leankg 0.16.7

Lightweight Knowledge Graph for AI-Assisted Development
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
# LeanKG Massive Graph — Engineering Requirement Document

**Version:** 2.2
**Date:** 2026-04-15
**Branch:** `feature/massive-graph`
**Worktree:** `.worktrees/feature/massive-graph`
**Status:** Draft — Pending Review

**v2.2 Changes:**
- Deep dive root cause analysis: service node detection failures (Section 12)
- Dynamic NODE TYPES and EDGE TYPES sidebar (data-driven from graph)
- Service topology loaded on root when multi-repo detected
- Service node click handler for drill-down navigation
- Folder/Directory type equivalence in filters
- Expanded node type catalog (Directory, Struct, Module, Constructor, Property, Decorator)

**v2.1 Changes:**
- Added `GET /api/graph/stats` endpoint (full histogram: nodes_by_type, edges_by_type, nodes_by_depth, folders, services)
- Added adaptive loading strategy: small DB loads all, large DB loads by layers, multi-repo loads per-service
- Added `depth` parameter to `GET /api/graph/data` for layer-based loading
- Updated data flow to stats-driven decision model
- Added Phase 0 implementation tasks for stats API
- Updated test plan with stats API tests and caching verification

---

## 1. Feature Overview

**Goal:** Enable LeanKG's web UI to handle massive codebases (100K+ elements, 50K+ relationships) by implementing lazy, hierarchical graph loading with progressive drill-down from services → folders → files → functions. Both single-repo and multi-repo projects must work.

**Architecture:** Lazy-loading graph API with DB-level filtering. UI starts at service/folder level, expanding on click. CodeViewer sidebar shows source for file/function nodes. Never loads full DB at once.

**Tech Stack:** Rust (Axum) backend + Vite/React (sigma.js) frontend + CozoDB

---

## 2. Validated Codebase Audit

### 2.1 Rust Backend — API Endpoints (src/web/handlers.rs — 3587 lines)

**Every endpoint audited. Full DB load = calls `g.all_elements()` or `g.all_relationships()`.**

| Route | Handler | Line | DB Load | Notes |
|-------|---------|------|---------|-------|
| `GET /` | `root_handler` || None | Serves embedded Vite UI |
| `GET /api/elements` | `api_elements` | 2250 | **FULL** | Returns all CodeElements |
| `GET /api/relationships` | `api_relationships` | 2269 | **FULL** | Returns all Relationships |
| `GET/POST /api/annotations` | `api_annotations` / CRUD | 2289 | Partial | CRUD on BusinessLogic table |
| `GET /api/search` | `api_search` | 2405 | **FULL** | Loads all elements, then filters in Rust by q/element_type/file_path |
| `GET /api/graph/data` | `api_graph_data` | 2447 | **FULL** | Loads ALL elements + ALL relationships. Builds full graph. Deduplicates edges. Adds `service_nodes` for `service_calls` rels. |
| `GET /api/graph/services` | `api_service_graph` || **DB-level** | Queries only `service_calls` relationships. Builds ServiceGraph with nodes/edges/weights. **Good pattern to follow.** |
| `GET /api/graph/service-topology` | `api_service_topology` | 1666 | **No DB** | Pure filesystem: scans for `.git` folders (multi-repo), walks config files for `dns:///` patterns. 30s timeout. Returns `ServiceTopology { nodes, relationships }`. |
| `GET /api/graph/expand-service` | `api_graph_expand_service` | 1897 | **FULL** | Loads ALL elements + ALL relationships, then filters in Rust by folder path prefix with `depth <= 2`. Accepts `?path=` or `?service=`. |
| `GET /api/graph/expand-cluster` | `api_graph_expand_cluster` | 2105 | **FULL** | Same pattern as expand-service. Loads ALL then filters by folder prefix depth <= 2. Accepts `?path=` or `?cluster=`. |
| `GET /api/graph/subgraph` | `api_graph_subgraph` | 2639 | **FULL** | Loads ALL elements + ALL relationships. Builds adjacency map in memory. BFS from root within N hops. Params: `root`, `depth` (default 3), `types` filter. |
| `GET /api/graph/clusters` | `api_graph_clusters` | 2805 | **FULL** | Loads ALL elements + ALL relationships. Groups by parent directory. Creates `cluster:` prefix nodes. Builds inter-cluster edges. |
| `GET /api/graph/layout` | `api_graph_layout` || Partial | Server-side ForceAtlas2 layout via `LayoutEngine`. |
| `GET /api/export/graph` | `api_export_graph` | 2566 | **FULL** | Loads ALL, formats for export. |
| `POST /api/query` | `api_query` || Passthrough | Raw CozoDB Datalog query. User controls scope. |
| `POST /api/project/switch` | `api_switch_path` || None | Switches project, spawns background indexing thread. |
| `GET /api/index/status` | `api_index_status` || None | Returns indexing progress state. |
| `GET /api/file` | `api_get_file` || None | Reads file from filesystem. Security: canonicalizes path. |

**Critical finding: 7 of 10 graph endpoints load ALL data from the database.**

### 2.2 GraphEngine Methods (src/graph/query.rs — 1574 lines)

| Method | Line | DB-level? | Notes |
|--------|------|-----------|-------|
| `all_elements()` || Full scan | Returns every CodeElement row |
| `all_relationships()` || Full scan | Returns every Relationship row |
| `get_children(parent_qualified)` | **354** | **YES** | Queries with `parent_qualified = $pq`. **Already exists!** Returns direct children. |
| `get_elements_by_file(file_path)` || **YES** | Filters by file_path at DB level |
| `search_by_name(name)` || Partial | Regex search with lowercase |
| `search_by_name_typed(name, element_type, limit)` || **YES** | Typed search with limit |
| `search_by_type(element_type)` || **YES** | Filter by element_type |
| `get_service_graph(current_service)` || **YES** | Queries only `service_calls` relationships |
| `get_call_graph_bounded(source, max_depth, max_results)` || **YES** | Bounded BFS in Datalog |
| `get_dependencies(file_path)` || **YES** | With cache |
| `get_dependents(target)` || **YES** | With cache |
| `get_callers(function_name, file_scope)` || **YES** | Callers of a function |
| `get_annotation(element_qualified)` || **YES** | Single annotation lookup |

**Key discovery:** `get_children()` already exists but is marked `#[allow(dead_code)]` — it's never called by any handler. It queries by `parent_qualified` field which is populated during indexing for functions/methods nested inside classes/files.

**Graph modules** (src/graph/mod.rs): cache, clustering, context, layout, persistent_cache, query, traversal

### 2.3 Data Models (src/db/models.rs — 305 lines)

```rust
pub struct CodeElement {
    pub qualified_name: String,    // e.g., "src/main.rs::main"
    pub element_type: String,       // "file", "function", "class", "method", etc.
    pub name: String,
    pub file_path: String,
    pub line_start: u32,
    pub line_end: u32,
    pub language: String,
    pub parent_qualified: Option<String>,  // ✅ Used by get_children()
    pub cluster_id: Option<String>,
    pub cluster_label: Option<String>,
    pub metadata: serde_json::Value,
}

pub struct Relationship {
    pub source_qualified: String,
    pub target_qualified: String,
    pub rel_type: String,           // 23 types including "service_calls"
    pub confidence: f64,
    pub metadata: serde_json::Value,
}
```

**`RelationshipType` enum** has 23 variants: imports, calls, tested_by, references, documented_by, service_calls, contains, defines, declares, implements, extends, uses, embeds, annotates, depends_on, configures, triggers, produces, consumes, manages, owns, represents, routes.

### 2.4 Vite WebUI Audit

| Component | File | Status | Notes |
|-----------|------|--------|-------|
| App shell | `ui/src/App.tsx` | Working | Loads service-topology first, fallback clusters. Manages `expandedNodeId`, `overviewData`. Has `handleExpandNode` (for `cluster:`/`service:` prefix) and `handleBackToOverview`. |
| GraphViewer | `ui/src/components/GraphViewer.tsx` | Working | Sigma.js wrapper. Double-click expand for cluster/service nodes. Shows CodeViewer for non-cluster/service selected nodes. Has zoom controls and "Expand" button. |
| CodeViewer | `ui/src/components/CodeViewer.tsx` | Working | Resizable right sidebar with react-syntax-highlighter. Fetches from `/api/file?path=`. Highlights line range from node metadata (startLine/endLine). |
| useSigma hook | `ui/src/hooks/useSigma.ts` | Working | Full Sigma.js lifecycle: FA2 layout with auto-duration, noverlap post-processing, nodeReducer (search dimming, selection highlighting, service node label visibility), edgeReducer (type visibility, selection highlighting). Click/doubleclick/stage click handlers. |
| useGraphFilters | `ui/src/hooks/useGraphFilters.ts` | Working | Manages visibleLabels, visibleEdgeTypes, depthFilter, zoomLevel (clusters/modules/files/functions), effectiveLabels. Zoom level presets control which node types are visible. |
| graph-adapter | `ui/src/lib/graph-adapter.ts` | Working | `createSigmaGraph()` builds graphology Graph from API data. Uses CONTAINS/DEFINES edges for parent-child hierarchy positioning (BFS layout). Louvain community detection. Has filter/expand helpers. |
| constants | `ui/src/lib/constants.ts` | Working | NODE_COLORS, NODE_SIZES, EDGE_STYLES. FILTERABLE_LABELS = [Service, Folder, File, Class, Function, Method, Interface, Enum]. DEFAULT_VISIBLE_LABELS = ['Service']. |
| edge-bundling | `ui/src/lib/edge-bundling.ts` | Working | Edge bundling for visual clarity. |

---

## 3. Gap Analysis — What Must Change

### 3.1 Backend Gaps

| # | Gap | Current State | Impact | Priority |
|---|-----|---------------|--------|----------|
| B1 | **7 endpoints load full DB** | `api_graph_data`, `api_graph_expand_service`, `api_graph_expand_cluster`, `api_graph_subgraph`, `api_graph_clusters`, `api_search`, `api_export_graph` all call `g.all_elements()` + `g.all_relationships()` then filter in Rust | OOM on 100K+ DBs | P0 |
| B2 | **No single-repo detection** | `api_service_topology` only finds multi-repo projects (scans for `.git` subfolders). Single repos get empty topology. | Single repos show nothing | P0 |
| B3 | **No lazy children API** | `get_children()` exists at `query.rs:354` but is dead code — no handler calls it. No `/api/graph/children` endpoint exists. | No progressive drill-down | P0 |
| B4 | **No element_type filter on expand** | `api_graph_expand_service` and `api_graph_expand_cluster` filter by path depth but not by element_type | UI loads too much data | P1 |
| B5 | **No pagination** | Expand endpoints return all matches at once. No limit/offset | UI lag/crash on large dirs | P1 |
| B6 | **No directory nodes in DB** | Tree-sitter indexer creates `file`, `function`, `class`, `method` elements but NOT `directory` elements. No `contains` edges for dir→dir or dir→file. | Can't render folder hierarchy from DB | P0 |

### 3.2 Frontend Gaps

| # | Gap | Current State | Impact | Priority |
|---|-----|---------------|--------|----------|
| F1 | **Single-repo renders empty** | App.tsx loads service-topology → gets empty for single-repo → falls back to clusters (which loads full DB) | Single-repo broken | P0 |
| F2 | **No click-to-expand hierarchy** | Double-click on cluster/service dumps all depth-2 nodes at once. No folder→subfolder→file→function progression | No progressive drill-down | P0 |
| F3 | **Node type filter doesn't affect API** | Sidebar checkboxes only hide/show existing nodes in sigma. They don't filter what the expand API returns | Over-fetching | P1 |
| F4 | **No breadcrumb** | Only "Back to Overview" button. No path navigation | Poor UX for deep drill-down | P2 |
| F5 | **Global loading overlay** | Loading blocks entire UI. No per-node loading state | Bad UX during expand | P2 |
| F6 | **DEFAULT_VISIBLE_LABELS only Service** | `constants.ts` defaults to `['Service']` only. For single-repo, should also show Folder, File | Single-repo shows too little | P1 |

---

## 4. Requirements (Acceptance Criteria)

### AC-1: Single Repo vs Multi-Repo Detection
- **Given** a project directory is indexed
- **When** the web UI loads `/api/graph/service-topology`
- **Then** the API detects project type:
  - **Multi-repo**: ≥2 child directories with `.git/` → return service nodes as today
  - **Single-repo**: 0 or 1 `.git/` children → return one service node (project name) with `children_folders` populated from top-level directories
- **And** response includes `project_type: "single_repo" | "multi_repo"`

### AC-2: Default Node Rendering
- **Given** the graph loads
- **When** initial data arrives
- **Then** service nodes are always rendered
- **And** for single-repo: child folder nodes are also rendered at level 1
- **And** node type filter defaults to: Service ✓, Folder ✓, File ✓ (Function ✗, Class ✗)

### AC-3: Click-to-Expand Drill-Down
- **Given** a service or folder node is rendered
- **When** user clicks the node
- **Then** the API returns only direct children of that node (one level deeper)
- **And** children include: folders, files, functions (filtered by active node type checkboxes)
- **And** edges are returned for all returned nodes
- **And** loading is shown only for the expanding node

### AC-4: CodeViewer for File/Function Nodes
- **Given** a file or function node is clicked
- **When** the node is a file type → right sidebar shows the full file source
- **And** when function/class → sidebar highlights the function's line range
- **And** the sidebar is resizable and closeable
- **Note:** This already works. No changes needed.

### AC-5: Always Have Edges
- **Given** any level of the graph
- **When** nodes are rendered
- **Then** edges between visible nodes are always included

### AC-6: No Full DB Loading
- **Given** a database with 100K+ elements
- **When** any API call is made
- **Then** the query uses DB-level filtering (path prefix + element_type + LIMIT)
- **And** response time is < 2 seconds
- **And** memory usage stays < 200MB on the server

---

## 5. API Design

### 5.0 Adaptive Loading Strategy

The UI adapts its loading behavior based on DB size and project type:

| Condition | Strategy | Initial Load | Drill-Down |
|-----------|----------|-------------|------------|
| **Single repo, small** (< 2000 nodes) | Load all at once | `GET /api/graph/data` (full) | Expand on click via `/api/graph/children` |
| **Single repo, large** (>= 2000 nodes) | Load by depth layers | `GET /api/graph/data?depth=0,1` (Service/Folder only) | Lazy-load deeper layers via `/api/graph/children` |
| **Multi repo** | Per-service drill-down | `GET /api/graph/service-topology` | Per-service expand via `/api/graph/expand-service` |

**Decision flow:**
```
UI loads -> GET /api/graph/stats (fast histogram)
         -> if total_nodes < 2000 && single_repo:
              GET /api/graph/data (load all)
         -> elif total_nodes >= 2000 && single_repo:
              GET /api/graph/data?depth=0,1 (top layers only)
         -> elif multi_repo:
              GET /api/graph/service-topology (already fast)
```

### 5.1 `GET /api/graph/stats` (NEW)

```
Returns a full histogram of the DB. Cheap CozoDB COUNT aggregations.
Used by the UI to decide loading strategy before fetching any graph data.
Should be cached (TTL 60s) to avoid repeated counting.

Response:
{
  success: true,
  data: {
    project_type: "single_repo" | "multi_repo",
    total_nodes: 7542,
    total_edges: 12890,
    nodes_by_type: {
      "file": 200,
      "function": 5000,
      "class": 300,
      "method": 1800,
      "interface": 42,
      ...
    },
    edges_by_type: {
      "CALLS": 8000,
      "IMPORTS": 3000,
      "DEFINES": 1500,
      "CONTAINS": 390,
      ...
    },
    nodes_by_depth: {
      "0": 5,      // root/service level
      "1": 20,     // folder level
      "2": 200,    // file level
      "3": 3000,   // function/class level
      "4": 4317    // deeper
    },
    folders: [
      { "path": "src/", "nodes": 4000, "edges": 7000 },
      { "path": "ui/src/", "nodes": 800, "edges": 1200 },
      { "path": "config/", "nodes": 50, "edges": 30 },
      ...
    ],
    services: [
      { "name": "be-merchant", "nodes": 2000, "edges": 3500 },
      { "name": "be-autos", "nodes": 1500, "edges": 2800 },
      ...
    ]
  }
}

Implementation:
  - Single CozoDB aggregation query using GROUP BY on element_type, rel_type
  - Depth computed from file_path segment count (split by '/', count segments)
  - Folders: DISTINCT file_path prefixes with COUNT
  - Services: only populated for multi-repo (from service-topology logic)
  - Cache result in AppState with 60s TTL
```

### 5.2 `GET /api/graph/service-topology` (MODIFY)

```
Current: Scans filesystem for .git dirs + dns:/// patterns. No DB access.
         Returns ServiceTopology { nodes, relationships }
         Returns empty for single-repo (no .git subfolders).

New: Add single-repo detection.
     If < 2 .git children found, treat as single-repo.
     Query DB for DISTINCT top-level file_path prefixes (depth=1).
     Create folder nodes with "folder:" prefix.
     Add CONTAINS edges from service node to folder nodes.

Query params:
  - show_orphans: bool (default false)
  - depth: int (default 1) — for single-repo, how many folder levels to include

Response:
{
  success: true,
  data: {
    nodes: [...],
    relationships: [...],
    project_type: "single_repo" | "multi_repo"
  }
}
```

### 5.3 `GET /api/graph/children` (NEW)

```
Returns direct children of a node, filtered by type. DB-level query.

Query params:
  - parent: string (required) — qualified_name or path prefix of parent
  - element_types: string (optional) — comma-separated: "directory,file,function,class"
  - limit: int (default 200, max 500)
  - offset: int (default 0)

Implementation: Build on existing GraphEngine::get_children() at query.rs:354.
                Add element_type filter and LIMIT/OFFSET to the CozoDB query.

Response:
{
  success: true,
  data: {
    nodes: [...],
    relationships: [...],
    total_count: 1234,
    has_more: true
  }
}
```

### 5.4 `GET /api/graph/expand-node` (NEW — unified expand)

```
Generic expand that works for any node type. Replaces expand-service + expand-cluster.

Query params:
  - node_id: string (required) — the node's ID
  - node_type: string (required) — "service" | "directory" | "file" | "cluster"
  - element_types: string (optional) — filter children by type
  - depth: int (default 1) — how many levels to expand
  - limit: int (default 200, max 500)

Response: same as /api/graph/children
```

### 5.5 `GET /api/graph/data` (MODIFY — add depth filter)

```
Current: Loads ALL elements + ALL relationships. No filtering.

New: Accept optional depth parameter for layer-based loading.

Query params:
  - depth: string (optional) — comma-separated depth levels: "0,1" means load only
    root and first-level nodes. Depth computed from file_path segment count.

Response: same as current GraphData format.
```

### 5.6 `GET /api/file` (KEEP — already working)

No changes needed. Already serves file content with path security.

---

## 6. Database Query Strategy

### 6.1 Current Pattern (BROKEN for large DBs)

```rust
// handlers.rs:1928 — api_graph_expand_service
let (elements_result, relationships_result) = match state.get_graph_engine().await {
    Ok(g) => (g.all_elements(), g.all_relationships()),  // LOADS EVERYTHING
    ...
};
// Then filters in Rust:
let filtered_elements: Vec<_> = all_elements.iter()
    .filter(|e| e.file_path.starts_with(&prefix))
    .collect();
```

### 6.2 New Pattern (DB-level filtering)

**Approach 1: Extend existing `get_children()` (query.rs:354)**

The method already exists and does DB-level filtering by `parent_qualified`. Extend it:

```rust
pub fn get_children_filtered(
    &self,
    parent_path: &str,
    element_types: Option<&[String]>,
    limit: Option<usize>,
    offset: Option<usize>,
) -> Result<ChildrenResult, Box<dyn std::error::Error>> {
    // CozoDB query with path prefix + optional type filter + limit
    let type_filter = match element_types {
        Some(types) => format!("et in [{}]", types.iter().map(|t| format!("\"{}\"", t)).collect::<Vec<_>>().join(", ")),
        None => String::new(),
    };

    let query = format!(r#"
        ?[qn, et, name, fp, ls, le, lang, pq, cid, clbl, meta] :=
            *code_elements[qn, et, name, fp, ls, le, lang, pq, cid, clbl, meta],
            starts_with_str(fp, $prefix),
            {type_filter}
        :limit $limit
        :offset $offset
    "#);
    // ... execute and return
}
```

**Approach 2: Path-prefix query for directory listing**

```cozo
?[qn, et, name, fp] :=
    *code_elements[qn, et, name, fp, ls, le, lang, pq, cid, clbl, meta],
    starts_with_str(fp, $prefix),
    et = $type
```

**Note on CozoDB:** `starts_with_str` is available. If not, use Rust-side filtering with cursor-based iteration (read in chunks via `:limit` + `:offset`).

### 6.3 Directory Nodes — Missing Data Problem

Current indexer (`src/indexer/extractor.rs`) does NOT create `directory` elements. It creates `file`, `function`, `class`, `method`, `interface`, `enum` elements only.

**Two options:**
1. **Add directory indexing to extractor** — During indexing, emit `directory` elements for each unique folder path, with `contains` edges to child files/dirs. This is the proper solution.
2. **Synthesize directory nodes on-the-fly** — In the `/api/graph/children` handler, scan file_path values for the given prefix, extract unique subdirectories, create synthetic folder nodes. No DB schema change needed. **Recommended for V1 — faster to implement, no indexer changes.**

---

## 7. UI Component Changes

### 7.1 App.tsx Changes

| Current | New |
|---------|-----|
| Loads service-topology → fallback clusters | Load service-topology with `project_type` detection |
| `expandedNodeId` for one level | `navigationStack: string[]` for multi-level drill-down |
| Global loading overlay | Per-node loading indicator |
| Expand button for cluster/service only | Click any node to expand (service, folder, file) |
| Node type filter only hides/shows existing nodes | Node type filter also controls what children to fetch |
| DEFAULT_VISIBLE_LABELS = ['Service'] | For single-repo: ['Service', 'Folder', 'File'] |

### 7.2 GraphViewer.tsx Changes

| Current | New |
|---------|-----|
| Double-click only on service/cluster | Click on any node triggers expand or code view |
| No breadcrumb | Breadcrumb bar showing navigation path |
| CodeViewer only for non-expandable nodes | CodeViewer for file/function nodes, expand for service/folder nodes |

### 7.3 New Components

| Component | Purpose |
|-----------|---------|
| `Breadcrumb.tsx` | Navigation path: `leankg > src > graph > query.rs` |
| `NodeExpandIndicator.tsx` | "+" badge on nodes that have unloaded children |

### 7.4 State Management

```typescript
navigationStack: string[]          // ["service:leankg", "src/", "src/graph/"]
currentData: GraphData             // Current visible nodes + edges
overviewData: GraphData            // Service-level data for "back to overview"
expandedNodes: Set<string>         // Nodes that have been expanded (avoid re-fetch)
nodeChildrenCache: Map<string, GraphData>  // Cache expanded children per node
selectedNode: string | null        // For CodeViewer sidebar
nodeTypeFilter: string[]           // Active node types for API filtering
```

---

## 8. Data Flow

### 8.0 Initial Load — Stats-Driven Decision

```
UI -> GET /api/graph/stats
API -> CozoDB COUNT aggregations (fast: < 100ms even on 100K DB)
API -> Return histogram: total_nodes, total_edges, nodes_by_type, folders, services
API -> Cache result with 60s TTL
UI -> Decide loading strategy:
  - total_nodes < 2000 && single_repo → Strategy A (load all)
  - total_nodes >= 2000 && single_repo → Strategy B (depth layers)
  - multi_repo → Strategy C (per-service)
```

### 8.1 Initial Load — Single Repo, Small (Strategy A: < 2000 nodes)

```
UI -> (already has stats) → total_nodes < 2000, single_repo
UI -> GET /api/graph/data
API -> g.all_elements() + g.all_relationships() (safe: small DB)
API -> Return full GraphData
UI -> Render all nodes, set visibleLabels = [Service, Folder, File]
UI -> Subsequent clicks use /api/graph/children for targeted expand
```

### 8.2 Initial Load — Single Repo, Large (Strategy B: >= 2000 nodes)

```
UI -> (already has stats) → total_nodes >= 2000, single_repo
UI -> GET /api/graph/data?depth=0,1
API -> DB: filter elements where file_path segment count <= depth
API -> Return Service + Folder + top-level File nodes only
UI -> Render top-level layers, set visibleLabels = [Service, Folder]
UI -> User clicks folder "src/" → lazy-load via /api/graph/children
UI -> User clicks file "query.rs" → shows CodeViewer
```

### 8.3 Initial Load — Multi Repo (Strategy C)

```
UI -> (already has stats) → multi_repo detected
UI -> GET /api/graph/service-topology
API -> Scan filesystem: find .git dirs (example-user, example-order, example-product)
API -> Walk config files: find dns:/// patterns
API -> Build ServiceTopology with SERVICE_CALLS edges
API -> Response: { nodes, relationships, project_type: "multi_repo" }
UI -> Render service nodes + SERVICE_CALLS edges
UI -> User clicks service → per-service expand via /api/graph/expand-service
```

### 8.4 Click-to-Expand

```
User clicks folder node "src/graph/"
UI -> Show per-node loading spinner
UI -> GET /api/graph/children?parent=src/graph/&element_types=directory,file,function&limit=200
API -> GraphEngine::get_children_filtered("src/graph/", ["directory","file","function"], 200, 0)
API -> DB: CozoDB query with starts_with_str(fp, "src/graph/") + type filter + LIMIT
API -> Synthesize directory nodes from file_path prefixes
API -> Get relationships for returned elements
API -> Response: { nodes, relationships, total_count, has_more }
UI -> Merge children into current graph
UI -> Remove spinner
UI -> Layout new nodes
```

### 8.5 CodeViewer (already works)

```
User clicks file node "src/graph/query.rs"
UI -> Open CodeViewer sidebar
UI -> GET /api/file?path=src/graph/query.rs
API -> Read file from filesystem
API -> Response: { content, language, total_lines }
UI -> Render with syntax highlighting

User clicks function node "query.rs::GraphEngine::new"
UI -> GET /api/file?path=src/graph/query.rs
UI -> Highlight lines 8-22 (from node metadata startLine/endLine)
```

---

## 9. Implementation Plan

### Phase 0: Stats API + Adaptive Loading (P0 — new)

| Task | File | Description |
|------|------|-------------|
| 0.1 | `src/graph/query.rs` | Add `get_db_stats()` method — runs CozoDB COUNT aggregations: total nodes, total edges, nodes_by_type, edges_by_type, nodes_by_depth, top folders by node count |
| 0.2 | `src/web/handlers.rs` | Add `api_graph_stats` handler — calls `get_db_stats()`, detects project_type, returns full histogram. Cache with 60s TTL in AppState. |
| 0.3 | `src/web/mod.rs` | Register route `GET /api/graph/stats` |
| 0.4 | `src/graph/query.rs` | Add depth parameter to element queries — compute depth from file_path segment count, support `?depth=0,1` filter |
| 0.5 | `src/web/handlers.rs` | Modify `api_graph_data` to accept optional `depth` query param for layer-based loading |
| 0.6 | `ui/src/App.tsx` | Add stats fetch on load → decide strategy A/B/C → load graph data accordingly |
| 0.7 | `ui/src/hooks/useGraphFilters.ts` | Expose loading strategy state (all / layers / per-service) |
| 0.8 | `ui/src/lib/constants.ts` | Add `SMALL_DB_THRESHOLD = 2000` constant |

### Phase 1: Backend — DB-Level Filtering (P0)

| Task | File | Description |
|------|------|-------------|
| 1.1 | `src/graph/query.rs` | Add `get_children_filtered()` method extending existing `get_children()` at line 354. Add path-prefix query, element_type filter, LIMIT/OFFSET. |
| 1.2 | `src/graph/query.rs` | Add `get_top_level_directories()` method — returns distinct folder prefixes at depth N from file_path values. |
| 1.3 | `src/web/handlers.rs` | Modify `api_service_topology` (line 1666) to detect single-repo. Query DB for top-level dirs. Return `project_type` field. |
| 1.4 | `src/web/handlers.rs` | Add `api_graph_children` handler — uses `get_children_filtered()`. Synthesizes directory nodes from file_path prefixes. Returns paginated response. |
| 1.5 | `src/web/handlers.rs` | Add `api_graph_expand_node` handler — unified expand routing by node_type. |
| 1.6 | `src/web/mod.rs` | Register routes for `/api/graph/children` and `/api/graph/expand-node`. |

### Phase 2: Frontend — Stats-Driven Adaptive Loading (P0)

| Task | File | Description |
|------|------|-------------|
| 2.1 | `ui/src/App.tsx` | Replace `expandedNodeId` with `navigationStack`. Fetch `/api/graph/stats` first, then decide loading strategy. For small single-repo: load all. For large single-repo: load top layers. For multi-repo: load service topology. |
| 2.2 | `ui/src/components/GraphViewer.tsx` | Update click handling: service/folder → expand, file/function → CodeViewer. Per-node loading indicator. |
| 2.3 | `ui/src/components/Breadcrumb.tsx` | New file. Navigation path with click-to-navigate. |
| 2.4 | `ui/src/lib/graph-adapter.ts` | Update merge logic for incremental node addition from children API. |
| 2.5 | `ui/src/lib/constants.ts` | Change DEFAULT_VISIBLE_LABELS based on project type + DB size. For small single-repo: all types. For large single-repo: [Service, Folder]. For multi-repo: [Service]. |

### Phase 3: Polish (P1-P2)

| Task | Description |
|------|-------------|
| 3.1 | Node type filter integration — pass checkbox state to expand API as `element_types` param |
| 3.2 | "Load more" button for paginated results when `has_more` is true |
| 3.3 | Per-node loading spinner instead of global overlay |

---

## 10. Testing Plan

### 10.1 Test Projects

| Test Case | Path | Type | Why |
|-----------|------|------|-----|
| **Multi-repo** | `examples/` | multi_repo | Contains `example-user/`, `example-order/`, `example-product/` — 3 Go microservices with `dns:///` patterns in `config/config.go`. Service topology: `example-product → example-user`, `example-product → example-order`, `example-order → example-user`. |
| **Single-repo** | `.worktrees/feature/massive-graph/` | single_repo | The LeanKG codebase itself (Rust + TypeScript). 339 elements, 262 relationships. Has `src/`, `ui/`, `config/`, `docs/` etc. as top-level directories. One `.git` (worktree). |

### 10.2 Multi-repo Setup (examples/)

The `examples/` directory has 3 Go microservices with inter-service gRPC calls:

```
examples/
├── example-user/       (port 10001 grpc)
│   └── config/config.go     — no dns:/// (leaf service)
├── example-order/      (port 10002 grpc)
│   └── config/config.go     — dns:///example-user:10001
├── example-product/    (port 10003 grpc)
│   └── config/config.go     — dns:///example-user:10001 + dns:///example-order:10002
├── go-api-service/     (standalone, not part of multi-repo topology)
├── java-api-service/
├── kotlin-api-service/
└── ...
```

**Service topology expected:**
```
example-product ──SERVICE_CALLS──> example-user
example-product ──SERVICE_CALLS──> example-order
example-order   ──SERVICE_CALLS──> example-user
```

**Note:** `examples/` directories do NOT have `.git/` subfolders. For multi-repo detection testing, must `git init` in each service:
```bash
cd examples && \
  git init example-user && \
  git init example-order && \
  git init example-product
```

### 10.3 Single-repo Setup (.worktrees/feature/massive-graph/)

The worktree itself is the test case. It has:
```
.worktrees/feature/massive-graph/
├── .git           (worktree file, not directory)
├── src/           (Rust source — 1574 lines in query.rs alone)
├── ui/src/        (TypeScript/React frontend)
├── config/        (YAML configs)
├── docs/          (Documentation)
└── examples/      (Example projects)
```

Expected initial load: project_type = "single_repo", nodes = [service:leankg, folder:src, folder:ui, folder:config, folder:docs, folder:examples]

### 10.4 Backend Testing

```bash
cd .worktrees/feature/massive-graph

# Build
cargo build

# Test 0: Stats API
curl -s 'http://localhost:8080/api/graph/stats' | jq '.data'
# Expected: { project_type: "single_repo", total_nodes: ~339, total_edges: ~262,
#             nodes_by_type: { file: N, function: N, ... }, edges_by_type: {...},
#             nodes_by_depth: {"1": N, "2": N, ...}, folders: [...] }

# Test 1: Single-repo detection
curl -s 'http://localhost:8080/api/graph/service-topology' | jq '.data.project_type'
# Expected: "single_repo"

# Test 2: Children API — expand src/
curl -s 'http://localhost:8080/api/graph/children?parent=src/&element_types=directory,file&limit=50' | jq '.data.nodes | length'
# Expected: < 50 (paginated)

# Test 3: Children API — has_more
curl -s 'http://localhost:8080/api/graph/children?parent=src/&limit=5' | jq '.data.has_more'
# Expected: true (src/ has more than 5 children)

# Test 4: Graph data with depth filter
curl -s 'http://localhost:8080/api/graph/data?depth=0,1' | jq '.data.nodes | length'
# Expected: small number (only root + first level)

# Test 5: File content
curl -s 'http://localhost:8080/api/file?path=src/graph/query.rs' | jq '.data.language'
# Expected: "rust"

# Test 6: Stats API caching (call twice, second should be faster)
time curl -s 'http://localhost:8080/api/graph/stats' > /dev/null
time curl -s 'http://localhost:8080/api/graph/stats' > /dev/null
# Expected: second call significantly faster (cache hit)
```

### 10.5 Multi-repo Testing

```bash
# Setup: init git in example services
cd examples
git init example-user && git init example-order && git init example-product

# Switch project to examples/
curl -X POST 'http://localhost:8080/api/project/switch' \
  -H 'Content-Type: application/json' \
  -d '{"path": "/Users/linh.doan/work/harvey/freepeak/leankg/examples"}'

# Test: Multi-repo detection
curl -s 'http://localhost:8080/api/graph/service-topology' | jq '.data.project_type'
# Expected: "multi_repo"

# Test: Service topology has correct edges
curl -s 'http://localhost:8080/api/graph/service-topology' | jq '.data.relationships[].rel_type'
# Expected: "SERVICE_CALLS" (3 edges)

# Cleanup
rm -rf examples/example-user/.git examples/example-order/.git examples/example-product/.git
```

### 10.6 Frontend Testing

```bash
cd ui/ && npm run dev
# Open http://localhost:5173

# Verify stats-driven loading:
#   0. Network tab: first call = GET /api/graph/stats (< 100ms)
#   1. For small DB: second call = GET /api/graph/data (full load)
#   2. For large DB: second call = GET /api/graph/data?depth=0,1 (layers)
#   3. For multi-repo: second call = GET /api/graph/service-topology

# Verify single-repo:
#   1. Service node + folder nodes (src/, ui/, config/, docs/) render on load
#   2. Click "src/" → children load (subfolders + files)
#   3. Click "src/graph/" → more children load
#   4. Click "query.rs" → CodeViewer opens with Rust syntax highlighting
#   5. Click "GraphEngine::new" → CodeViewer highlights lines 8-22
#   6. Breadcrumb: leankg > src > graph > query.rs
#   7. No full-page loading after initial load
#   8. Node type checkboxes filter children on next expand

# Verify multi-repo:
#   1. Switch project to examples/
#   2. 3 service nodes render: example-user, example-order, example-product
#   3. SERVICE_CALLS edges between them
#   4. Click service → shows internal structure
```

---

## 11. Risks and Mitigations

| Risk | Likelihood | Impact | Mitigation |
|------|-----------|--------|------------|
| Stats COUNT query slow on 100K+ DB | Low | Medium | Cache with 60s TTL. Use CozoDB aggregation (not Rust-side counting). If still slow, maintain running counts on insert. |
| CozoDB `starts_with_str` not efficient for prefix scans | Medium | High | Fall back to Rust-side filtering with cursor pagination (`:limit` + `:offset`) |
| Large directory (>1000 files) | High | Medium | LIMIT + pagination + "Load more" UI |
| Sigma.js performance with > 500 nodes | Medium | Medium | Virtual rendering, hide labels on zoom out, limit visible nodes |
| `parent_qualified` field not populated for all elements | High | High | Fallback to file_path prefix extraction for directory listing |
| Path normalization (relative vs absolute, `./` prefix) | High | Low | Normalize all paths at API boundary — strip `./` prefix consistently |
| No `.git` in examples/ for multi-repo testing | Certain | Low | `git init` in each service for testing; document in test plan |
| Stats cache stale after indexing | Medium | Low | Invalidate cache on `api_switch_path` and after indexing completes |
| Depth threshold (2000) may need tuning per project | Medium | Low | Make threshold configurable via leankg.yaml `web.stats_threshold` |

---

## 12. Root Cause Analysis: Service Node Detection & Display Issues

**Date:** 2026-04-15
**Investigator:** AI Agent Deep Dive

### 12.1 Problem Statement

Service nodes and SERVICE_CALLS edges are not displayed correctly in the WebUI. The right sidebar NODE TYPES list doesn't match the EDGE TYPES list. Clicking service nodes does nothing.

### 12.2 Root Causes Identified

#### RC-1: Services are never stored as CodeElement records (CRITICAL)

**File:** `src/indexer/microservice.rs:313-328`

The microservice extractor creates `Relationship` rows with `rel_type = "service_calls"`, where `source_qualified` and `target_qualified` hold service names like `"be-gateway"`. However, no corresponding `CodeElement` with `element_type: "service"` is ever created. This means:

- `all_elements()` never returns service nodes
- `get_children_filtered()` can't find services
- Search APIs can't locate services
- Only the specialized `get_service_graph()` and `extract_service_topology()` can surface them

**Impact:** Services are invisible to the standard graph query pipeline.

#### RC-2: Two disconnected service detection paths (CRITICAL)

**Files:** `src/web/handlers.rs:1609` vs `src/web/handlers.rs:1668`

Two completely separate APIs detect services differently:

| API | Method | Node ID Format | Data Source |
|-----|--------|---------------|-------------|
| `/api/graph/services` | `get_service_graph()` | `"be-gateway"` (plain) | CozoDB relationships |
| `/api/graph/service-topology` | `extract_service_topology()` | `"service:be-gateway"` (prefixed) | Filesystem scan |

The frontend `useSigma.ts:223` checks both patterns (`node.startsWith('service:') || data.nodeType === 'Service'`), but the main graph data endpoint (`/api/graph/children`) returns neither format.

#### RC-3: UI loads `/api/graph/children` at root — returns no services (HIGH)

**File:** `ui/src/App.tsx:83-85`

The initial load calls `loadChildren('')` which hits `/api/graph/children?parent=`. This endpoint queries `code_elements` via `get_children_filtered()` — since services don't exist as `CodeElement` records, only `Folder` and `File` nodes appear. No `Service` nodes are ever returned at root level.

**This is why service nodes are invisible.** The service topology data exists at `/api/graph/service-topology` but the UI never calls it on initial load.

#### RC-4: NODE TYPES sidebar is hardcoded (HIGH)

**File:** `ui/src/lib/constants.ts:37-46` (old `FILTERABLE_LABELS`)

The sidebar displayed a static list: `['Service', 'Folder', 'File', 'Class', 'Function', 'Method', 'Interface', 'Enum']`. This:
- Doesn't include `Struct`, `Directory`, `Module`, `Constructor`, `Property`
- Shows types that may not exist in the current graph
- Doesn't reflect the actual data returned by the API

#### RC-5: EDGE TYPES sidebar is hardcoded (MEDIUM)

**File:** `ui/src/lib/constants.ts:25-33`

The edge type sidebar shows `CONTAINS, DEFINES, IMPORTS, CALLS, SERVICE_CALLS, EXTENDS, IMPLEMENTS` regardless of what edges exist in the current graph view. Missing: `REFERENCES, DOCUMENTED_BY, TESTED_BY`.

#### RC-6: No click handler for Service nodes (MEDIUM)

**File:** `ui/src/App.tsx:87-98` (old code)

When clicking a service node, the `handleNodeClick` fell through without any case for `elementType === 'service'`. Clicking did nothing — no drill-down into the service's internals.

### 12.3 Fixes Applied

| Fix | Files Changed | Description |
|-----|--------------|-------------|
| **F1: Dynamic Node Types** | `constants.ts`, `App.tsx`, `useGraphFilters.ts` | Node types are now discovered from actual graph data via `useMemo`. Replaced static `FILTERABLE_LABELS` with `DEFAULT_NODE_TYPE_ORDER` + data-driven discovery. Added `Directory`, `Struct`, `Module`, `Constructor`, `Property`, `Decorator`. |
| **F2: Dynamic Edge Types** | `App.tsx` | Edge types are now discovered from actual relationships data. Added `REFERENCES`, `DOCUMENTED_BY`, `TESTED_BY` to EDGE_STYLES. |
| **F3: Service topology on root load** | `App.tsx` | `loadChildren('')` now first tries `/api/graph/service-topology`. If the project has multiple services (>1 node), it uses that data. Falls back to `/api/graph/children` for single-repo projects. |
| **F4: Service click handler** | `App.tsx` | Added `if (elementType === 'service' || nodeId.startsWith('service:'))` handler that navigates into the service's folder path. |
| **F5: Folder/Directory equivalence** | `graph-adapter.ts` | `filterGraphByLabels` now treats `Folder` and `Directory` as equivalent — toggling one toggles both. |
| **F6: Expanded color/size maps** | `constants.ts` | Added `Directory`, `Module`, `Constructor`, `Property`, `Decorator`, `Config` to both `NODE_COLORS` and `NODE_SIZES`. |

### 12.4 Architecture Diagram (After Fix)

```
Initial Load (root)
    ├── Try /api/graph/service-topology
    │   ├── Multi-repo detected (>1 .git dirs)
    │   │   └── Return Service + Folder nodes with SERVICE_CALLS + CONTAINS edges
    │   └── Single-repo (fallback)
    │       └── Return Service(project) + Folder children with CONTAINS edges
    └── If no services → /api/graph/children?parent=
        └── Return Folder + File + Function nodes from DB

User clicks Service node
    └── Navigate into service folder path → /api/graph/children?parent=<service-path>

User clicks Folder/Directory node
    └── Navigate deeper → /api/graph/children?parent=<path>

Sidebar: Node Types / Edge Types
    └── Discovered dynamically from current graph data
```

### 12.5 Remaining Issues (Future Work)

| Issue | Severity | Description |
|-------|----------|-------------|
| Services not stored as CodeElement | Critical | The indexer should create `CodeElement { element_type: "service", qualified_name: "service:be-gateway", ... }` when extracting microservice relationships. This would unify all queries. |
| `get_children_filtered` ignores services | High | Should also check `service_calls` relationships to find child services at a given path level |
| Clustering ignores service_calls | Medium | `clustering.rs` should include `service_calls` in community detection |
| Node ID format inconsistency | Medium | Standardize on `"service:<name>"` prefix format across all APIs |

---

*Last updated: 2026-04-15 — ERD v2.2 with Root Cause Analysis and fixes for service node detection & display.*