diff --git a/cookbook b/cookbook
new file mode 100755
index 000000000..71f32b822
--- /dev/null
+++ b/cookbook
@@ -0,0 +1,47 @@
+set -euo pipefail
+
+usage() {
+  cat <<'EOF'
+usage:
+  cookbook test [path/to/notebook.ipynb] [--ai-compare] [--timeout=900] [--kernel=python3]
+
+notes:
+  - If no notebook path is provided, it runs ALL notebooks.
+  - --ai-compare sets AI_COMPARE=1 (uses your GOOGLE_API_KEY).
+  - Outputs go to reports/*.compare.json
+EOF
+}
+
+cmd="${1:-}"; shift || true
+case "$cmd" in
+  test)
+    NB_ARG=""
+    AI_COMPARE="${AI_COMPARE:-0}"
+    TIMEOUT="900"
+    KERNEL="python3"
+
+    for a in "$@"; do
+      case "$a" in
+        --ai-compare) AI_COMPARE=1 ;;
+        --timeout=*) TIMEOUT="${a#*=}" ;;
+        --kernel=*) KERNEL="${a#*=}" ;;
+        *.ipynb) NB_ARG="$a" ;;
+      esac
+    done
+
+    if [[ -n "$NB_ARG" ]]; then export NB="$NB_ARG"; fi
+    export AI_COMPARE="$AI_COMPARE"
+    export NB_TIMEOUT="$TIMEOUT"
+    export NB_KERNEL="$KERNEL"
+
+    exec python tests/test_nbclient.py
+    ;;
+  ""|-h|--help|help)
+    usage
+    ;;
+  *)
+    echo "Unknown subcommand: $cmd" >&2
+    usage
+    exit 2
+    ;;
+esac
\ No newline at end of file
diff --git a/tests/test_nbclient.py b/tests/test_nbclient.py
new file mode 100644
index 000000000..36a31be44
--- /dev/null
+++ b/tests/test_nbclient.py
@@ -0,0 +1,318 @@
+# Copyright 2025 Google LLC
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os, re, json, time, pathlib, subprocess, nbformat, requests, sys
+from nbclient import NotebookClient
+from nbclient.exceptions import CellExecutionError
+
+ROOT = pathlib.Path(__file__).resolve().parents[1]
+REPORTS = ROOT / "reports"
+REPORTS.mkdir(parents=True, exist_ok=True)
+
+def _discover():
+    nb = os.getenv("NB")
+    if nb:
+        nb_list = re.split(r"[,\s]+", nb)    
+        paths = []
+        for n in nb_list:
+            if not n: continue
+            p = (ROOT / n) if not n.startswith(("/", "./")) else pathlib.Path(n)
+            paths.append(p.resolve())
+        return paths
+    return sorted(p for p in ROOT.rglob("*.ipynb") if ".ipynb_checkpoints" not in str(p))
+
+def _run(cmd):
+    return subprocess.run(cmd, check=False, text=True, capture_output=True)
+
+def _ensure_requirements(nb_path):
+    try:
+        import pipreqsnb
+    except ImportError:
+        inst = _run([os.sys.executable, "-m", "pip", "install", "pipreqsnb"])
+        if inst.returncode:
+            print(f"[pip install] warn: install failed for pipreqsnb:\n{inst.stdout}\n{inst.stderr}")
+            return
+
+    req_out = REPORTS / f"reqs.txt"
+    gen = _run(["pipreqsnb", str(nb_path), "--savepath", str(req_out)])
+    if gen.returncode: return
+    if not req_out.exists() or req_out.stat().st_size == 0: return
+    inst = _run([os.sys.executable, "-m", "pip", "install", "-r", str(req_out)])
+    if inst.returncode:
+        print(f"[pip install] warn: install failed for {req_out}:\n{inst.stdout}\n{inst.stderr}")
+    _run(["rm", str(req_out)])
+
+_USERDATA_RE = re.compile(r"userdata\.get\s*\(\s*(['\"])([^'\"]+)\1\s*(?:,\s*([^)]+))?\s*\)")
+
+def _patch_colab_userdata(nb):
+    for cell in nb.cells:
+        if cell.cell_type != "code": continue
+        src = cell.source or ""
+        lines, had_os = [], False
+        for line in src.splitlines():
+            if line.strip().startswith(("from google.colab import userdata", "import google.colab")): continue
+            if re.match(r"^\s*import\s+os(\s|$)", line): had_os = True
+            lines.append(line)
+        src = "\n".join(lines)
+        def _sub(m):
+            key, default = m.group(2), m.group(3)
+            return f"os.getenv('{key}', {default})" if default else f"os.getenv('{key}')"
+        if "userdata.get(" in src:
+            src2 = _USERDATA_RE.sub(_sub, src)
+            if src2 != src and not had_os: src = "import os\n" + src2
+            else: src = src2
+        cell.source = src
+    return nb
+
+def _summarize_outputs(outputs):
+    buf, img_count, err = [], 0, None
+    for out in outputs or []:
+        ot = out.get("output_type")
+        if ot == "stream": buf.append(out.get("text", ""))
+        elif ot in ("execute_result", "display_data"):
+            data = out.get("data", {})
+            text = data.get("text/plain") or ""
+            buf.append("".join(text) if isinstance(text, list) else str(text))
+        elif ot == "error": err = {"ename": out.get("ename"), "evalue": out.get("evalue")}
+    return {"text": "".join(buf).strip(), "images": img_count, "error": err}
+
+def _collect_cell_snapshots(nb):
+    return [{"index": i, "type": c.cell_type, "code": c.source,
+             "summary": _summarize_outputs(c.get("outputs")) if c.cell_type == "code" else None}
+            for i, c in enumerate(nb.cells)]
+
+def _gemini_compare_batches(file_name, diffs, batch_size=20, progress_callback=None):
+    if not diffs: return [], []
+    api_key = os.getenv("GOOGLE_API_KEY")
+    if not api_key: return [], ["<AI compare skipped: GOOGLE_API_KEY missing>"]
+
+    model = os.getenv("GEMINI_MODEL", "gemini-2.5-flash")
+    url = f"https://generativelanguage.googleapis.com/v1beta/models/{model}:generateContent?key={api_key}"
+    system_text = ("""
+        You are an output-diff judge for notebook cells. Given code, OLD output, and NEW output, 
+        classify each cell into EXACTLY one of:
+        - wrong (Wrong or totally different)
+        - slightly_changed (Slightly different)
+        - ok_cells (Outputs are effectively the same. Allowed minor variations include: 
+            - wording/phrasing differences, 
+            - timing/speed values (e.g., download progress), 
+            - harmless pip/installation warnings.
+          Do NOT count these as real changes.)
+        Return ONLY a JSON array of objects: {index:int, bucket:string, note:string}. 
+        No extra keys, no narration.
+    """)
+
+
+    def _coerce_json(text):
+        t = text.strip()
+        if t.startswith("```"):
+            t = t.strip("`")
+            if "\n" in t: t = t.split("\n", 1)[1]
+        start, end = t.find("["), t.rfind("]")
+        if start != -1 and end != -1 and end > start:
+            t = t[start:end+1]
+        t = re.sub(r",\s*(\]|\})", r"\1", t)
+        return json.loads(t)
+
+    results, raw_texts = [], []
+    total_batches = (len(diffs) + batch_size - 1) // batch_size
+    for i in range(0, len(diffs), batch_size):
+        if progress_callback:
+            progress_callback(i // batch_size + 1, total_batches)
+
+        batch = diffs[i:i+batch_size]
+        blocks = [f"Cell {d['index']}\n```python\n{d['code'] or ''}\n```\n"
+                  f"OLD OUTPUT:\n{d['old_text'] or '(empty)'}\n\n"
+                  f"NEW OUTPUT:\n{d['new_text'] or '(empty)'}\n----\n" for d in batch]
+
+        payload = {
+            "system_instruction": {"parts": [{"text": system_text}]},
+            "contents": [{"role": "user", "parts": [{"text": f"File: {file_name}\nEvaluate these cells:\n\n{''.join(blocks)}"}]}],
+            "generationConfig": {"temperature": 0.1, "maxOutputTokens": 8192, "response_mime_type": "application/json"}
+        }
+
+        try:
+            resp = requests.post(url, json=payload, timeout=90)
+            resp.raise_for_status()
+            data = resp.json()
+            part = data.get("candidates", [{}])[0].get("content", {}).get("parts", [{}])[0]
+            parsed = part if 'text' not in part else _coerce_json(part.get('text', '[]'))
+
+            for item in parsed:
+                idx = int(item.get("index"))
+                raw_bucket = (item.get("bucket") or "").strip().lower()
+                bucket = ("ok_cells" if raw_bucket in ("ok", "same", "almost_same", "ok_cells") else
+                         "slightly_changed" if raw_bucket in ("slightly", "slightly_changed") else "wrong")
+                results.append({"index": idx, "bucket": bucket, "note": (item.get("note") or "").strip()})
+        except Exception as e:
+            raw_texts.append(f"<AI compare error: {e}>")
+
+    return results
+
+def _clear_screen():
+    os.system('cls' if os.name == 'nt' else 'clear')
+
+class TestProgressUI:
+    def __init__(self, total_files):
+        self.total_files = total_files
+        self.history = []
+        self.current_file = ""
+        self.current_step = ""
+        self.files_done = 0
+
+    def start_file(self, path):
+        self.current_file = str(path.relative_to(ROOT))
+        self.current_step = "Initializing..."
+        # Add a "running" entry for the current file
+        running_report = {
+            "file": str(path),
+            "status": "running",
+            "buckets": {"ok_cells": {}, "slightly_changed": {}, "wrong": {}}
+        }
+        # Remove any previous running entry for this file
+        self.history = [r for r in self.history if r["file"] != str(path)]
+        self.history.append(running_report)
+        self.draw()
+
+    def update_progress(self, step_message):
+        self.current_step = step_message
+        self.draw()
+
+    def finish_file(self, report):
+        self.history = [r for r in self.history if r["file"] != report["file"]]
+        self.history.append(report)
+        self.current_step = "Completed."
+        self.current_file = ""
+        self.files_done += 1
+
+    def draw(self, final_summary=False):
+        _clear_screen()
+        percent = (self.files_done / self.total_files * 100) if self.total_files > 0 else 0
+        status_line = f"Test suite finished for {self.total_files} notebooks.\n" if final_summary \
+            else f"Running test suite for {self.total_files} notebooks... ({percent:.0f}% complete)\n"
+        sys.stdout.write(status_line)
+
+        if self.current_file:
+            sys.stdout.write(f"Current: {self.current_file}\nStep: {self.current_step}\n\n")
+
+        ai_compare = os.getenv("AI_COMPARE", "0") == "1"
+        if ai_compare:
+            header = f"{'File':<50} {'Status':<10} {'OK':>5} {'Slightly Changed':>18} {'Wrong':>7}"
+        else:
+            header = f"{'File':<50} {'Status':<10}"
+
+        sys.stdout.write("\n" + header + "\n" + "─" * len(header) + "\n")
+        for r in self.history:
+            status = r['status']
+            file_rel = str(pathlib.Path(r["file"]).relative_to(ROOT))
+            if ai_compare:
+                buckets = r.get("buckets", {})
+                ok, slight, wrong = len(buckets.get("ok_cells", {})), len(buckets.get("slightly_changed", {})), len(buckets.get("wrong", {}))
+                if status == 'running':
+                    ok_str, slight_str, wrong_str = ('...', '...', '...')
+                elif status == 'passed':
+                    ok_str, slight_str, wrong_str = (str(ok), str(slight), str(wrong))
+                else:
+                    ok_str, slight_str, wrong_str = ('-', '-', str(wrong or 1))
+                sys.stdout.write(f"{file_rel:<50} {status:<10} {ok_str:>5} {slight_str:>18} {wrong_str:>7}\n")
+            else:
+                sys.stdout.write(f"{file_rel:<50} {status:<10}\n")
+        sys.stdout.flush()
+
+def run_notebook_test(nb_path, ui):
+    """Contains the logic for running a single notebook."""
+    ui.start_file(nb_path)
+    ui.update_progress("Ensuring requirements...")
+    _ensure_requirements(nb_path)
+
+    nb = nbformat.read(nb_path, as_version=4)
+    old_snaps = _collect_cell_snapshots(nb)
+    nb = _patch_colab_userdata(nb)
+
+    client = NotebookClient(nb, timeout=int(os.getenv("NB_TIMEOUT", "900")),
+                            kernel_name=os.getenv("NB_KERNEL", "python3"),
+                            allow_errors=True, record_timing=True, store_widget_state=False)
+
+    started = time.time()
+    first_error_msg = None
+    code_cells_with_indices = [(i, c) for i, c in enumerate(nb.cells) if c.cell_type == 'code']
+    total_code_cells = len(code_cells_with_indices)
+
+    with client.setup_kernel():
+        for i, (cell_index, cell) in enumerate(code_cells_with_indices):
+            ui.update_progress(f"Executing cell {i + 1}/{total_code_cells}...")
+            try:
+                client.execute_cell(cell, cell_index)
+            except CellExecutionError as e:
+                if not first_error_msg:
+                    first_error_msg = str(e).split("\n", 1)[0]
+
+    new_snaps = _collect_cell_snapshots(nb)
+    rel = str(nb_path.relative_to(ROOT)).replace("/", "__")
+    file_report = {
+        "file": str(nb_path),
+        "duration_sec": round(time.time() - started, 3),
+        "status": "failed" if first_error_msg else "passed",
+        "buckets": {"ok_cells": {}, "slightly_changed": {}, "wrong": {}}
+    }
+
+    if first_error_msg:
+        for c in new_snaps:
+            if c["type"] == "code" and c["summary"] and c["summary"]["error"]:
+                file_report["buckets"]["wrong"][str(c["index"])] = {
+                    "cell_code": c["code"], "ai_note": "Execution error in this run."
+                }
+
+    if os.getenv("AI_COMPARE", "0") == "1" and not first_error_msg:
+        diffs = [{"index": new["index"], "code": new["code"],
+                  "old_text": (old.get("summary") or {}).get("text", ""),
+                  "new_text": (new.get("summary") or {}).get("text", "")}
+                 for old, new in zip(old_snaps, new_snaps) if new["type"] == "code"]
+
+        def ai_progress(batch_num, total_batches):
+            ui.update_progress(f"AI comparing batch {batch_num}/{total_batches}...")
+
+        ai_results = _gemini_compare_batches(str(nb_path), diffs, progress_callback=ai_progress)
+        by_cell = {r["index"]: r for r in ai_results}
+
+        for d in diffs:
+            r = by_cell.get(d["index"])
+            if not r: continue
+            bucket = r["bucket"]
+            file_report["buckets"].setdefault(bucket, {})[str(d["index"])] = {
+                "cell_code": d["code"], "old_text": d["old_text"], "new_text": d["new_text"], "ai_note": (r.get("note") or "")[:280]
+            }
+
+    (REPORTS / f"{rel}.compare.json").write_text(json.dumps(file_report, indent=2), encoding="utf-8")
+    ui.finish_file(file_report)
+    if first_error_msg:
+        print(f"\nERROR in {nb_path.name}: {first_error_msg}\n")
+
+
+def main():
+    """Main execution function."""
+    notebooks = _discover()
+    if not notebooks:
+        print("No notebooks found to test.")
+        sys.exit(0)
+
+    ui = TestProgressUI(len(notebooks))
+
+    for nb_path in notebooks:
+        run_notebook_test(nb_path, ui)
+
+    ui.draw(final_summary=True)
+
+if __name__ == "__main__":
+    main()
\ No newline at end of file