docs: update README for model tracking feature

Document new Models tab, /nodes-stats/models endpoint, model_usage
table, updated How It Works diagram, and file structure.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

README.md

@@ -4,14 +4,16 @@
   <img src="docs/logo.svg" width="120" alt="Node Stats logo">
 </p>
 
-A ComfyUI custom node package that silently tracks which nodes and packages you actually use. Helps identify unused packages that are safe to remove — keeping your ComfyUI install lean.
+A ComfyUI custom node package that silently tracks which nodes, packages, and model files you actually use. Helps identify unused packages and models that are safe to remove — keeping your ComfyUI install lean.
 
 ## Features
 
 - **Silent tracking** — hooks into every prompt submission, zero config needed
+- **Node & model tracking** — tracks custom node packages and model files (checkpoints, VAEs, ControlNets, etc.) separately
 - **Per-package classification** — packages are sorted into tiers based on usage recency
-- **Smart aging** — packages gradually move from "recently unused" to "safe to remove" over time
-- **Uninstall detection** — removed packages are flagged separately, historical data preserved
+- **Per-model classification** — models grouped by type (checkpoints, vae, …) with the same recency tiers
+- **Smart aging** — items gradually move from "recently unused" to "safe to remove" over time
+- **Uninstall detection** — removed packages/models are flagged separately, historical data preserved
 - **Expandable detail** — click any package to see individual node-level stats
 - **Non-blocking** — DB writes happen in a background thread, no impact on workflow execution
 
@@ -55,11 +57,17 @@ Restart ComfyUI. Tracking starts immediately and silently.
 
 ### UI
 
-Click the **Node Stats** button (bar chart icon) in the ComfyUI top menu bar. A dialog shows:
+Click the **Node Stats** button (bar chart icon) in the ComfyUI top menu bar. A dialog opens with two tabs:
 
-- **Summary bar** with counts for each classification tier
-- **Sections** for each tier, sorted from most actionable to least
-- **Expandable rows** — click any package to see per-node execution counts and timestamps
+**Nodes tab**
+- Summary bar with counts for each classification tier
+- Sections for each tier, sorted from most actionable to least
+- Expandable rows — click any package to see per-node execution counts and timestamps
+
+**Models tab**
+- Summary bar with counts for each tier across all model types
+- Sections per model type (checkpoints, vae, controlnet, …)
+- Per-model table showing execution count, last used date, and status
 
 ### API
 
@@ -67,6 +75,7 @@ Click the **Node Stats** button (bar chart icon) in the ComfyUI top menu bar. A
 |----------|--------|-------------|
 | `/nodes-stats/packages` | GET | Per-package aggregated stats with classification |
 | `/nodes-stats/usage` | GET | Raw per-node usage data |
+| `/nodes-stats/models` | GET | Per-type model stats with classification |
 | `/nodes-stats/reset` | POST | Clear all tracked data |
 
 ```bash
@@ -114,42 +123,46 @@ curl http://localhost:8188/nodes-stats/packages | python3 -m json.tool
 ## How It Works
 
 ```
-Queue Prompt ──> Prompt Handler ──> Extract class_types ──> Background Thread
+Queue Prompt ──> Prompt Handler ──> Extract class_types + prompt ──> Background Thread
                                                                               │
-                                         ┌─────────────────────────┘
+                                              ┌───────────────────────────────┘
                                               ▼
                                          SQLite DB
                                         usage_stats.db
-                                    ┌──────────┐
-                                    │node_usage │  per-node counts & timestamps
-                                    │prompt_log │  full node list per prompt
-                                    └──────────┘
+                                         ┌─────────────┐
+                                         │ node_usage  │  per-node counts & timestamps
+                                         │ prompt_log  │  full node list per prompt
+                                         │ model_usage │  per-model counts & timestamps
+                                         └─────────────┘
                                               │
-            GET /nodes-stats/packages ◄──┘
+          GET /nodes-stats/packages ◄─────────┤
+          GET /nodes-stats/models   ◄─────────┘
                     │
                     ▼
-               Mapper merges DB data
-               with NODE_CLASS_MAPPINGS
+          Merge DB data with installed
+          nodes/models, classify by recency
                     │
                     ▼
-              Classify by recency ──> JSON response ──> UI Dialog
+          JSON response ──> UI Dialog (Nodes tab / Models tab)
 ```
 
 1. Registers a prompt handler via `PromptServer.instance.add_on_prompt_handler()`
-2. On every prompt submission, extracts `class_type` from each node in the workflow
+2. On every prompt submission, extracts `class_type` from each node and the full prompt dict
 3. Offloads recording to a background thread (non-blocking)
 4. Maps each class_type to its source package using `RELATIVE_PYTHON_MODULE`
-5. Upserts per-node counts and timestamps into SQLite
-6. On stats request, merges DB data with current node registry and classifies by recency
+5. Detects model file selections by introspecting each node's `INPUT_TYPES()` for folder-dropdown inputs, then resolves filenames via `folder_paths`
+6. Upserts per-node and per-model counts and timestamps into SQLite
+7. On stats request, merges DB data with current node registry / installed models and classifies by recency
 
 ## Data Storage
 
-All data is stored in `usage_stats.db` in the package directory.
+All data is stored in `<ComfyUI user dir>/nodes_stats/usage_stats.db` (survives extension reinstalls).
 
 | Table | Contents |
 |-------|----------|
 | `node_usage` | Per-node: class_type, package, execution count, first/last seen |
 | `prompt_log` | Per-prompt: timestamp, JSON array of all class_types used |
+| `model_usage` | Per-model: filename, type, execution count, first/last seen |
 
 Use `POST /nodes-stats/reset` to clear all data and start fresh.
 
@@ -157,8 +170,9 @@ Use `POST /nodes-stats/reset` to clear all data and start fresh.
 
 ```
 __init__.py        Entry point: prompt handler, API routes
-mapper.py          class_type → package name mapping
+mapper.py          class_type → package mapping; model filename → type mapping
 tracker.py         SQLite persistence and stats aggregation
-js/nodes_stats.js  Frontend: menu button + stats dialog
+js/nodes_stats.js  Frontend: menu button + stats dialog (Nodes/Models tabs)
 pyproject.toml     Package metadata
+tests/             Unit tests for tracker and mapper
 ```

__init__.py

@@ -4,7 +4,7 @@ import threading
 from aiohttp import web
 from server import PromptServer
 
-from .mapper import NodePackageMapper
+from .mapper import NodePackageMapper, ModelMapper
 from .tracker import UsageTracker
 
 logger = logging.getLogger(__name__)
@@ -14,10 +14,11 @@ WEB_DIRECTORY = "./js"
 
 mapper = NodePackageMapper()
 tracker = UsageTracker()
+model_mapper = ModelMapper()
 
 
 def on_prompt_handler(json_data):
-    """Called on every prompt submission. Extracts class_types and records usage."""
+    """Called on every prompt submission. Extracts class_types and queues recording."""
     try:
         prompt = json_data.get("prompt", {})
         class_types = set()
@@ -26,9 +27,11 @@ def on_prompt_handler(json_data):
             if ct:
                 class_types.add(ct)
         if class_types:
+            # Pass the full prompt to the thread — model extraction (which calls
+            # INPUT_TYPES() on every node) happens off the main request thread.
             threading.Thread(
-                target=tracker.record_usage,
-                args=(class_types, mapper),
+                target=_record_prompt,
+                args=(class_types, prompt),
                 daemon=True,
             ).start()
     except Exception:
@@ -36,6 +39,19 @@ def on_prompt_handler(json_data):
     return json_data
 
 
+def _record_prompt(class_types, prompt):
+    try:
+        tracker.record_usage(class_types, mapper)
+    except Exception:
+        logger.warning("nodes-stats: error recording node usage", exc_info=True)
+    try:
+        models = model_mapper.extract_models_from_prompt(prompt)
+        if models:
+            tracker.record_model_usage(models)
+    except Exception:
+        logger.warning("nodes-stats: error recording model usage", exc_info=True)
+
+
 PromptServer.instance.add_on_prompt_handler(on_prompt_handler)
 
 
@@ -62,11 +78,23 @@ async def get_node_stats(request):
         return web.json_response({"error": "internal error"}, status=500)
 
 
+@routes.get("/nodes-stats/models")
+async def get_model_stats(request):
+    try:
+        installed_by_type = model_mapper.get_all_models()
+        stats = tracker.get_model_stats(installed_by_type)
+        return web.json_response(stats)
+    except Exception:
+        logger.error("nodes-stats: error getting model stats", exc_info=True)
+        return web.json_response({"error": "internal error"}, status=500)
+
+
 @routes.post("/nodes-stats/reset")
 async def reset_stats(request):
     try:
         tracker.reset()
         mapper.invalidate()
+        model_mapper.invalidate()
         return web.json_response({"status": "ok"})
     except Exception:
         logger.error("nodes-stats: error resetting stats", exc_info=True)

docs/plans/2026-04-08-model-tracking-design.md

@@ -0,0 +1,117 @@
+# Model Usage Tracking — Design
+
+**Date:** 2026-04-08
+**Status:** Approved
+
+## Goal
+
+Extend ComfyUI Node Stats to also track which model files are used across prompts, and surface models that are never used — making it easy to identify files that can be safely deleted.
+
+**Scope:** All model folder types registered in `folder_paths` except LoRAs (and non-model folders: `configs`, `custom_nodes`, `temp`, `output`, `input`). Discovered dynamically so custom-node-added types are picked up automatically.
+
+## Design Decisions
+
+- **Tracked per filename globally** — `dreamshaper.safetensors` is one entry regardless of which node loads it. `model_type` is stored for display grouping only.
+- **Detection via node introspection (Approach A)** — At prompt time, for each node look up its `INPUT_TYPES()`, find inputs whose type is a list (ComfyUI folder dropdown pattern), cross-reference with `folder_paths` to identify the folder type, extract the selected value.
+- **UI: new tab** in the existing dialog alongside the current "Nodes" tab.
+- **Same tier classification** as packages: `used`, `unused_new`, `consider_removing`, `safe_to_remove`, `uninstalled`.
+- **Existing node stats are unaffected** — additive only.
+
+## Data Collection
+
+In `on_prompt_handler` (`__init__.py`), after extracting `class_types`, also extract model references:
+
+1. For each node in the prompt, look up its class in `nodes.NODE_CLASS_MAPPINGS`
+2. Call `INPUT_TYPES()` and find inputs whose declared type is a `list`
+3. Map that list back to its `folder_paths` folder type (via reverse lookup built at startup)
+4. Extract the actual selected value from the prompt's `inputs` dict
+5. Record `(model_name, model_type)` via background thread
+
+A `ModelMapper` class (in `mapper.py` or new file) handles the reverse lookup: `folder_type → set of filenames`, cached and invalidated on reset.
+
+Excluded folder types (not model files):
+```python
+EXCLUDED_FOLDER_TYPES = {"loras", "configs", "custom_nodes", "temp", "output", "input", "upscale_models"}
+```
+> Note: `upscale_models` may be included/excluded per preference; default include.
+
+## Storage
+
+New table in existing `usage_stats.db`:
+
+```sql
+CREATE TABLE IF NOT EXISTS model_usage (
+    model_name TEXT PRIMARY KEY,
+    model_type TEXT NOT NULL,
+    count INTEGER NOT NULL DEFAULT 0,
+    first_seen TEXT NOT NULL,
+    last_seen TEXT NOT NULL
+);
+CREATE INDEX IF NOT EXISTS idx_model_usage_type ON model_usage(model_type);
+```
+
+Same upsert pattern as `node_usage`. `reset` clears this table too.
+
+## API
+
+New endpoint:
+
+```
+GET /nodes-stats/models
+```
+
+Returns a list of model types, each containing classified models:
+
+```json
+[
+  {
+    "model_type": "checkpoints",
+    "models": [
+      {
+        "model_name": "dreamshaper.safetensors",
+        "count": 42,
+        "first_seen": "2026-01-01T00:00:00+00:00",
+        "last_seen": "2026-03-01T00:00:00+00:00",
+        "installed": true,
+        "status": "used"
+      },
+      {
+        "model_name": "old_model.ckpt",
+        "count": 0,
+        "first_seen": null,
+        "last_seen": null,
+        "installed": true,
+        "status": "safe_to_remove"
+      }
+    ]
+  }
+]
+```
+
+"All installed models" sourced from `folder_paths.get_filename_list(type)` for each tracked type. Models on disk but never seen appear with `count: 0`.
+
+Existing endpoints unchanged:
+- `GET /nodes-stats/packages`
+- `GET /nodes-stats/usage`
+- `POST /nodes-stats/reset` — extended to also clear `model_usage`
+
+## UI
+
+Two tabs at top of dialog: **Nodes** (existing content, untouched) | **Models** (new).
+
+Models tab layout:
+- Summary badge bar: counts per status tier across all model types
+- One section per model type (only types with ≥1 model shown), titled e.g. "Checkpoints", "VAE", "ControlNet"
+- Within each section: models sorted by status tier (safe_to_remove → consider_removing → unused_new → used), then alphabetically
+- Each row: model name | execution count | last used date | status color
+- No expandable rows (models are leaves)
+- Uninstalled models: collapsed section at the bottom, same pattern as node packages
+
+## File Changes
+
+| File | Change |
+|------|--------|
+| `tracker.py` | Add `model_usage` table to schema; add `record_model_usage()`, `get_model_stats()` methods; extend `reset()` |
+| `mapper.py` | Add `ModelMapper` class with folder-type reverse lookup and model filename introspection |
+| `__init__.py` | Extend `on_prompt_handler` to extract and record model usage; add `GET /nodes-stats/models` endpoint |
+| `js/nodes_stats.js` | Add tab switcher UI; add Models tab rendering (summary badges + per-type sections) |

js/nodes_stats.js

@@ -34,23 +34,27 @@ app.registerExtension({
 });
 
 async function showStatsDialog() {
-  let data;
+  let data, modelData;
   try {
-    const resp = await fetch("/nodes-stats/packages");
-    if (!resp.ok) {
-      alert("Failed to load node stats: HTTP " + resp.status);
-      return;
-    }
-    data = await resp.json();
-    if (!Array.isArray(data)) {
-      alert("Failed to load node stats: unexpected response format");
+    const [pkgResp, modelResp] = await Promise.all([
+      fetch("/nodes-stats/packages"),
+      fetch("/nodes-stats/models"),
+    ]);
+    if (!pkgResp.ok) { alert("Failed to load node stats: HTTP " + pkgResp.status); return; }
+    if (!modelResp.ok) { alert("Failed to load model stats: HTTP " + modelResp.status); return; }
+    data = await pkgResp.json();
+    modelData = await modelResp.json();
+    if (!Array.isArray(data) || !Array.isArray(modelData)) {
+      alert("Failed to load stats: unexpected response format");
       return;
     }
   } catch (e) {
-    alert("Failed to load node stats: " + e.message);
+    alert("Failed to load stats: " + e.message);
     return;
   }
 
+  const custom = data.filter((p) => p.package !== "__builtin__");
+
   // Remove existing dialog if any
   const existing = document.getElementById("nodes-stats-dialog");
   if (existing) existing.remove();
@@ -67,69 +71,55 @@ async function showStatsDialog() {
   dialog.style.cssText =
     "background:#1e1e1e;color:#ddd;border-radius:8px;padding:24px;max-width:800px;width:90%;max-height:85vh;overflow-y:auto;font-family:monospace;font-size:13px;";
 
-  const custom = data.filter((p) => p.package !== "__builtin__");
-  const safeToRemove = custom.filter((p) => p.status === "safe_to_remove");
-  const considerRemoving = custom.filter((p) => p.status === "consider_removing");
-  const unusedNew = custom.filter((p) => p.status === "unused_new");
-  const used = custom.filter((p) => p.status === "used");
-  const uninstalled = custom.filter((p) => p.status === "uninstalled");
-
   let html = `<div style="display:flex;justify-content:space-between;align-items:center;margin-bottom:16px;">
-    <h2 style="margin:0;color:#fff;font-size:18px;">Node Package Stats</h2>
+    <h2 style="margin:0;color:#fff;font-size:18px;">Usage Stats</h2>
     <button id="nodes-stats-close" style="background:none;border:none;color:#888;font-size:20px;cursor:pointer;">&times;</button>
   </div>`;
 
-  html += `<div style="display:flex;gap:10px;margin-bottom:20px;flex-wrap:wrap;">
-    <div style="background:#3a1a1a;padding:8px 14px;border-radius:4px;border-left:3px solid #e44;">
-      <span style="font-size:20px;font-weight:bold;color:#e44;">${safeToRemove.length}</span>
-      <span style="color:#c99;margin-left:6px;">safe to remove</span>
-    </div>
-    <div style="background:#2a2215;padding:8px 14px;border-radius:4px;border-left:3px solid #e90;">
-      <span style="font-size:20px;font-weight:bold;color:#e90;">${considerRemoving.length}</span>
-      <span style="color:#ca8;margin-left:6px;">consider removing</span>
-    </div>
-    <div style="background:#1a1a2a;padding:8px 14px;border-radius:4px;border-left:3px solid #68f;">
-      <span style="font-size:20px;font-weight:bold;color:#68f;">${unusedNew.length}</span>
-      <span style="color:#99b;margin-left:6px;">unused &lt;1 month</span>
-    </div>
-    <div id="nodes-stats-used-badge" style="background:#1a2a1a;padding:8px 14px;border-radius:4px;border-left:3px solid #4a4;cursor:default;user-select:none;">
-      <span style="font-size:20px;font-weight:bold;color:#4a4;">${used.length}</span>
-      <span style="color:#9c9;margin-left:6px;">used</span>
-    </div>
+  // Tab switcher — wired via addEventListener after insertion, no onclick globals
+  html += `
+  <div style="display:flex;gap:0;margin-bottom:20px;border-bottom:1px solid #333;">
+    <button id="ns-tab-nodes"
+      style="background:none;border:none;border-bottom:2px solid #4a4;color:#4a4;padding:8px 18px;cursor:pointer;font-family:monospace;font-size:13px;font-weight:bold;">
+      Nodes
+    </button>
+    <button id="ns-tab-models"
+      style="background:none;border:none;border-bottom:2px solid transparent;color:#888;padding:8px 18px;cursor:pointer;font-family:monospace;font-size:13px;">
+      Models
+    </button>
   </div>`;
 
-  if (safeToRemove.length > 0) {
-    html += sectionHeader("Safe to Remove", "Unused for 2+ months", "#e44");
-    html += buildTable(safeToRemove, "safe_to_remove");
-  }
+  // Nodes tab content (existing content, wrapped)
+  html += `<div id="ns-content-nodes">`;
+  html += buildNodesTabContent(custom);
+  html += `</div>`;
 
-  if (considerRemoving.length > 0) {
-    html += sectionHeader("Consider Removing", "Unused for 1-2 months", "#e90");
-    html += buildTable(considerRemoving, "consider_removing");
-  }
-
-  if (unusedNew.length > 0) {
-    html += sectionHeader("Recently Unused", "Unused for less than 1 month", "#68f");
-    html += buildTable(unusedNew, "unused_new");
-  }
-
-  if (used.length > 0) {
-    html += sectionHeader("Used", "", "#4a4");
-    html += buildTable(used, "used");
-  }
-
-  if (uninstalled.length > 0) {
-    html += sectionHeader("Uninstalled", "Previously tracked, no longer installed", "#555");
-    html += buildTable(uninstalled, "uninstalled");
-  }
+  // Models tab content
+  html += `<div id="ns-content-models" style="display:none;">`;
+  html += buildModelsTabContent(modelData);
+  html += `</div>`;
 
   dialog.innerHTML = html;
   overlay.appendChild(dialog);
   document.body.appendChild(overlay);
 
-  document
-    .getElementById("nodes-stats-close")
-    .addEventListener("click", () => overlay.remove());
+  // Tab switch — local function, no window pollution
+  function switchTab(tab) {
+    dialog.querySelector("#ns-content-nodes").style.display = tab === "nodes" ? "" : "none";
+    dialog.querySelector("#ns-content-models").style.display = tab === "models" ? "" : "none";
+    const nodeBtn = dialog.querySelector("#ns-tab-nodes");
+    const modelBtn = dialog.querySelector("#ns-tab-models");
+    nodeBtn.style.borderBottomColor = tab === "nodes" ? "#4a4" : "transparent";
+    nodeBtn.style.color = tab === "nodes" ? "#4a4" : "#888";
+    nodeBtn.style.fontWeight = tab === "nodes" ? "bold" : "normal";
+    modelBtn.style.borderBottomColor = tab === "models" ? "#4a4" : "transparent";
+    modelBtn.style.color = tab === "models" ? "#4a4" : "#888";
+    modelBtn.style.fontWeight = tab === "models" ? "bold" : "normal";
+  }
+  dialog.querySelector("#ns-tab-nodes").addEventListener("click", () => switchTab("nodes"));
+  dialog.querySelector("#ns-tab-models").addEventListener("click", () => switchTab("models"));
+
+  dialog.querySelector("#nodes-stats-close").addEventListener("click", () => overlay.remove());
 
   // Toggle expandable rows
   dialog.querySelectorAll(".pkg-row").forEach((row) => {
@@ -148,7 +138,7 @@ async function showStatsDialog() {
   // Easter egg: click "used" badge 5 times to show podium
   let eggClicks = 0;
   let eggTimer = null;
-  const usedBadge = document.getElementById("nodes-stats-used-badge");
+  const usedBadge = dialog.querySelector("#nodes-stats-used-badge");
   if (usedBadge) {
     usedBadge.addEventListener("click", () => {
       eggClicks++;
@@ -165,6 +155,116 @@ async function showStatsDialog() {
   }
 }
 
+function buildNodesTabContent(custom) {
+  const safeToRemove     = custom.filter((p) => p.status === "safe_to_remove");
+  const considerRemoving = custom.filter((p) => p.status === "consider_removing");
+  const unusedNew        = custom.filter((p) => p.status === "unused_new");
+  const used             = custom.filter((p) => p.status === "used");
+  const uninstalled      = custom.filter((p) => p.status === "uninstalled");
+
+  let html = `<div style="display:flex;gap:10px;margin-bottom:20px;flex-wrap:wrap;">
+    <div style="background:#3a1a1a;padding:8px 14px;border-radius:4px;border-left:3px solid #e44;">
+      <span style="font-size:20px;font-weight:bold;color:#e44;">${safeToRemove.length}</span>
+      <span style="color:#c99;margin-left:6px;">safe to remove</span>
+    </div>
+    <div style="background:#2a2215;padding:8px 14px;border-radius:4px;border-left:3px solid #e90;">
+      <span style="font-size:20px;font-weight:bold;color:#e90;">${considerRemoving.length}</span>
+      <span style="color:#ca8;margin-left:6px;">consider removing</span>
+    </div>
+    <div style="background:#1a1a2a;padding:8px 14px;border-radius:4px;border-left:3px solid #68f;">
+      <span style="font-size:20px;font-weight:bold;color:#68f;">${unusedNew.length}</span>
+      <span style="color:#99b;margin-left:6px;">unused &lt;1 month</span>
+    </div>
+    <div id="nodes-stats-used-badge" style="background:#1a2a1a;padding:8px 14px;border-radius:4px;border-left:3px solid #4a4;cursor:default;user-select:none;">
+      <span style="font-size:20px;font-weight:bold;color:#4a4;">${used.length}</span>
+      <span style="color:#9c9;margin-left:6px;">used</span>
+    </div>
+  </div>`;
+
+  if (safeToRemove.length > 0)     html += sectionHeader("Safe to Remove", "Unused for 2+ months", "#e44")              + buildTable(safeToRemove, "safe_to_remove");
+  if (considerRemoving.length > 0) html += sectionHeader("Consider Removing", "Unused for 1-2 months", "#e90")           + buildTable(considerRemoving, "consider_removing");
+  if (unusedNew.length > 0)        html += sectionHeader("Recently Unused", "Unused for less than 1 month", "#68f")      + buildTable(unusedNew, "unused_new");
+  if (used.length > 0)             html += sectionHeader("Used", "", "#4a4")                                             + buildTable(used, "used");
+  if (uninstalled.length > 0)      html += sectionHeader("Uninstalled", "Previously tracked, no longer installed", "#555") + buildTable(uninstalled, "uninstalled");
+
+  return html;
+}
+
+function buildModelsTabContent(modelData) {
+  // Flatten for summary counts
+  const allModels      = modelData.flatMap((g) => g.models);
+  const safeCount      = allModels.filter((m) => m.status === "safe_to_remove").length;
+  const considerCount  = allModels.filter((m) => m.status === "consider_removing").length;
+  const unusedNewCount = allModels.filter((m) => m.status === "unused_new").length;
+  const usedCount      = allModels.filter((m) => m.status === "used").length;
+
+  let html = `<div style="display:flex;gap:10px;margin-bottom:20px;flex-wrap:wrap;">
+    <div style="background:#3a1a1a;padding:8px 14px;border-radius:4px;border-left:3px solid #e44;">
+      <span style="font-size:20px;font-weight:bold;color:#e44;">${safeCount}</span>
+      <span style="color:#c99;margin-left:6px;">safe to remove</span>
+    </div>
+    <div style="background:#2a2215;padding:8px 14px;border-radius:4px;border-left:3px solid #e90;">
+      <span style="font-size:20px;font-weight:bold;color:#e90;">${considerCount}</span>
+      <span style="color:#ca8;margin-left:6px;">consider removing</span>
+    </div>
+    <div style="background:#1a1a2a;padding:8px 14px;border-radius:4px;border-left:3px solid #68f;">
+      <span style="font-size:20px;font-weight:bold;color:#68f;">${unusedNewCount}</span>
+      <span style="color:#99b;margin-left:6px;">unused &lt;1 month</span>
+    </div>
+    <div style="background:#1a2a1a;padding:8px 14px;border-radius:4px;border-left:3px solid #4a4;">
+      <span style="font-size:20px;font-weight:bold;color:#4a4;">${usedCount}</span>
+      <span style="color:#9c9;margin-left:6px;">used</span>
+    </div>
+  </div>`;
+
+  if (allModels.length === 0) {
+    html += `<p style="color:#666;">No models tracked yet. Run a workflow to start.</p>`;
+    return html;
+  }
+
+  for (const group of modelData) {
+    if (group.models.length === 0) continue;
+    const title = group.model_type.charAt(0).toUpperCase() + group.model_type.slice(1).replace(/_/g, " ");
+    html += sectionHeader(title, `${group.models.length} model${group.models.length !== 1 ? "s" : ""}`, "#4a4");
+    html += buildModelTable(group.models);
+  }
+
+  return html;
+}
+
+function buildModelTable(models) {
+  let html = `<table style="width:100%;border-collapse:collapse;margin-bottom:12px;">
+    <thead><tr style="color:#888;text-align:left;border-bottom:1px solid #333;">
+      <th style="padding:6px 8px;">Model</th>
+      <th style="padding:6px 8px;text-align:right;">Executions</th>
+      <th style="padding:6px 8px;">Last Used</th>
+      <th style="padding:6px 8px;">Status</th>
+    </tr></thead><tbody>`;
+
+  for (const m of models) {
+    const { bg, hover } = STATUS_COLORS[m.status] || STATUS_COLORS.used;
+    const lastSeen = m.last_seen ? new Date(m.last_seen).toLocaleDateString() : "—";
+    const statusLabel = {
+      safe_to_remove:    { text: "safe to remove",    color: "#e44" },
+      consider_removing: { text: "consider removing", color: "#e90" },
+      unused_new:        { text: "unused <1mo",       color: "#68f" },
+      used:              { text: "used",               color: "#4a4" },
+      uninstalled:       { text: "uninstalled",        color: "#555" },
+    }[m.status] || { text: m.status, color: "#888" };
+
+    html += `<tr style="background:${bg};border-bottom:1px solid #222;"
+      onmouseover="this.style.background='${hover}'" onmouseout="this.style.background='${bg}'">
+      <td style="padding:6px 8px;color:#fff;">${escapeHtml(m.model_name)}</td>
+      <td style="padding:6px 8px;text-align:right;">${m.count}</td>
+      <td style="padding:6px 8px;color:#888;">${lastSeen}</td>
+      <td style="padding:6px 8px;"><span style="color:${statusLabel.color};font-size:11px;">${statusLabel.text}</span></td>
+    </tr>`;
+  }
+
+  html += `</tbody></table>`;
+  return html;
+}
+
 function sectionHeader(title, subtitle, color) {
   let html = `<h3 style="color:${color};margin:16px 0 8px;font-size:14px;">${title}`;
   if (subtitle) html += ` <span style="color:#666;font-size:12px;font-weight:normal;">— ${subtitle}</span>`;

mapper.py

@@ -66,3 +66,119 @@ class NodePackageMapper:
     def invalidate(self):
         """Force rebuild on next access (e.g. after node reload)."""
         self._map = None
+
+
+# Folder types that are not model files and should not be tracked
+EXCLUDED_FOLDER_TYPES = {
+    "loras",
+    "configs",
+    "custom_nodes",
+    "temp",
+    "output",
+    "input",
+    "annotators",
+    "assets",
+}
+
+
+class ModelMapper:
+    """Tracks which folder_paths model types exist and resolves filenames to types."""
+
+    def __init__(self):
+        self._folder_files = None   # {folder_type: frozenset(filenames)}
+        self._reverse = None        # {filename: folder_type}
+
+    def _build(self):
+        try:
+            import folder_paths
+
+            folder_files = {}
+            for folder_type in folder_paths.folder_names_and_paths:
+                if folder_type in EXCLUDED_FOLDER_TYPES:
+                    continue
+                try:
+                    files = folder_paths.get_filename_list(folder_type)
+                except Exception:
+                    files = []
+                if files:
+                    folder_files[folder_type] = frozenset(files)
+
+            # Build reverse map: filename -> folder_type (last write wins on collision)
+            reverse = {}
+            for folder_type, files in folder_files.items():
+                for f in files:
+                    reverse[f] = folder_type
+
+            # Assign atomically only on success
+            self._folder_files = folder_files
+            self._reverse = reverse
+
+        except Exception:
+            logger.warning("ModelMapper: failed to build model map", exc_info=True)
+            self._folder_files = {}
+            self._reverse = {}
+
+    def _ensure(self):
+        if self._folder_files is None:
+            self._build()
+
+    def get_model_type(self, filename):
+        """Return the folder type for a filename, or None if not tracked."""
+        self._ensure()
+        return self._reverse.get(filename)
+
+    def get_all_models(self):
+        """Return {folder_type: [filename, ...]} for all tracked types."""
+        self._ensure()
+        return {k: sorted(v) for k, v in self._folder_files.items()}
+
+    def extract_models_from_prompt(self, prompt):
+        """Scan a prompt dict and return (model_name, model_type) pairs.
+
+        For each node, inspects INPUT_TYPES() to find list-type (folder dropdown)
+        inputs, then resolves the selected value against the folder_paths reverse map.
+        """
+        self._ensure()
+        try:
+            import nodes as comfy_nodes
+        except ImportError:
+            return []
+
+        seen = set()
+        results = []
+
+        for node_data in prompt.values():
+            class_type = node_data.get("class_type")
+            node_inputs = node_data.get("inputs", {})
+            if not class_type or not node_inputs:
+                continue
+
+            node_cls = comfy_nodes.NODE_CLASS_MAPPINGS.get(class_type)
+            if node_cls is None:
+                continue
+
+            try:
+                input_types = node_cls.INPUT_TYPES()
+                for category in ("required", "optional"):
+                    for input_name, input_def in input_types.get(category, {}).items():
+                        if not isinstance(input_def, (list, tuple)) or not input_def:
+                            continue
+                        # ComfyUI folder dropdowns have a list as their type
+                        if not isinstance(input_def[0], list):
+                            continue
+                        value = node_inputs.get(input_name)
+                        if not isinstance(value, str) or value in seen:
+                            continue
+                        model_type = self.get_model_type(value)
+                        if model_type:
+                            seen.add(value)
+                            results.append((value, model_type))
+            except Exception:
+                continue
+
+        return results
+
+    def invalidate(self):
+        """Force rebuild on next access."""
+        self._folder_files = None
+        self._reverse = None

tests/conftest.py

@@ -0,0 +1,15 @@
+# tests/conftest.py
+import sys
+import os
+from unittest.mock import MagicMock
+
+# Put the project root on sys.path so tests can import tracker, mapper directly
+sys.path.insert(0, os.path.dirname(os.path.dirname(__file__)))
+
+# Stub ComfyUI-only modules before any test file imports project code
+_folder_paths_mock = MagicMock()
+_folder_paths_mock.get_user_directory.return_value = "/tmp"
+sys.modules["folder_paths"] = _folder_paths_mock
+for mod in ("nodes", "server"):
+    if mod not in sys.modules:
+        sys.modules[mod] = MagicMock()

tests/test_model_mapper.py

@@ -0,0 +1,96 @@
+import pytest
+from unittest.mock import patch, MagicMock
+from mapper import ModelMapper
+
+
+FAKE_FOLDER_NAMES = {
+    "checkpoints": ([], {}),
+    "vae": ([], {}),
+    "loras": ([], {}),
+    "configs": ([], {}),
+}
+
+FAKE_FILES = {
+    "checkpoints": ["dream.safetensors", "v15.ckpt"],
+    "vae": ["vae.safetensors"],
+    "loras": ["style.safetensors"],
+}
+
+
+def _make_mapper():
+    # conftest.py already put a MagicMock in sys.modules["folder_paths"],
+    # so we can configure it directly here.
+    import folder_paths as fp
+    fp.folder_names_and_paths = FAKE_FOLDER_NAMES
+    fp.get_filename_list.side_effect = lambda t: FAKE_FILES.get(t, [])
+    m = ModelMapper()
+    m._build()
+    return m
+
+
+def test_get_model_type_known():
+    m = _make_mapper()
+    assert m.get_model_type("dream.safetensors") == "checkpoints"
+    assert m.get_model_type("vae.safetensors") == "vae"
+
+
+def test_loras_excluded():
+    m = _make_mapper()
+    assert m.get_model_type("style.safetensors") is None
+
+
+def test_get_all_models():
+    m = _make_mapper()
+    all_models = m.get_all_models()
+    assert "checkpoints" in all_models
+    assert "vae" in all_models
+    assert "loras" not in all_models
+    assert "dream.safetensors" in all_models["checkpoints"]
+
+
+def test_unknown_filename_returns_none():
+    m = _make_mapper()
+    assert m.get_model_type("nonexistent.ckpt") is None
+
+
+def test_extract_models_from_prompt(monkeypatch):
+    m = _make_mapper()
+
+    fake_node_cls = MagicMock()
+    fake_node_cls.INPUT_TYPES.return_value = {
+        "required": {
+            "ckpt_name": (["dream.safetensors", "v15.ckpt"],),
+            "steps": ("INT", {"default": 20}),
+        }
+    }
+
+    fake_prompt = {
+        "1": {
+            "class_type": "CheckpointLoaderSimple",
+            "inputs": {"ckpt_name": "dream.safetensors", "steps": 20},
+        }
+    }
+
+    import nodes as comfy_nodes
+    monkeypatch.setattr(comfy_nodes, "NODE_CLASS_MAPPINGS", {"CheckpointLoaderSimple": fake_node_cls})
+    results = m.extract_models_from_prompt(fake_prompt)
+
+    assert ("dream.safetensors", "checkpoints") in results
+
+
+def test_extract_models_skips_non_list_inputs(monkeypatch):
+    m = _make_mapper()
+
+    fake_node_cls = MagicMock()
+    fake_node_cls.INPUT_TYPES.return_value = {
+        "required": {
+            "text": ("STRING", {}),
+        }
+    }
+    fake_prompt = {"1": {"class_type": "CLIPTextEncode", "inputs": {"text": "hello"}}}
+
+    import nodes as comfy_nodes
+    monkeypatch.setattr(comfy_nodes, "NODE_CLASS_MAPPINGS", {"CLIPTextEncode": fake_node_cls})
+    results = m.extract_models_from_prompt(fake_prompt)
+
+    assert results == []

tests/test_model_tracker.py

@@ -0,0 +1,103 @@
+import pytest
+import tempfile
+import os
+from unittest.mock import patch
+from datetime import datetime, timezone, timedelta
+from tracker import UsageTracker
+
+
+@pytest.fixture
+def tracker(tmp_path):
+    return UsageTracker(db_path=str(tmp_path / "test.db"))
+
+
+def test_record_and_retrieve_model_usage(tracker):
+    tracker.record_model_usage([("dreamshaper.safetensors", "checkpoints")])
+    tracker.record_model_usage([("dreamshaper.safetensors", "checkpoints")])
+
+    raw = tracker.get_raw_model_stats()
+    assert len(raw) == 1
+    assert raw[0]["model_name"] == "dreamshaper.safetensors"
+    assert raw[0]["model_type"] == "checkpoints"
+    assert raw[0]["count"] == 2
+
+
+def test_record_multiple_models(tracker):
+    tracker.record_model_usage([
+        ("dreamshaper.safetensors", "checkpoints"),
+        ("vae.safetensors", "vae"),
+    ])
+    raw = tracker.get_raw_model_stats()
+    assert len(raw) == 2
+
+
+def test_reset_clears_model_usage(tracker):
+    tracker.record_model_usage([("model.safetensors", "checkpoints")])
+    tracker.reset()
+    assert tracker.get_raw_model_stats() == []
+
+
+def test_empty_models_returns_empty(tracker):
+    assert tracker.get_raw_model_stats() == []
+
+
+def test_get_model_stats_used(tracker):
+    tracker.record_model_usage([("model.safetensors", "checkpoints")])
+    installed = {"checkpoints": ["model.safetensors"]}
+    result = tracker.get_model_stats(installed)
+    assert len(result) == 1
+    assert result[0]["model_type"] == "checkpoints"
+    assert result[0]["models"][0]["status"] == "used"
+    assert result[0]["models"][0]["count"] == 1
+
+
+def test_get_model_stats_never_used_new(tracker):
+    installed = {"checkpoints": ["unused.safetensors"]}
+    result = tracker.get_model_stats(installed)
+    assert result[0]["models"][0]["status"] == "unused_new"
+    assert result[0]["models"][0]["count"] == 0
+
+
+def test_get_model_stats_uninstalled(tracker):
+    tracker.record_model_usage([("gone.safetensors", "checkpoints")])
+    installed = {}  # no longer on disk
+    result = tracker.get_model_stats(installed)
+    assert result[0]["models"][0]["status"] == "uninstalled"
+    assert result[0]["models"][0]["installed"] is False
+
+
+def test_get_model_stats_sorted_by_status(tracker):
+    tracker.record_model_usage([("active.safetensors", "checkpoints")])
+    installed = {"checkpoints": ["active.safetensors", "unused.safetensors"]}
+    result = tracker.get_model_stats(installed)
+    models = result[0]["models"]
+    statuses = [m["status"] for m in models]
+    # unused_new (2) comes before used (3) in STATUS_ORDER
+    assert statuses.index("unused_new") < statuses.index("used")
+
+
+def test_get_model_stats_safe_to_remove(tracker):
+    """A model last used 70 days ago should be classified safe_to_remove."""
+    tracker.record_model_usage([("old.safetensors", "checkpoints")])
+    installed = {"checkpoints": ["old.safetensors"]}
+
+    # Patch datetime in tracker so "now" is 70 days after last_seen
+    future_now = datetime.now(timezone.utc) + timedelta(days=70)
+    with patch("tracker.datetime") as mock_dt:
+        mock_dt.now.return_value = future_now
+        result = tracker.get_model_stats(installed)
+
+    assert result[0]["models"][0]["status"] == "safe_to_remove"
+
+
+def test_get_model_stats_consider_removing(tracker):
+    """A model last used 40 days ago should be classified consider_removing."""
+    tracker.record_model_usage([("medium.safetensors", "checkpoints")])
+    installed = {"checkpoints": ["medium.safetensors"]}
+
+    future_now = datetime.now(timezone.utc) + timedelta(days=40)
+    with patch("tracker.datetime") as mock_dt:
+        mock_dt.now.return_value = future_now
+        result = tracker.get_model_stats(installed)
+
+    assert result[0]["models"][0]["status"] == "consider_removing"

tracker.py

@@ -40,8 +40,17 @@ CREATE TABLE IF NOT EXISTS prompt_log (
     class_types TEXT NOT NULL
 );
 
+CREATE TABLE IF NOT EXISTS model_usage (
+    model_name TEXT PRIMARY KEY,
+    model_type TEXT NOT NULL,
+    count INTEGER NOT NULL DEFAULT 0,
+    first_seen TEXT NOT NULL,
+    last_seen TEXT NOT NULL
+);
+
 CREATE INDEX IF NOT EXISTS idx_node_usage_package ON node_usage(package);
 CREATE INDEX IF NOT EXISTS idx_prompt_log_timestamp ON prompt_log(timestamp);
+CREATE INDEX IF NOT EXISTS idx_model_usage_type ON model_usage(model_type);
 """
 
 
@@ -116,6 +125,47 @@ class UsageTracker:
             finally:
                 conn.close()
 
+    def record_model_usage(self, models):
+        """Record usage of model files from a single prompt.
+
+        models: list of (model_name, model_type) tuples
+        """
+        if not models:
+            return
+        now = datetime.now(timezone.utc).isoformat()
+        with self._lock:
+            self._ensure_db()
+            conn = self._connect()
+            try:
+                for model_name, model_type in models:
+                    conn.execute(
+                        """INSERT INTO model_usage (model_name, model_type, count, first_seen, last_seen)
+                           VALUES (?, ?, 1, ?, ?)
+                           ON CONFLICT(model_name) DO UPDATE SET
+                               count = count + 1,
+                               last_seen = excluded.last_seen,
+                               model_type = excluded.model_type""",
+                        (model_name, model_type, now, now),
+                    )
+                conn.commit()
+            finally:
+                conn.close()
+
+    def get_raw_model_stats(self):
+        """Return raw per-model usage rows from DB."""
+        with self._lock:
+            self._ensure_db()
+            conn = self._connect()
+            try:
+                conn.row_factory = sqlite3.Row
+                rows = conn.execute(
+                    "SELECT model_name, model_type, count, first_seen, last_seen "
+                    "FROM model_usage ORDER BY count DESC"
+                ).fetchall()
+                return [dict(r) for r in rows]
+            finally:
+                conn.close()
+
     def get_node_stats(self):
         """Return raw per-node usage data."""
         with self._lock:
@@ -216,6 +266,98 @@ class UsageTracker:
         result.sort(key=lambda p: p["total_executions"])
         return result
 
+    def get_model_stats(self, installed_by_type):
+        """Return per-type grouped model stats with tier classification.
+
+        installed_by_type: {model_type: [model_name, ...]} from ModelMapper
+        """
+        db_rows = self.get_raw_model_stats()
+        db_models = {r["model_name"]: r for r in db_rows}
+
+        now = datetime.now(timezone.utc)
+        one_month_ago = (now - timedelta(days=30)).isoformat()
+        two_months_ago = (now - timedelta(days=60)).isoformat()
+        tracking_start = self._get_first_prompt_time()
+
+        STATUS_ORDER = {
+            "safe_to_remove": 0,
+            "consider_removing": 1,
+            "unused_new": 2,
+            "used": 3,
+            "uninstalled": 4,
+        }
+
+        result_by_type = {}
+
+        # Process installed models
+        for model_type, filenames in installed_by_type.items():
+            entries = []
+            for model_name in filenames:
+                if model_name in db_models:
+                    row = db_models[model_name]
+                    last_seen = row["last_seen"]
+                    if last_seen < two_months_ago:
+                        status = "safe_to_remove"
+                    elif last_seen < one_month_ago:
+                        status = "consider_removing"
+                    else:
+                        status = "used"
+                    entry = {
+                        "model_name": model_name,
+                        "model_type": model_type,
+                        "count": row["count"],
+                        "first_seen": row["first_seen"],
+                        "last_seen": last_seen,
+                        "installed": True,
+                        "status": status,
+                    }
+                else:
+                    if tracking_start is None:
+                        status = "unused_new"
+                    elif tracking_start < two_months_ago:
+                        status = "safe_to_remove"
+                    elif tracking_start < one_month_ago:
+                        status = "consider_removing"
+                    else:
+                        status = "unused_new"
+                    entry = {
+                        "model_name": model_name,
+                        "model_type": model_type,
+                        "count": 0,
+                        "first_seen": None,
+                        "last_seen": None,
+                        "installed": True,
+                        "status": status,
+                    }
+                entries.append(entry)
+            result_by_type[model_type] = entries
+
+        # Add uninstalled (in DB but not on disk)
+        installed_names = {
+            name for names in installed_by_type.values() for name in names
+        }
+        for model_name, row in db_models.items():
+            if model_name not in installed_names:
+                model_type = row["model_type"]
+                result_by_type.setdefault(model_type, []).append({
+                    "model_name": model_name,
+                    "model_type": model_type,
+                    "count": row["count"],
+                    "first_seen": row["first_seen"],
+                    "last_seen": row["last_seen"],
+                    "installed": False,
+                    "status": "uninstalled",
+                })
+
+        # Sort each type's models by status tier then name
+        result = []
+        for model_type in sorted(result_by_type):
+            models = result_by_type[model_type]
+            models.sort(key=lambda m: (STATUS_ORDER.get(m["status"], 5), m["model_name"]))
+            result.append({"model_type": model_type, "models": models})
+
+        return result
+
     def _get_first_prompt_time(self):
         """Return the timestamp of the earliest recorded prompt, or None."""
         with self._lock:
@@ -237,6 +379,7 @@ class UsageTracker:
             try:
                 conn.execute("DELETE FROM node_usage")
                 conn.execute("DELETE FROM prompt_log")
+                conn.execute("DELETE FROM model_usage")
                 conn.commit()
             finally:
                 conn.close()