diff --git a/bincio/serve/routers/admin.py b/bincio/serve/routers/admin.py
index 4b294a0..ede2d27 100644
--- a/bincio/serve/routers/admin.py
+++ b/bincio/serve/routers/admin.py
@@ -10,7 +10,7 @@ from pathlib import Path
 from typing import Any
 
 from fastapi import APIRouter, Cookie, HTTPException, Request
-from fastapi.responses import JSONResponse, StreamingResponse
+from fastapi.responses import FileResponse, JSONResponse, StreamingResponse
 
 from bincio.serve import deps, tasks
 from bincio.serve.models import ResetPasswordCodeResponse
@@ -58,6 +58,16 @@ def _wipe_user_activities(user_dir: Path) -> int:
     return deleted
 
 
+@router.get("/api/admin/stats")
+async def admin_stats(bincio_session: str | None = Cookie(default=None)) -> FileResponse:
+    """Serve the latest usage stats figure. Admin only."""
+    deps._require_admin(bincio_session)
+    path = deps._get_data_dir().parent / "stats" / "latest.png"
+    if not path.exists():
+        raise HTTPException(404, "Stats not yet generated — run scripts/usage_stats.py first")
+    return FileResponse(path, media_type="image/png", headers={"Cache-Control": "no-cache, no-store"})
+
+
 @router.get("/api/admin/users")
 async def admin_users(bincio_session: str | None = Cookie(default=None)) -> JSONResponse:
     deps._require_admin(bincio_session)
diff --git a/scripts/usage_stats.py b/scripts/usage_stats.py
new file mode 100644
index 0000000..44d59a4
--- /dev/null
+++ b/scripts/usage_stats.py
@@ -0,0 +1,294 @@
+#!/usr/bin/env python3
+# /// script
+# dependencies = ["matplotlib>=3.9", "pandas>=2.2"]
+# ///
+"""
+Bincio usage statistics — parses nginx access logs and produces a
+multi-panel matplotlib figure saved as a PNG.
+
+Run locally:  uv run scripts/usage_stats.py
+On VPS cron:  0 3 * * 1  cd /opt/bincio && uv run scripts/usage_stats.py
+Output:       /var/bincio/stats/latest.png  (served at /api/admin/stats)
+"""
+from __future__ import annotations
+
+import argparse
+import gzip
+import re
+import sys
+from datetime import datetime
+from pathlib import Path
+from urllib.parse import urlparse
+
+import matplotlib
+matplotlib.use("Agg")
+import matplotlib.pyplot as plt
+import matplotlib.ticker as ticker
+import numpy as np
+import pandas as pd
+
+# ── Config ────────────────────────────────────────────────────────────────────
+
+LOG_DIR    = Path("/var/log/nginx")
+OUTPUT_DIR = Path("/var/bincio/stats")
+OUTPUT     = OUTPUT_DIR / "latest.png"
+
+# ── Log parsing ───────────────────────────────────────────────────────────────
+
+_LOG_RE = re.compile(
+    r'(?P<ip>\S+) \S+ \S+ \[(?P<time>[^\]]+)\] '
+    r'"(?P<method>\S+) (?P<path>\S+) [^"]+" '
+    r'(?P<status>\d+) \S+ '
+    r'"(?P<referer>[^"]*)" "(?P<ua>[^"]*)"'
+)
+_TS_FMT = "%d/%b/%Y:%H:%M:%S %z"
+
+# ── Bot filtering ─────────────────────────────────────────────────────────────
+
+_BOT_UA_RE = re.compile(
+    r"bot|crawl|spider|scan|nmap|masscan|zgrab|python-requests|"
+    r"curl|wget|nikto|nuclei|go-http|censys|shodan|paloalto|expanse|"
+    r"dataforseo|semrush|ahrefs|mj12|dotbot|petalbot|fuzz|dirbuster",
+    re.I,
+)
+_BOT_PATH_RE = re.compile(
+    r"^/(wp-|phpmyadmin|xmlrpc|\.env|\.git|setup\.php|"
+    r"SDK/|actuator/|cgi-bin/|PROPFIND|\.well-known/acme)",
+    re.I,
+)
+
+def _is_bot(ua: str, path: str) -> bool:
+    if not ua or ua == "-":
+        return True
+    if _BOT_UA_RE.search(ua):
+        return True
+    if _BOT_PATH_RE.search(path):
+        return True
+    return False
+
+# ── Feature mapping (from Referer header) ─────────────────────────────────────
+
+# Evaluated in order: first match wins. None label = exclude.
+_FEATURE_MAP: list[tuple[str, str | None, str | None]] = [
+    ("planner.bincio.org",  None,          "planner"),
+    ("wiki.bincio.org",     None,          "wiki"),
+    ("activity.bincio.org", "/admin/",     None),         # exclude admin polling
+    ("activity.bincio.org", "/activity/",  "activity"),
+    ("activity.bincio.org", "/segments/",  "segments"),
+    ("activity.bincio.org", "/stats/",     "stats"),
+    ("activity.bincio.org", "/explore/",   "explore"),
+    ("activity.bincio.org", "/ideas/",     "ideas"),
+    ("activity.bincio.org", "/u/",         "profile"),
+    ("activity.bincio.org", None,          "feed"),
+    ("bincio.org",          None,          "hub"),
+]
+
+FEATURE_COLORS = {
+    "feed":     "#60a5fa",
+    "activity": "#4ade80",
+    "segments": "#facc15",
+    "planner":  "#f97316",
+    "wiki":     "#a855f7",
+    "ideas":    "#f43f5e",
+    "explore":  "#34d399",
+    "profile":  "#94a3b8",
+    "hub":      "#64748b",
+    "stats":    "#e879a0",
+}
+
+def _feature(referer: str) -> str | None:
+    if not referer or referer == "-":
+        return None
+    try:
+        p = urlparse(referer)
+        host = p.netloc.lower().lstrip("www.")
+        path = p.path
+    except Exception:
+        return None
+    for h, prefix, label in _FEATURE_MAP:
+        if host == h:
+            if prefix is None or path.startswith(prefix):
+                return label
+    return None
+
+# ── Loading ───────────────────────────────────────────────────────────────────
+
+def load_logs(log_dir: Path) -> pd.DataFrame:
+    rows = []
+    files = sorted(log_dir.glob("access.log*"), reverse=True)
+    if not files:
+        print(f"No log files found in {log_dir}", file=sys.stderr)
+        sys.exit(1)
+
+    for f in files:
+        opener = gzip.open if f.suffix == ".gz" else open
+        try:
+            with opener(f, "rt", errors="replace") as fh:
+                for line in fh:
+                    m = _LOG_RE.match(line)
+                    if not m:
+                        continue
+                    ua   = m.group("ua")
+                    path = m.group("path")
+                    if _is_bot(ua, path):
+                        continue
+                    try:
+                        ts = datetime.strptime(m.group("time"), _TS_FMT)
+                    except ValueError:
+                        continue
+                    rows.append({
+                        "ts":      ts,
+                        "ip":      m.group("ip"),
+                        "method":  m.group("method"),
+                        "path":    path,
+                        "status":  int(m.group("status")),
+                        "referer": m.group("referer"),
+                    })
+        except Exception as exc:
+            print(f"Warning: skipping {f.name}: {exc}", file=sys.stderr)
+
+    if not rows:
+        print("No usable log entries found after bot filtering.", file=sys.stderr)
+        sys.exit(1)
+
+    df = pd.DataFrame(rows)
+    df["ts"]      = pd.to_datetime(df["ts"], utc=True)
+    df["hour"]    = df["ts"].dt.hour
+    df["dow"]     = df["ts"].dt.dayofweek          # 0 = Monday
+    df["feature"] = df["referer"].map(_feature)
+    return df
+
+# ── Figure ────────────────────────────────────────────────────────────────────
+
+BG    = "#09090b"
+FG    = "#e4e4e7"
+GRID  = "#27272a"
+BLUE  = "#60a5fa"
+
+def _style_ax(ax: plt.Axes) -> None:
+    ax.set_facecolor(BG)
+    ax.tick_params(colors=FG, labelsize=9)
+    for spine in ax.spines.values():
+        spine.set_edgecolor(GRID)
+
+def make_figure(df: pd.DataFrame, output: Path) -> None:
+    # ── daily logins ──────────────────────────────────────────────────────────
+    login_mask  = (df["method"] == "POST") & (df["path"] == "/api/auth/login") & (df["status"] == 200)
+    full_range  = pd.date_range(df["ts"].min(), df["ts"].max(), freq="D", tz="UTC")
+    daily_logins = df[login_mask].set_index("ts").resample("D").size().reindex(full_range, fill_value=0)
+    rolling7     = daily_logins.rolling(7, center=True, min_periods=1).mean()
+
+    # ── feature usage (weekly) ────────────────────────────────────────────────
+    feat_df = df[df["feature"].notna()].copy()
+    feat_weekly = (
+        feat_df.set_index("ts")
+        .groupby([pd.Grouper(freq="W-MON"), "feature"])
+        .size()
+        .unstack(fill_value=0)
+    )
+    feat_order = [f for f in FEATURE_COLORS if f in feat_weekly.columns]
+    feat_weekly = feat_weekly[feat_order]
+
+    # ── heatmap: hour × weekday (API requests, no admin) ─────────────────────
+    api_df = df[
+        df["path"].str.startswith("/api/") &
+        ~df["path"].str.startswith("/api/admin")
+    ]
+    heat = (
+        api_df.groupby(["dow", "hour"]).size()
+        .unstack(fill_value=0)
+        .reindex(index=range(7), columns=range(24), fill_value=0)
+    )
+
+    # ── layout ────────────────────────────────────────────────────────────────
+    plt.style.use("dark_background")
+    fig = plt.figure(figsize=(15, 10), facecolor=BG)
+    gs  = fig.add_gridspec(2, 2, hspace=0.42, wspace=0.30,
+                           left=0.07, right=0.97, top=0.92, bottom=0.08)
+    ax_log  = fig.add_subplot(gs[0, 0])
+    ax_heat = fig.add_subplot(gs[0, 1])
+    ax_feat = fig.add_subplot(gs[1, :])
+
+    for ax in (ax_log, ax_heat, ax_feat):
+        _style_ax(ax)
+
+    # Panel 1 — daily logins + rolling mean
+    ax_log.bar(daily_logins.index, daily_logins.values,
+               color=BLUE, alpha=0.30, width=pd.Timedelta(hours=20))
+    ax_log.plot(daily_logins.index, rolling7.values,
+                color=BLUE, linewidth=2, label="7-day avg")
+    ax_log.set_title("Daily logins", color=FG, fontsize=11, pad=8)
+    ax_log.set_ylabel("count", color=FG, fontsize=9)
+    ax_log.yaxis.set_major_locator(ticker.MaxNLocator(integer=True, nbins=5))
+    ax_log.tick_params(axis="x", rotation=25)
+    ax_log.legend(fontsize=8, framealpha=0.15, facecolor=BG, edgecolor=GRID)
+    ax_log.grid(axis="y", color=GRID, linewidth=0.5)
+
+    # Panel 2 — heatmap
+    days = ["Mon", "Tue", "Wed", "Thu", "Fri", "Sat", "Sun"]
+    im = ax_heat.imshow(heat.values, aspect="auto", cmap="Blues",
+                        interpolation="nearest", origin="upper")
+    ax_heat.set_xticks(range(0, 24, 3))
+    ax_heat.set_xticklabels([f"{h:02d}h" for h in range(0, 24, 3)], color=FG, fontsize=8)
+    ax_heat.set_yticks(range(7))
+    ax_heat.set_yticklabels(days, color=FG, fontsize=9)
+    ax_heat.set_title("API requests: hour × weekday (UTC)", color=FG, fontsize=11, pad=8)
+    cb = fig.colorbar(im, ax=ax_heat, fraction=0.046, pad=0.04)
+    cb.ax.tick_params(labelcolor=FG, labelsize=8)
+
+    # Panel 3 — feature usage stacked area
+    if not feat_weekly.empty:
+        n = len(feat_weekly)
+        x = np.arange(n)
+        bottom = np.zeros(n)
+        for feat in feat_order:
+            vals = feat_weekly[feat].values.astype(float)
+            ax_feat.fill_between(x, bottom, bottom + vals,
+                                 color=FEATURE_COLORS[feat], alpha=0.80, label=feat)
+            bottom += vals
+
+        week_labels = [str(w.date()) for w in feat_weekly.index]
+        step = max(1, n // 12)
+        ax_feat.set_xticks(x[::step])
+        ax_feat.set_xticklabels(week_labels[::step], rotation=30, ha="right",
+                                fontsize=8, color=FG)
+        ax_feat.set_xlim(0, n - 1)
+        ax_feat.set_title("Feature usage per week  (from Referer)", color=FG, fontsize=11, pad=8)
+        ax_feat.set_ylabel("API requests", color=FG, fontsize=9)
+        ax_feat.legend(loc="upper left", fontsize=8, framealpha=0.15,
+                       facecolor=BG, edgecolor=GRID, ncol=len(feat_order))
+        ax_feat.grid(axis="y", color=GRID, linewidth=0.5)
+    else:
+        ax_feat.text(0.5, 0.5, "No feature data (no Referer headers yet)",
+                     ha="center", va="center", color=FG, transform=ax_feat.transAxes)
+
+    total_logins = int(login_mask.sum())
+    span_days    = (df["ts"].max() - df["ts"].min()).days + 1
+    fig.suptitle(
+        f"bincio — {total_logins} logins over {span_days} days "
+        f"· generated {datetime.utcnow().strftime('%Y-%m-%d')}",
+        color=FG, fontsize=12, y=0.97,
+    )
+
+    output.parent.mkdir(parents=True, exist_ok=True)
+    fig.savefig(output, dpi=150, facecolor=BG, bbox_inches="tight")
+    print(f"Saved → {output}")
+    plt.close(fig)
+
+# ── Entry point ───────────────────────────────────────────────────────────────
+
+def main() -> None:
+    ap = argparse.ArgumentParser(description="Generate bincio usage stats figure.")
+    ap.add_argument("--log-dir", type=Path, default=LOG_DIR,  metavar="DIR")
+    ap.add_argument("--output",  type=Path, default=OUTPUT,   metavar="FILE")
+    args = ap.parse_args()
+
+    print("Loading logs…", file=sys.stderr)
+    df = load_logs(args.log_dir)
+    span = (df["ts"].max() - df["ts"].min()).days + 1
+    print(f"  {len(df):,} non-bot requests over {span} days", file=sys.stderr)
+    make_figure(df, args.output)
+
+
+if __name__ == "__main__":
+    main()