convoviz 0.2.12__py3-none-any.whl

convoviz/analysis/graphs.py

@@ -0,0 +1,855 @@
+"""Graph generation for conversation analytics.
+
+Goals:
+- Professional, consistent styling across plots.
+- High-signal summaries by default (avoid output spam).
+- Correct time bucketing (based on *message timestamps*, not conversation creation time).
+"""
+
+from __future__ import annotations
+
+from collections import defaultdict
+from collections.abc import Callable, Iterable
+from datetime import UTC, datetime
+from pathlib import Path
+
+import matplotlib.dates as mdates
+import matplotlib.font_manager as fm
+import matplotlib.ticker as mticker
+from matplotlib.axes import Axes
+from matplotlib.figure import Figure
+from matplotlib.image import AxesImage
+from tqdm import tqdm
+
+from convoviz.config import GraphConfig, get_default_config
+from convoviz.models import ConversationCollection
+from convoviz.utils import get_asset_path
+
+WEEKDAYS = ["Monday", "Tuesday", "Wednesday", "Thursday", "Friday", "Saturday", "Sunday"]
+
+
+def _load_font(config: GraphConfig) -> fm.FontProperties:
+    font_path = get_asset_path(f"fonts/{config.font_name}")
+    return fm.FontProperties(fname=str(font_path)) if font_path.exists() else fm.FontProperties()
+
+
+def _style_axes(ax: Axes, config: GraphConfig) -> None:
+    # Clean look
+    ax.set_facecolor("white")
+    ax.spines["top"].set_visible(False)
+    ax.spines["right"].set_visible(False)
+    ax.spines["left"].set_color("#d0d7de")
+    ax.spines["bottom"].set_color("#d0d7de")
+    ax.tick_params(colors="#24292f")
+    ax.yaxis.set_major_locator(mticker.MaxNLocator(nbins=6, integer=True))
+
+    if config.grid:
+        ax.grid(axis="y", linestyle="-", linewidth=0.8, alpha=0.35, color="#8c959f")
+    ax.set_axisbelow(True)
+
+
+def _apply_tick_font(ax: Axes, font_prop: fm.FontProperties) -> None:
+    for label in ax.get_xticklabels() + ax.get_yticklabels():
+        label.set_fontproperties(font_prop)
+
+
+def _setup_single_axes(config: GraphConfig) -> tuple[Figure, Axes, fm.FontProperties]:
+    fig = Figure(figsize=config.figsize, dpi=config.dpi, facecolor="white")
+    ax: Axes = fig.add_subplot()
+    font_prop = _load_font(config)
+    _style_axes(ax, config)
+    return fig, ax, font_prop
+
+
+def _ts_to_dt(ts: float, config: GraphConfig) -> datetime:
+    """Convert epoch timestamps into aware datetimes based on config."""
+    dt_utc = datetime.fromtimestamp(ts, UTC)
+    return dt_utc if config.timezone == "utc" else dt_utc.astimezone()
+
+
+def _tz_label(config: GraphConfig) -> str:
+    return "UTC" if config.timezone == "utc" else "Local"
+
+
+def _month_start(dt: datetime) -> datetime:
+    return dt.replace(day=1, hour=0, minute=0, second=0, microsecond=0)
+
+
+def _year_start(dt: datetime) -> datetime:
+    return dt.replace(month=1, day=1, hour=0, minute=0, second=0, microsecond=0)
+
+
+def _day_start(dt: datetime) -> datetime:
+    return dt.replace(hour=0, minute=0, second=0, microsecond=0)
+
+
+def _iter_month_starts(start: datetime, end: datetime) -> list[datetime]:
+    start = _month_start(start)
+    end = _month_start(end)
+    months: list[datetime] = []
+    cur = start
+    while cur <= end:
+        months.append(cur)
+        year, month = cur.year, cur.month
+        cur = cur.replace(year=year + 1, month=1) if month == 12 else cur.replace(month=month + 1)
+    return months
+
+
+def _fill_missing_months(counts: dict[datetime, int]) -> tuple[list[datetime], list[int]]:
+    if not counts:
+        return [], []
+    keys = sorted(counts.keys())
+    months = _iter_month_starts(keys[0], keys[-1])
+    return months, [counts.get(m, 0) for m in months]
+
+
+def _aggregate_counts_by_month(
+    timestamps: Iterable[float],
+    config: GraphConfig,
+) -> dict[datetime, int]:
+    counts: defaultdict[datetime, int] = defaultdict(int)
+    for ts in timestamps:
+        dt = _ts_to_dt(ts, config)
+        counts[_month_start(dt)] += 1
+    return dict(counts)
+
+
+def _moving_average(values: list[int], window: int) -> list[float]:
+    if window <= 1:
+        return [float(v) for v in values]
+    if len(values) < window:
+        return []
+    out: list[float] = []
+    running = sum(values[:window])
+    out.append(running / window)
+    for i in range(window, len(values)):
+        running += values[i] - values[i - window]
+        out.append(running / window)
+    return out
+
+
+def generate_week_barplot(
+    timestamps: list[float],
+    title: str,
+    config: GraphConfig | None = None,
+) -> Figure:
+    """Create a bar graph showing message distribution across weekdays.
+
+    Args:
+        timestamps: List of Unix timestamps
+        title: Title for the graph
+        config: Optional graph configuration
+
+    Returns:
+        Matplotlib Figure object
+    """
+    cfg = config or get_default_config().graph
+    fig, ax, font_prop = _setup_single_axes(cfg)
+
+    weekday_counts: dict[str, int] = dict.fromkeys(WEEKDAYS, 0)
+    for ts in timestamps:
+        dt = _ts_to_dt(ts, cfg)
+        weekday_counts[WEEKDAYS[dt.weekday()]] += 1
+
+    x = list(range(len(WEEKDAYS)))
+    y = [weekday_counts[d] for d in WEEKDAYS]
+
+    bars = ax.bar(x, y, color=cfg.color, alpha=0.9, width=0.72)
+    if cfg.show_counts:
+        ax.bar_label(bars, padding=3, fontsize=9, fontproperties=font_prop, color="#24292f")
+
+    ax.set_title(f"{title} · Weekday pattern", fontproperties=font_prop, fontsize=14, pad=14)
+    ax.set_xlabel("Weekday", fontproperties=font_prop)
+    ax.set_ylabel("User prompts", fontproperties=font_prop)
+    ax.set_xticks(x)
+    ax.set_xticklabels(WEEKDAYS, rotation=35, ha="right", fontproperties=font_prop)
+    _apply_tick_font(ax, font_prop)
+
+    fig.tight_layout()
+    return fig
+
+
+def generate_hour_barplot(
+    timestamps: list[float],
+    title: str,
+    config: GraphConfig | None = None,
+) -> Figure:
+    """Create a bar graph showing message distribution across hours of the day (0-23).
+
+    Args:
+        timestamps: List of Unix timestamps
+        title: Title for the graph
+        config: Optional graph configuration
+
+    Returns:
+        Matplotlib Figure object
+    """
+    cfg = config or get_default_config().graph
+    fig, ax, font_prop = _setup_single_axes(cfg)
+
+    hour_counts: dict[int, int] = dict.fromkeys(range(24), 0)
+    for ts in timestamps:
+        dt = _ts_to_dt(ts, cfg)
+        hour_counts[dt.hour] += 1
+
+    y = [hour_counts[i] for i in range(24)]
+
+    bars = ax.bar(range(24), y, color=cfg.color, alpha=0.9, width=0.72)
+    if cfg.show_counts:
+        ax.bar_label(bars, padding=2, fontsize=7, fontproperties=font_prop, color="#24292f")
+
+    ax.set_title(
+        f"{title} · Hourly pattern ({_tz_label(cfg)})",
+        fontproperties=font_prop,
+        fontsize=14,
+        pad=14,
+    )
+    ax.set_xlabel(f"Hour of day ({_tz_label(cfg)})", fontproperties=font_prop)
+    ax.set_ylabel("User prompts", fontproperties=font_prop)
+
+    ax.set_xticks(range(0, 24, 2))
+    ax.set_xticklabels([f"{i:02d}:00" for i in range(0, 24, 2)], fontproperties=font_prop)
+    _apply_tick_font(ax, font_prop)
+
+    fig.tight_layout()
+    return fig
+
+
+def generate_model_piechart(
+    collection: ConversationCollection,
+    config: GraphConfig | None = None,
+) -> Figure:
+    """Create a model usage chart.
+
+    Note: kept for backwards compatibility (historically a pie chart). We now render a
+    more readable horizontal bar chart with percentages.
+    """
+    cfg = config or get_default_config().graph
+    model_counts: defaultdict[str, int] = defaultdict(int)
+
+    for conv in collection.conversations:
+        model = conv.model or "Unknown"
+        model_counts[model] += 1
+
+    total = sum(model_counts.values())
+    fig, ax, font_prop = _setup_single_axes(cfg)
+
+    if total == 0:
+        ax.text(0.5, 0.5, "No data", ha="center", va="center", fontproperties=font_prop)
+        ax.set_axis_off()
+        return fig
+
+    # Group minor models to keep the plot readable.
+    threshold = 0.05
+    refined_counts: dict[str, int] = {}
+    other_count = 0
+    for model, count in model_counts.items():
+        if count / total < threshold:
+            other_count += count
+        else:
+            refined_counts[model] = count
+    if other_count:
+        refined_counts["Other"] = other_count
+
+    items = sorted(refined_counts.items(), key=lambda x: x[1], reverse=True)
+    labels = [k for k, _ in items][:10]
+    counts = [v for _, v in items][:10]
+    y = list(range(len(labels)))[::-1]
+
+    bars = ax.barh(y, counts[::-1], color=cfg.color, alpha=0.9, height=0.6)
+    ax.set_yticks(y)
+    ax.set_yticklabels(labels[::-1], fontproperties=font_prop)
+    ax.set_xlabel("Conversations", fontproperties=font_prop)
+    ax.set_title("Model usage", fontproperties=font_prop, fontsize=14, pad=14)
+
+    for bar, count in zip(bars, counts[::-1], strict=True):
+        pct = 100 * (count / total)
+        ax.text(
+            bar.get_width(),
+            bar.get_y() + bar.get_height() / 2,
+            f"  {count}  ({pct:.1f}%)",
+            va="center",
+            ha="left",
+            fontproperties=font_prop,
+            fontsize=9,
+            color="#24292f",
+        )
+
+    _apply_tick_font(ax, font_prop)
+    fig.tight_layout()
+    return fig
+
+
+def generate_length_histogram(
+    collection: ConversationCollection,
+    config: GraphConfig | None = None,
+) -> Figure:
+    """Create a histogram showing distribution of conversation lengths.
+
+    Caps the X-axis at the 95th percentile to focus on typical lengths.
+
+    Args:
+        collection: Collection of conversations
+        config: Optional graph configuration
+
+    Returns:
+        Matplotlib Figure object
+    """
+    cfg = config or get_default_config().graph
+    lengths = [conv.message_count("user") for conv in collection.conversations]
+    fig, ax, font_prop = _setup_single_axes(cfg)
+
+    if not lengths:
+        ax.text(0.5, 0.5, "No data", ha="center", va="center", fontproperties=font_prop)
+        ax.set_axis_off()
+        return fig
+
+    sorted_lengths = sorted(lengths)
+    p50 = sorted_lengths[int(0.50 * (len(sorted_lengths) - 1))]
+    p90 = sorted_lengths[int(0.90 * (len(sorted_lengths) - 1))]
+    p95 = sorted_lengths[int(0.95 * (len(sorted_lengths) - 1))]
+    cap = max(int(p95), 5)
+    plot_lengths = [min(L, cap) for L in lengths]
+
+    bin_count = min(24, max(10, cap // 2))
+    ax.hist(
+        plot_lengths,
+        bins=bin_count,
+        color=cfg.color,
+        alpha=0.85,
+        rwidth=0.9,
+        edgecolor="white",
+        linewidth=0.5,
+    )
+
+    ax.axvline(p50, color="#24292f", linewidth=1.2, alpha=0.8)
+    ax.axvline(p90, color="#cf222e", linewidth=1.2, alpha=0.8)
+    ax.text(
+        p50,
+        ax.get_ylim()[1] * 0.95,
+        f"median={p50}",
+        rotation=90,
+        va="top",
+        ha="right",
+        fontproperties=font_prop,
+        fontsize=9,
+        color="#24292f",
+    )
+    ax.text(
+        p90,
+        ax.get_ylim()[1] * 0.95,
+        f"p90={p90}",
+        rotation=90,
+        va="top",
+        ha="right",
+        fontproperties=font_prop,
+        fontsize=9,
+        color="#cf222e",
+    )
+
+    ax.set_title("Conversation length (user prompts)", fontproperties=font_prop, fontsize=14, pad=14)
+    ax.set_xlabel("User prompts per conversation", fontproperties=font_prop)
+    ax.set_ylabel("Conversations", fontproperties=font_prop)
+    ax.set_xlim(left=0, right=cap)
+    _apply_tick_font(ax, font_prop)
+
+    fig.tight_layout()
+    return fig
+
+
+def generate_conversation_lifetime_histogram(
+    collection: ConversationCollection,
+    config: GraphConfig | None = None,
+) -> Figure:
+    """Create a histogram of conversation lifetimes (update_time - create_time)."""
+    cfg = config or get_default_config().graph
+    fig, ax, font_prop = _setup_single_axes(cfg)
+
+    lifetimes_days: list[float] = []
+    for conv in collection.conversations:
+        delta = conv.update_time - conv.create_time
+        lifetimes_days.append(max(0.0, delta.total_seconds() / 86_400))
+
+    if not lifetimes_days:
+        ax.text(0.5, 0.5, "No data", ha="center", va="center", fontproperties=font_prop)
+        ax.set_axis_off()
+        return fig
+
+    sorted_vals = sorted(lifetimes_days)
+    p50 = sorted_vals[int(0.50 * (len(sorted_vals) - 1))]
+    p90 = sorted_vals[int(0.90 * (len(sorted_vals) - 1))]
+    p95 = sorted_vals[int(0.95 * (len(sorted_vals) - 1))]
+    cap = max(float(p95), 1.0)
+    plot_vals = [min(v, cap) for v in lifetimes_days]
+
+    ax.hist(
+        plot_vals,
+        bins=24,
+        color=cfg.color,
+        alpha=0.85,
+        rwidth=0.9,
+        edgecolor="white",
+        linewidth=0.5,
+    )
+    ax.axvline(p50, color="#24292f", linewidth=1.2, alpha=0.8)
+    ax.axvline(p90, color="#cf222e", linewidth=1.2, alpha=0.8)
+
+    ax.set_title("Conversation lifetimes (days)", fontproperties=font_prop, fontsize=14, pad=14)
+    ax.set_xlabel("Days between first and last message", fontproperties=font_prop)
+    ax.set_ylabel("Conversations", fontproperties=font_prop)
+    ax.set_xlim(left=0, right=cap)
+    ax.text(
+        0.99,
+        0.98,
+        f"median={p50:.1f}d\np90={p90:.1f}d",
+        transform=ax.transAxes,
+        ha="right",
+        va="top",
+        fontproperties=font_prop,
+        fontsize=9,
+        color="#57606a",
+    )
+    _apply_tick_font(ax, font_prop)
+
+    fig.tight_layout()
+    return fig
+
+
+def generate_monthly_activity_barplot(
+    collection: ConversationCollection,
+    config: GraphConfig | None = None,
+) -> Figure:
+    """Create a bar chart showing total prompt count per month.
+
+    Important: this is computed from *message timestamps* (actual activity), not from
+    the conversation creation month.
+    """
+    cfg = config or get_default_config().graph
+    timestamps = collection.timestamps("user")
+    fig, ax, font_prop = _setup_single_axes(cfg)
+
+    if not timestamps:
+        ax.text(0.5, 0.5, "No data", ha="center", va="center", fontproperties=font_prop)
+        ax.set_axis_off()
+        return fig
+
+    month_counts = _aggregate_counts_by_month(timestamps, cfg)
+    months, values = _fill_missing_months(month_counts)
+    if not months:
+        ax.text(0.5, 0.5, "No data", ha="center", va="center", fontproperties=font_prop)
+        ax.set_axis_off()
+        return fig
+
+    x = mdates.date2num(months)
+    ax.bar(x, values, width=25, color=cfg.color, alpha=0.25, edgecolor="none")
+
+    smooth = _moving_average(values, window=3)
+    if smooth:
+        ax.plot(x[2:], smooth, color=cfg.color, linewidth=2.2, alpha=0.9)
+
+    locator = mdates.AutoDateLocator(minticks=4, maxticks=10)
+    ax.xaxis.set_major_locator(locator)
+    ax.xaxis.set_major_formatter(mdates.ConciseDateFormatter(locator))
+
+    ax.set_title("Monthly activity (user prompts)", fontproperties=font_prop, fontsize=14, pad=14)
+    ax.set_xlabel(f"Month ({_tz_label(cfg)})", fontproperties=font_prop)
+    ax.set_ylabel("User prompts", fontproperties=font_prop)
+    _apply_tick_font(ax, font_prop)
+
+    fig.tight_layout()
+    return fig
+
+
+def generate_daily_activity_lineplot(
+    collection: ConversationCollection,
+    config: GraphConfig | None = None,
+) -> Figure:
+    """Create a line chart showing user prompt count per day (with a rolling mean)."""
+    cfg = config or get_default_config().graph
+    timestamps = collection.timestamps("user")
+
+    fig, ax, font_prop = _setup_single_axes(cfg)
+    if not timestamps:
+        ax.text(0.5, 0.5, "No data", ha="center", va="center", fontproperties=font_prop)
+        ax.set_axis_off()
+        return fig
+
+    counts: defaultdict[datetime, int] = defaultdict(int)
+    for ts in timestamps:
+        dt = _ts_to_dt(ts, cfg)
+        day = dt.replace(hour=0, minute=0, second=0, microsecond=0)
+        counts[day] += 1
+
+    days = sorted(counts.keys())
+    values = [counts[d] for d in days]
+
+    x = mdates.date2num(days)
+    ax.bar(x, values, width=0.9, color=cfg.color, alpha=0.18, edgecolor="none")
+    ax.plot(x, values, color=cfg.color, linewidth=1.2, alpha=0.25)
+    smooth = _moving_average(values, window=7)
+    if smooth:
+        ax.plot(x[6:], smooth, color=cfg.color, linewidth=2.4, alpha=0.95)
+
+    locator = mdates.AutoDateLocator()
+    ax.xaxis.set_major_locator(locator)
+    ax.xaxis.set_major_formatter(mdates.ConciseDateFormatter(locator))
+    ax.set_title("Daily activity (user prompts)", fontproperties=font_prop, fontsize=14, pad=14)
+    ax.set_xlabel(f"Day ({_tz_label(cfg)})", fontproperties=font_prop)
+    ax.set_ylabel("User prompts", fontproperties=font_prop)
+
+    for label in ax.get_xticklabels() + ax.get_yticklabels():
+        label.set_fontproperties(font_prop)
+
+    fig.tight_layout()
+    return fig
+
+
+def generate_activity_heatmap(
+    collection: ConversationCollection,
+    config: GraphConfig | None = None,
+) -> Figure:
+    """Create a heatmap of activity by weekday × hour (user prompts)."""
+    cfg = config or get_default_config().graph
+    timestamps = collection.timestamps("user")
+
+    fig, ax, font_prop = _setup_single_axes(cfg)
+    if not timestamps:
+        ax.text(0.5, 0.5, "No data", ha="center", va="center", fontproperties=font_prop)
+        ax.set_axis_off()
+        return fig
+
+    grid: list[list[int]] = [[0 for _ in range(24)] for _ in range(7)]
+    for ts in timestamps:
+        dt = _ts_to_dt(ts, cfg)
+        grid[dt.weekday()][dt.hour] += 1
+
+    # Keep the axes frame for the heatmap.
+    ax.grid(False)
+    for side in ["top", "right", "left", "bottom"]:
+        ax.spines[side].set_visible(False)
+
+    img: AxesImage = ax.imshow(grid, aspect="auto", cmap="Blues", interpolation="nearest")
+
+    ax.set_title(
+        f"Activity heatmap (weekday × hour, {_tz_label(cfg)})",
+        fontproperties=font_prop,
+        fontsize=14,
+        pad=14,
+    )
+    ax.set_xlabel(f"Hour of day ({_tz_label(cfg)})", fontproperties=font_prop)
+    ax.set_ylabel("Weekday", fontproperties=font_prop)
+
+    ax.set_xticks(list(range(0, 24, 2)))
+    ax.set_xticklabels([f"{h:02d}" for h in range(0, 24, 2)], fontproperties=font_prop)
+    ax.set_yticks(list(range(7)))
+    ax.set_yticklabels(WEEKDAYS, fontproperties=font_prop)
+
+    cbar = fig.colorbar(img, ax=ax, fraction=0.046, pad=0.04)
+    cbar.set_label("User prompts", fontproperties=font_prop)
+    for t in cbar.ax.get_yticklabels():
+        t.set_fontproperties(font_prop)
+
+    fig.tight_layout()
+    return fig
+
+
+def generate_summary_dashboard(
+    collection: ConversationCollection,
+    config: GraphConfig | None = None,
+) -> Figure:
+    """Create a compact, high-signal overview dashboard."""
+    cfg = config or get_default_config().graph
+    font_prop = _load_font(cfg)
+
+    fig = Figure(figsize=(14, 9), dpi=cfg.dpi, facecolor="white")
+    gs = fig.add_gridspec(3, 2, height_ratios=[1.2, 1.0, 1.0], width_ratios=[1.25, 1.0])
+
+    ax_ts: Axes = fig.add_subplot(gs[0, :])
+    ax_heat: Axes = fig.add_subplot(gs[1:, 0])
+    ax_model: Axes = fig.add_subplot(gs[1, 1])
+    ax_len: Axes = fig.add_subplot(gs[2, 1])
+
+    for ax in (ax_ts, ax_model, ax_len):
+        _style_axes(ax, cfg)
+        _apply_tick_font(ax, font_prop)
+
+    # Header
+    user_ts = collection.timestamps("user")
+    conv_count = len(collection.conversations)
+    prompt_count = len(user_ts)
+
+    fig.text(
+        0.01,
+        0.985,
+        "ChatGPT usage overview",
+        fontproperties=font_prop,
+        fontsize=18,
+        va="top",
+        ha="left",
+        color="#24292f",
+    )
+
+    if user_ts:
+        dts = [_ts_to_dt(ts, cfg) for ts in user_ts]
+        date_range = f"{min(dts).date().isoformat()} → {max(dts).date().isoformat()}"
+    else:
+        date_range = "No activity"
+
+    fig.text(
+        0.01,
+        0.955,
+        f"{conv_count} conversations · {prompt_count} user prompts · {date_range} · {_tz_label(cfg)}",
+        fontproperties=font_prop,
+        fontsize=10,
+        va="top",
+        ha="left",
+        color="#57606a",
+    )
+
+    # Monthly activity (timeseries)
+    if user_ts:
+        month_counts = _aggregate_counts_by_month(user_ts, cfg)
+        months, values = _fill_missing_months(month_counts)
+        x = mdates.date2num(months)
+        ax_ts.bar(x, values, width=25, color=cfg.color, alpha=0.20, edgecolor="none")
+        smooth = _moving_average(values, window=3)
+        if smooth:
+            ax_ts.plot(x[2:], smooth, color=cfg.color, linewidth=2.6, alpha=0.95)
+
+        locator = mdates.AutoDateLocator(minticks=4, maxticks=10)
+        ax_ts.xaxis.set_major_locator(locator)
+        ax_ts.xaxis.set_major_formatter(mdates.ConciseDateFormatter(locator))
+        ax_ts.set_title("Monthly activity (user prompts)", fontproperties=font_prop, fontsize=13, pad=10)
+        ax_ts.set_ylabel("User prompts", fontproperties=font_prop)
+        ax_ts.set_xlabel(f"Month ({_tz_label(cfg)})", fontproperties=font_prop)
+        _apply_tick_font(ax_ts, font_prop)
+    else:
+        ax_ts.text(0.5, 0.5, "No data", ha="center", va="center", fontproperties=font_prop)
+        ax_ts.set_axis_off()
+
+    # Heatmap
+    if user_ts:
+        grid: list[list[int]] = [[0 for _ in range(24)] for _ in range(7)]
+        for ts in user_ts:
+            dt = _ts_to_dt(ts, cfg)
+            grid[dt.weekday()][dt.hour] += 1
+
+        ax_heat.grid(False)
+        for side in ["top", "right", "left", "bottom"]:
+            ax_heat.spines[side].set_visible(False)
+        img = ax_heat.imshow(grid, aspect="auto", cmap="Blues", interpolation="nearest")
+        ax_heat.set_title(
+            f"Weekday × hour heatmap ({_tz_label(cfg)})",
+            fontproperties=font_prop,
+            fontsize=13,
+            pad=10,
+        )
+        ax_heat.set_xlabel("Hour", fontproperties=font_prop)
+        ax_heat.set_ylabel("Weekday", fontproperties=font_prop)
+        ax_heat.set_xticks(list(range(0, 24, 3)))
+        ax_heat.set_xticklabels([f"{h:02d}" for h in range(0, 24, 3)], fontproperties=font_prop)
+        ax_heat.set_yticks(list(range(7)))
+        ax_heat.set_yticklabels(WEEKDAYS, fontproperties=font_prop)
+        cbar = fig.colorbar(img, ax=ax_heat, fraction=0.046, pad=0.04)
+        cbar.set_label("Prompts", fontproperties=font_prop)
+        for t in cbar.ax.get_yticklabels():
+            t.set_fontproperties(font_prop)
+    else:
+        ax_heat.text(0.5, 0.5, "No data", ha="center", va="center", fontproperties=font_prop)
+        ax_heat.set_axis_off()
+
+    # Model usage (reuse existing generator logic by drawing into its own axes)
+    model_counts: defaultdict[str, int] = defaultdict(int)
+    for conv in collection.conversations:
+        model_counts[conv.model or "Unknown"] += 1
+    total_models = sum(model_counts.values())
+    if total_models:
+        items = sorted(model_counts.items(), key=lambda x: x[1], reverse=True)
+        labels = [k for k, _ in items][:8]
+        counts = [v for _, v in items][:8]
+        y = list(range(len(labels)))[::-1]
+        bars = ax_model.barh(y, counts[::-1], color=cfg.color, alpha=0.9, height=0.6)
+        ax_model.set_yticks(y)
+        ax_model.set_yticklabels(labels[::-1], fontproperties=font_prop)
+        ax_model.set_xlabel("Conversations", fontproperties=font_prop)
+        ax_model.set_title("Models", fontproperties=font_prop, fontsize=13, pad=10)
+        for bar, count in zip(bars, counts[::-1], strict=True):
+            pct = 100 * (count / total_models)
+            ax_model.text(
+                bar.get_width(),
+                bar.get_y() + bar.get_height() / 2,
+                f"  {pct:.0f}%",
+                va="center",
+                ha="left",
+                fontproperties=font_prop,
+                fontsize=9,
+                color="#57606a",
+            )
+        _apply_tick_font(ax_model, font_prop)
+    else:
+        ax_model.text(0.5, 0.5, "No data", ha="center", va="center", fontproperties=font_prop)
+        ax_model.set_axis_off()
+
+    # Conversation length mini-hist
+    lengths = [conv.message_count("user") for conv in collection.conversations]
+    if lengths:
+        sorted_lengths = sorted(lengths)
+        cap = max(int(sorted_lengths[int(0.95 * (len(sorted_lengths) - 1))]), 5)
+        plot_lengths = [min(L, cap) for L in lengths]
+        ax_len.hist(
+            plot_lengths,
+            bins=min(16, max(8, cap // 2)),
+            color=cfg.color,
+            alpha=0.85,
+            rwidth=0.9,
+            edgecolor="white",
+            linewidth=0.5,
+        )
+        ax_len.set_title("Conversation length", fontproperties=font_prop, fontsize=13, pad=10)
+        ax_len.set_xlabel("User prompts", fontproperties=font_prop)
+        ax_len.set_ylabel("Conversations", fontproperties=font_prop)
+        ax_len.set_xlim(left=0, right=cap)
+        _apply_tick_font(ax_len, font_prop)
+    else:
+        ax_len.text(0.5, 0.5, "No data", ha="center", va="center", fontproperties=font_prop)
+        ax_len.set_axis_off()
+
+    fig.subplots_adjust(top=0.93, left=0.06, right=0.98, bottom=0.06, hspace=0.4, wspace=0.25)
+    return fig
+
+
+def generate_summary_graphs(
+    collection: ConversationCollection,
+    output_dir: Path,
+    config: GraphConfig | None = None,
+    *,
+    progress_bar: bool = False,
+) -> None:
+    """Generate all summary-level graphs.
+
+    Args:
+        collection: Collection of conversations
+        output_dir: Directory to save the graphs
+        config: Optional graph configuration
+    """
+    if not collection.conversations:
+        return
+
+    cfg = config or get_default_config().graph
+
+    user_ts = collection.timestamps("user")
+
+    tasks: list[tuple[str, str, Callable[[], Figure]]] = [
+        ("Overview", "overview.png", lambda: generate_summary_dashboard(collection, cfg)),
+        ("Activity heatmap", "activity_heatmap.png", lambda: generate_activity_heatmap(collection, cfg)),
+        ("Daily activity", "daily_activity.png", lambda: generate_daily_activity_lineplot(collection, cfg)),
+        ("Monthly activity", "monthly_activity.png", lambda: generate_monthly_activity_barplot(collection, cfg)),
+        ("Model usage", "model_usage.png", lambda: generate_model_piechart(collection, cfg)),
+        ("Conversation lengths", "conversation_lengths.png", lambda: generate_length_histogram(collection, cfg)),
+        (
+            "Conversation lifetimes",
+            "conversation_lifetimes.png",
+            lambda: generate_conversation_lifetime_histogram(collection, cfg),
+        ),
+    ]
+
+    if user_ts:
+        tasks.extend(
+            [
+                (
+                    "Weekday pattern",
+                    "weekday_pattern.png",
+                    lambda: generate_week_barplot(user_ts, "All time", cfg),
+                ),
+                (
+                    "Hourly pattern",
+                    "hourly_pattern.png",
+                    lambda: generate_hour_barplot(user_ts, "All time", cfg),
+                ),
+            ]
+        )
+
+    for _, filename, build in tqdm(
+        tasks,
+        desc="Creating summary graphs",
+        disable=not progress_bar,
+    ):
+        fig = build()
+        fig.savefig(output_dir / filename, facecolor="white")
+
+
+def generate_graphs(
+    collection: ConversationCollection,
+    output_dir: Path,
+    config: GraphConfig | None = None,
+    *,
+    progress_bar: bool = False,
+) -> None:
+    """Generate weekly, hourly, and summary graphs.
+
+    Args:
+        collection: Collection of conversations
+        output_dir: Directory to save the graphs
+        config: Optional graph configuration
+        progress_bar: Whether to show progress bars
+    """
+    output_dir.mkdir(parents=True, exist_ok=True)
+    cfg = config or get_default_config().graph
+
+    # Summary graphs (default: small, high-signal set)
+    generate_summary_graphs(collection, output_dir, cfg, progress_bar=progress_bar)
+
+    # Optional breakdowns (can generate lots of files; off by default)
+    if not collection.conversations:
+        return
+
+    timestamps = collection.timestamps("user")
+    if not timestamps:
+        return
+
+    breakdown_root = output_dir / "Breakdowns"
+    if cfg.generate_monthly_breakdowns:
+        monthly_dir = breakdown_root / "Monthly"
+        monthly_dir.mkdir(parents=True, exist_ok=True)
+
+        month_groups: defaultdict[datetime, list[float]] = defaultdict(list)
+        for ts in timestamps:
+            dt = _ts_to_dt(ts, cfg)
+            month_groups[_month_start(dt)].append(ts)
+
+        for month, ts_list in tqdm(
+            sorted(month_groups.items(), key=lambda x: x[0]),
+            desc="Creating monthly breakdown graphs",
+            disable=not progress_bar,
+        ):
+            slug = month.strftime("%Y-%m")
+            title = month.strftime("%b %Y")
+            generate_week_barplot(ts_list, title, cfg).savefig(
+                monthly_dir / f"{slug}_weekday.png", facecolor="white"
+            )
+            generate_hour_barplot(ts_list, title, cfg).savefig(
+                monthly_dir / f"{slug}_hourly.png", facecolor="white"
+            )
+
+    if cfg.generate_yearly_breakdowns:
+        yearly_dir = breakdown_root / "Yearly"
+        yearly_dir.mkdir(parents=True, exist_ok=True)
+
+        year_groups: defaultdict[datetime, list[float]] = defaultdict(list)
+        for ts in timestamps:
+            dt = _ts_to_dt(ts, cfg)
+            year_groups[_year_start(dt)].append(ts)
+
+        for year, ts_list in tqdm(
+            sorted(year_groups.items(), key=lambda x: x[0]),
+            desc="Creating yearly breakdown graphs",
+            disable=not progress_bar,
+        ):
+            slug = year.strftime("%Y")
+            title = year.strftime("%Y")
+            generate_week_barplot(ts_list, title, cfg).savefig(
+                yearly_dir / f"{slug}_weekday.png", facecolor="white"
+            )
+            generate_hour_barplot(ts_list, title, cfg).savefig(
+                yearly_dir / f"{slug}_hourly.png", facecolor="white"
+            )