/@biomejs+biome@*
RUN rm -rf node_modules/@biomejs

# ── TypeScript compiler ───────────────────────────────────────────────────────
# tsx (the runtime) uses bundled esbuild for transpilation; it does NOT import
# the typescript package at runtime. Both installed versions are safe to remove:
#   4.2.4 — old transitive dep brought in by some package
#   5.9.3 — devDependency
# Savings: 58 + 23 = 81 MB
RUN rm -rf \
    node_modules/.bun/typescript@4.2.4 \
    node_modules/.bun/typescript@5.9.3
RUN rm -f node_modules/typescript

# ── Rolldown build binaries ───────────────────────────────────────────────────
# Used by tsdown (the bundler) at build time only. Provides zero runtime value.
# Savings: 20 (gnu) + 21 (musl) = 41 MB
RUN rm -rf \
    node_modules/.bun/@rolldown+binding-linux-x64-gnu@* \
    node_modules/.bun/@rolldown+binding-linux-x64-musl@*

# ── Three.js (all 3 versions) + VRM + react-three/fiber ─────────────────────
# Three.js is a browser 3D rendering library. The Node.js server dist/entry.js
# has zero references to 'three', 'babylonjs', or 'lucide-react' (confirmed).
# All 3D functionality is compiled into apps/app/dist/ by Vite.
# Savings: 38 + 38 + 24 (three) + ~12 (@pixiv + @react-three) = ~112 MB
RUN rm -rf \
    node_modules/.bun/three@* \
    node_modules/.bun/@pixiv+three-vrm@* \
    node_modules/.bun/@pixiv+three-vrm-core@* \
    node_modules/.bun/@pixiv+three-vrm-materials-hdr-emissive-multiplier@* \
    node_modules/.bun/@pixiv+three-vrm-materials-mtoon@* \
    node_modules/.bun/@pixiv+three-vrm-materials-v0compat@* \
    node_modules/.bun/@pixiv+three-vrm-node-constraint@* \
    node_modules/.bun/@pixiv+three-vrm-springbone@* \
    node_modules/.bun/@pixiv+types-vrm-0.0@* \
    node_modules/.bun/@pixiv+types-vrmc-* \
    node_modules/.bun/@react-three+fiber@*
RUN rm -rf node_modules/three node_modules/@pixiv node_modules/@react-three

# ── lucide-react: browser icon library (2 versions) ──────────────────────────
# Both are bundled by Vite into apps/app/dist/. Not imported server-side.
# Savings: 45 + 42 = 87 MB
RUN rm -rf node_modules/.bun/lucide-react@*
RUN rm -f node_modules/lucide-react

# ── @babylonjs/core: browser 3D game engine ──────────────────────────────────
# Savings: 67 MB
RUN rm -rf node_modules/.bun/@babylonjs+* node_modules/.bun/@dimforge+rapier3d-compat@*
RUN rm -rf node_modules/@babylonjs node_modules/@dimforge

# ── @mediapipe/pose: browser pose-estimation model ───────────────────────────
# plugin-vision uses this for in-browser pose detection. The model weights and
# WASM are browser-side only. The server process does not load mediapipe.
# Savings: 51 MB
RUN rm -rf node_modules/.bun/@mediapipe+*
RUN rm -rf node_modules/@mediapipe

# ── glibc vs musl: remove wrong-platform native pre-builds ───────────────────
# node:22-slim is Debian (glibc). The musl/Alpine pre-built .node files will
# never be selected by the runtime. Remove them to reclaim space.
# If you switch to node:22-alpine, swap the patterns (delete gnu, keep musl).
# Savings: 30 (canvas-musl) + 17 (sharp-libvips-musl) = 47 MB
RUN rm -rf \
    node_modules/.bun/@napi-rs+canvas-linux-x64-musl@* \
    node_modules/.bun/@img+sharp-libvips-linuxmusl-x64@*
RUN rm -rf \
    node_modules/@napi-rs/canvas-linux-x64-musl \
    "node_modules/@img/sharp-libvips-linuxmusl-x64"

# ── Storybook: UI component workshop (devDep of apps/app) ────────────────────
# Savings: ~38 MB
RUN rm -rf \
    node_modules/.bun/@storybook+* \
    node_modules/.bun/storybook@*
RUN rm -rf node_modules/@storybook node_modules/storybook

# ── vitest + testing infrastructure ──────────────────────────────────────────
# Savings: ~12 MB
RUN rm -rf \
    node_modules/.bun/vitest@* \
    node_modules/.bun/@vitest+* \
    node_modules/.bun/react-test-renderer@* \
    node_modules/.bun/@testing-library+*
RUN rm -rf \
    node_modules/vitest \
    node_modules/@vitest \
    node_modules/react-test-renderer \
    node_modules/@testing-library

# ── ESLint + Babel compiler stack (devDeps / build-time only) ────────────────
# eslint: 4 MB. @babel/core, parser, types, traverse, generator: ~16 MB.
# We intentionally keep @babel/runtime (~1 MB) as some runtime packages import
# it for regenerator-runtime. Only the compiler toolchain is removed.
# Savings: ~20 MB
RUN rm -rf \
    node_modules/.bun/eslint@* \
    node_modules/.bun/@babel+core@* \
    node_modules/.bun/@babel+parser@* \
    node_modules/.bun/@babel+types@* \
    node_modules/.bun/@babel+traverse@* \
    node_modules/.bun/@babel+generator@* \
    node_modules/.bun/@babel+template@* \
    node_modules/.bun/@babel+helper-module-transforms@* \
    node_modules/.bun/@babel+helper-compilation-targets@* \
    node_modules/.bun/@babel+compat-data@* \
    node_modules/.bun/@typescript-eslint+* \
    node_modules/.bun/eslint-plugin-import@* \
    node_modules/.bun/eslint-config-prettier@*

# ── apps/app/public/: confirmed duplicate of apps/app/dist/ ──────────────────
# Vite copies the public/ directory into dist/ during every build.
# Verified: dist/vrms (72 MB) == public/vrms (72 MB)
#           dist/worlds (57 MB) == public/worlds (57 MB)
# The full public asset list (vrms, worlds, animations, vrm-decoders, favicons,
# og-image, logos) is present under dist/. Safe to drop the source copy.
# Savings: 141 MB
RUN rm -rf apps/app/public

# ── Source files not needed in the production runtime ────────────────────────
# root src/  = TypeScript source of the milady wrapper (compiled → dist/)
# test/      = unit tests
# apps/app/src  = Vite app source (compiled → apps/app/dist/)
# apps/app/electrobun = Electron build artifacts
# Unused sub-apps: homepage, home, landing, chrome-extension, ui
RUN rm -rf \
    src/ \
    test/ \
    coverage/ \
    dev.log \
    install.sh \
    install.ps1 \
    apps/app/src \
    apps/app/test \
    apps/app/public_src \
    apps/app/electrobun \
    apps/homepage \
    apps/home \
    apps/landing \
    apps/chrome-extension \
    apps/ui

# ── Build/dev config files ────────────────────────────────────────────────────
RUN rm -f \
    biome.json \
    vitest.config.ts \
    vitest.e2e.config.ts \
    vitest.live.config.ts \
    tsdown.config.ts \
    tsdown.electron.config.ts \
    tsdown.electron.config.test.ts \
    knip.jsonc

# ── Source maps from node_modules ────────────────────────────────────────────
# .map files are used by error stack-trace mappers in development. In production
# they just consume space. Estimated savings: 30–60 MB.
RUN find node_modules/.bun -name "*.map" -delete 2>/dev/null || true

# ── TypeScript declaration files from node_modules ───────────────────────────
# .d.ts files provide IDE type info. Node.js never reads them at runtime.
# Estimated savings: 20–40 MB.
RUN find node_modules/.bun -name "*.d.ts" -delete 2>/dev/null || true

# ── test/ directories inside third-party packages ────────────────────────────
RUN find node_modules/.bun -mindepth 3 -maxdepth 5 -type d \
    \( -name "test" -o -name "__tests__" -o -name "tests" -o -name "fixtures" \) \
    -exec rm -rf {} + 2>/dev/null || true

# ── Sweep dangling symlinks caused by the removals above ─────────────────────
# -xtype l matches symlinks whose targets no longer exist.
RUN find node_modules -maxdepth 4 -xtype l -delete 2>/dev/null || true


# ╔═════════════════════════════════════════════════════════════════════════════╗
# ║  TIER 2 — Optional ML stack removal (uncomment to enable)                 ║
# ║  Each block is independent; mix-and-match as needed.                      ║
# ╚═════════════════════════════════════════════════════════════════════════════╝

# ── onnxruntime-node: local ONNX model inference ─────────────────────────────
# Disables: plugin-local-embedding ONNX path (falls back to remote embeddings)
# Disables: any plugin that performs local inference via onnxruntime
# Savings: 208 MB
#
# RUN rm -rf \
#     node_modules/.bun/onnxruntime-node@* \
#     node_modules/.bun/onnxruntime-common@*
# RUN rm -rf node_modules/onnxruntime-node node_modules/onnxruntime-common

# ── @tensorflow: TensorFlow.js + backends ────────────────────────────────────
# Disables: plugin-vision TF inference path, TF-based embedding models
# Note: @tensorflow/tfjs-backend-webgl and webgpu are browser-only anyway;
# this block removes the CPU/WASM backends that do run server-side.
# Savings: ~200 MB
#
# RUN rm -rf \
#     node_modules/.bun/@tensorflow+tfjs@* \
#     node_modules/.bun/@tensorflow+tfjs-core@* \
#     node_modules/.bun/@tensorflow+tfjs-layers@* \
#     node_modules/.bun/@tensorflow+tfjs-converter@* \
#     node_modules/.bun/@tensorflow+tfjs-backend-cpu@* \
#     node_modules/.bun/@tensorflow+tfjs-backend-webgl@* \
#     node_modules/.bun/@tensorflow+tfjs-backend-webgpu@* \
#     node_modules/.bun/@tensorflow+tfjs-backend-wasm@* \
#     node_modules/.bun/@tensorflow+tfjs-data@* \
#     node_modules/.bun/@tensorflow+tfjs-node@* \
#     node_modules/.bun/@tensorflow-models+*
# RUN rm -rf node_modules/@tensorflow node_modules/@tensorflow-models

# ── @huggingface/transformers: local HuggingFace models ──────────────────────
# Disables: HF-backed local transformer / embedding models
# Savings: 48 MB
#
# RUN rm -rf node_modules/.bun/@huggingface+transformers@*
# RUN rm -rf node_modules/@huggingface


# ─── Stage 3: Runtime ─────────────────────────────────────────────────────────
FROM node:22-slim

RUN apt-get update && apt-get install -y --no-install-recommends \
    ca-certificates curl \
    && rm -rf /var/lib/apt/lists/*

# tsx is required: node --import tsx loads @elizaos/agent raw .ts files.
# Install globally so the node binary can find it via --import.
RUN npm install -g tsx

WORKDIR /app

# Copy the pruned application tree (only what survived Stage 2)
COPY --from=pruner /app /app

ENV NODE_ENV=production
ENV MILADY_PORT=2138
ENV MILADY_API_BIND=0.0.0.0

EXPOSE 2138

HEALTHCHECK --interval=30s --timeout=10s --start-period=60s --retries=3 \
    CMD curl -f http://localhost:${MILADY_PORT}/api/health || exit 1

CMD ["node", "--import", "tsx", "milady.mjs", "start"]