diff --git a/.forgejo/workflows/ci.yml b/.forgejo/workflows/ci.yml
index 557f119..0d145c3 100644
--- a/.forgejo/workflows/ci.yml
+++ b/.forgejo/workflows/ci.yml
@@ -121,6 +121,9 @@ jobs:
           echo "value=$TAGS" >> $GITHUB_OUTPUT
           echo "build_version=$BUILD_VERSION" >> $GITHUB_OUTPUT
 
+      - name: Free disk space
+        run: docker system prune -f || true
+
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v4
 
@@ -138,3 +141,5 @@ jobs:
           push: true
           tags: ${{ steps.tags.outputs.value }}
           build-args: BUILD_VERSION=${{ steps.tags.outputs.build_version }}
+          cache-from: type=registry,ref=${{ env.IMAGE }}:cache
+          cache-to: type=registry,ref=${{ env.IMAGE }}:cache,mode=max
diff --git a/Dockerfile b/Dockerfile
index 1d5d203..a2cd616 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -1,3 +1,4 @@
+# syntax=docker/dockerfile:1
 # Stage 1: Build Vue frontend
 FROM node:22-alpine AS build-frontend
 WORKDIR /build
@@ -12,16 +13,20 @@ WORKDIR /app
 
 COPY pyproject.toml .
 COPY src/ src/
-RUN pip install --no-cache-dir .
+RUN --mount=type=cache,target=/root/.cache/pip \
+    pip install .
 # Voice dependencies (faster-whisper, Kokoro TTS, soundfile) — activated at runtime via VOICE_ENABLED
-RUN pip install --no-cache-dir faster-whisper kokoro soundfile
+# Cache mount keeps wheels across builds so the heavy torch/CUDA deps aren't re-downloaded every time.
+RUN --mount=type=cache,target=/root/.cache/pip \
+    pip install faster-whisper kokoro soundfile
 
 # Build the fable-mcp wheel so it can be served for download
 COPY fable-mcp/ fable-mcp/
-RUN pip install --no-cache-dir build hatchling \
+RUN --mount=type=cache,target=/root/.cache/pip \
+    pip install build hatchling \
     && python -m build --wheel ./fable-mcp --outdir /app/dist/ \
     && pip uninstall -y build \
-    && rm -rf fable-mcp/ /root/.cache/pip
+    && rm -rf fable-mcp/
 
 COPY --from=build-frontend /build/dist/ src/fabledassistant/static/
 COPY alembic.ini .