diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 2ae103b..b00b867 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -70,6 +70,9 @@ jobs:
       run: |
         myst build --html
 
+    - name: Build local PyPI warehouse for piplite
+      run: python build_pypi.py
+
     - name: Upload build output
       uses: actions/upload-pages-artifact@v3
       with:
diff --git a/build_pypi.py b/build_pypi.py
new file mode 100644
index 0000000..1544629
--- /dev/null
+++ b/build_pypi.py
@@ -0,0 +1,142 @@
+#!/usr/bin/env python3
+"""Download wheels and build a local PyPI index for piplite.
+
+Creates a pypi/ directory with wheel files and an all.json index that
+piplite uses to resolve packages locally, eliminating 404 warnings when
+the JupyterLite kernel starts on the live site.
+
+Usage:
+    python build_pypi.py [output_dir]
+
+    output_dir defaults to lectures/_build/html/pypi
+"""
+
+import datetime
+import hashlib
+import json
+import re
+import subprocess
+import sys
+import tempfile
+import zipfile
+from pathlib import Path
+
+# Packages required by active lectures (see lectures/myst.yml TOC)
+# long_run_growth.md:      openpyxl
+# inflation_history.md:    xlrd, openpyxl
+# french_rev.md:           openpyxl, requests
+# markov_chains_I.md:      quantecon_wasm
+# markov_chains_II.md:     quantecon_wasm
+# eigen_II.md:             quantecon_wasm
+# networks.md:             quantecon_wasm, quantecon-book-networks, pandas-datareader
+PACKAGES = [
+    "openpyxl",
+    "requests",
+    "xlrd",
+    "quantecon_wasm",
+    "quantecon-book-networks",
+    "pandas-datareader",
+]
+
+
+def download_wheels(packages, dest_dir):
+    """Download pure-Python wheels for the given packages (no deps)."""
+    subprocess.check_call([
+        sys.executable, "-m", "pip", "download",
+        "--no-deps",
+        "--only-binary=:all:",
+        "--dest", str(dest_dir),
+        *packages,
+    ])
+
+
+def get_wheel_metadata(whl_path):
+    """Extract name, version, and requires_python from a wheel's METADATA."""
+    with zipfile.ZipFile(whl_path) as zf:
+        for entry in zf.namelist():
+            if entry.endswith("/METADATA"):
+                text = zf.read(entry).decode("utf-8")
+                info = {"requires_python": None}
+                for line in text.splitlines():
+                    if line.startswith("Name: "):
+                        info["name"] = line[6:].strip()
+                    elif line.startswith("Version: "):
+                        info["version"] = line[9:].strip()
+                    elif line.startswith("Requires-Python: "):
+                        info["requires_python"] = line[17:].strip()
+                    elif line == "":
+                        break  # end of headers
+                return info
+    raise ValueError(f"No METADATA found in {whl_path}")
+
+
+def build_index(wheel_dir):
+    """Build a piplite-compatible all.json index from wheels in a directory."""
+    all_json = {}
+    now = (
+        datetime.datetime.now(tz=datetime.timezone.utc)
+        .isoformat()
+        .split("+")[0] + "Z"
+    )
+
+    for whl_path in sorted(Path(wheel_dir).glob("*.whl")):
+        meta = get_wheel_metadata(whl_path)
+        whl_bytes = whl_path.read_bytes()
+        whl_sha256 = hashlib.sha256(whl_bytes).hexdigest()
+        whl_md5 = hashlib.md5(whl_bytes).hexdigest()
+
+        # PEP 503 normalized name
+        normalized = re.sub(r"[-_.]+", "-", meta["name"]).lower()
+
+        release = {
+            "comment_text": "",
+            "digests": {"sha256": whl_sha256, "md5": whl_md5},
+            "downloads": -1,
+            "filename": whl_path.name,
+            "has_sig": False,
+            "md5_digest": whl_md5,
+            "packagetype": "bdist_wheel",
+            "python_version": "py3",
+            "requires_python": meta["requires_python"],
+            "size": whl_path.stat().st_size,
+            "upload_time": now,
+            "upload_time_iso_8601": now,
+            "url": f"./{whl_path.name}",
+            "yanked": False,
+            "yanked_reason": None,
+        }
+
+        if normalized not in all_json:
+            all_json[normalized] = {"releases": {}}
+        all_json[normalized]["releases"][meta["version"]] = [release]
+
+    return all_json
+
+
+def main():
+    output_dir = (
+        Path(sys.argv[1]) if len(sys.argv) > 1
+        else Path("lectures/_build/html/pypi")
+    )
+    output_dir.mkdir(parents=True, exist_ok=True)
+
+    with tempfile.TemporaryDirectory() as tmpdir:
+        print(f"Downloading wheels: {', '.join(PACKAGES)}")
+        download_wheels(PACKAGES, tmpdir)
+
+        for whl in sorted(Path(tmpdir).glob("*.whl")):
+            dest = output_dir / whl.name
+            dest.write_bytes(whl.read_bytes())
+            print(f"  {whl.name} ({whl.stat().st_size:,} bytes)")
+
+    index = build_index(output_dir)
+    index_path = output_dir / "all.json"
+    index_path.write_text(json.dumps(index, indent=2, sort_keys=True))
+
+    total = sum(f.stat().st_size for f in output_dir.glob("*.whl"))
+    count = sum(len(v["releases"]) for v in index.values())
+    print(f"\nBuilt {index_path} with {count} packages ({total:,} bytes)")
+
+
+if __name__ == "__main__":
+    main()