diff --git a/.changelog/next/fixed-issue-4196.md b/.changelog/next/fixed-issue-4196.md
new file mode 100644
index 0000000000..f79c2b3e4c
--- /dev/null
+++ b/.changelog/next/fixed-issue-4196.md
@@ -0,0 +1 @@
+- Brain graph views now load lightweight projections, keeping large song sheets out of graph data while preserving node summaries and tags.
diff --git a/server/services/brainGraph.js b/server/services/brainGraph.js
index e2465db6bc..f6d9824edb 100644
--- a/server/services/brainGraph.js
+++ b/server/services/brainGraph.js
@@ -15,9 +15,9 @@
  * O(n²) tag explosion can't happen.
  *
  * The two edge-bearing views need every record's tags and summary, so they go
- * through `loadNodes()` (a full read of each entity store). The search index
- * needs only `{ id, label, brainType }`, so it reads the cached field
- * projections in `brainSearchIndex` instead — focusing the graph's search box
+ * through `loadNodes()` (the cached field projections for each entity store).
+ * The search index needs only `{ id, label, brainType }`, so it reads those
+ * same projections in `brainSearchIndex` instead — focusing the graph's search box
  * no longer walks and JSON-parses every record body and Daily Log entry on disk
  * (issue #3507).
  */
@@ -87,9 +87,9 @@ const journalDate = (entry) => entry.id || entry.date;
 // Load all non-archived brain entities as graph nodes (no edges).
 async function loadNodes() {
   const nodes = [];
-  for (const type of ENTITY_TYPES) {
-    const records = await brainStorage.getAll(type);
-    for (const record of records) {
+  const perType = await Promise.all(ENTITY_TYPES.map((type) => getBrainProjections(type, { ranked: false })));
+  ENTITY_TYPES.forEach((type, i) => {
+    for (const record of perType[i]) {
       if (record.archived) continue;
       nodes.push({
         id: record.id,
@@ -101,7 +101,7 @@ async function loadNodes() {
         status: record.status
       });
     }
-  }
+  });
 
   // Goals (identity system): active goals only (completed/abandoned excluded)
   const goalsData = await getGoals().catch(() => null);
diff --git a/server/services/brainGraph.test.js b/server/services/brainGraph.test.js
index f081c24b47..24b3b898e9 100644
--- a/server/services/brainGraph.test.js
+++ b/server/services/brainGraph.test.js
@@ -149,6 +149,11 @@ describe('getBrainGraphSearchIndex', () => {
         content: 'z'.repeat(5000),
         segments: [{ text: 'z'.repeat(5000) }]
       }];
+      if (type === 'songs') return [{
+        id: 's1',
+        title: 'Example song',
+        content: { format: 'chordpro', text: 'x'.repeat(5000) }
+      }];
       return [];
     });
 
@@ -158,6 +163,9 @@ describe('getBrainGraphSearchIndex', () => {
     expect(memory.embedding).toBeUndefined();
     expect(memory.attachments).toBeUndefined();
 
+    const [song] = await getBrainProjections('songs', { ranked: false });
+    expect(song.content).toBeUndefined();
+
     // The Daily Log body is the biggest record in the brain — the graph only
     // needs "is this day non-empty?", so the text must never reach the cache.
     expect(await getBrainProjections('journals', { ranked: false })).toEqual([
@@ -186,6 +194,25 @@ describe('getBrainGraphOverview', () => {
     expect(result.nodes.map(n => n.label).sort()).toEqual(['Alice', 'Phoenix']);
   });
 
+  it('keeps projected summaries, tags, and status for edge-bearing views', async () => {
+    onlyType('people', [{
+      id: 'p1',
+      name: 'Ada Placeholder',
+      description: 'Example description',
+      tags: ['example'],
+      status: 'active'
+    }]);
+
+    const result = await getBrainGraphOverview({ limit: 100 });
+
+    expect(result.nodes).toMatchObject([{
+      id: 'p1',
+      summary: 'Example description',
+      tags: ['example'],
+      status: 'active'
+    }]);
+  });
+
   it('summarizes a SongBook node with its artist, never its content object (#4105)', async () => {
     onlyType('songs', [{
       id: 's1',
diff --git a/server/services/brainSearchIndex.js b/server/services/brainSearchIndex.js
index dfb27e6e05..e4ef0c7c5c 100644
--- a/server/services/brainSearchIndex.js
+++ b/server/services/brainSearchIndex.js
@@ -47,37 +47,50 @@ import { safeDate } from '../lib/fileUtils.js';
  *     from `capturedText` / `name` / `context` / `title` / `oneLiner` / `notes`
  *     / `nextAction` / `content` / `mood` / `url` / `description`.
  *   - `getBrainGraphSearchIndex` (server/services/brainGraph.js) derives each
- *     node's label from `name || title` and drops archived records, so the
- *     graph entity types also project `name`, `title` and `archived`.
+ *     node's label from `name || title` and drops archived records, while the
+ *     edge-bearing graph views also need tags, status, and summary fields.
  * `journals` is graph-only (the Daily Log is not a unified-search source), and
  * projects no body — see `journalHasBody` below. `songs` (SongBook) is
- * graph-only too, and projects only its label field: the sheet body lives in
- * `content.text` (up to 200k chars of tab/ChordPro per song) and nothing here
- * renders it, so it must never reach the cache.
+ * graph-only too, and its sheet body lives in `content.text` (up to 200k
+ * chars of tab/ChordPro per song). The graph projection keeps only a string
+ * `content` value, so the SongBook object never reaches the cache.
  */
+const GRAPH_PROJECTION_FIELDS = Object.freeze([
+  'name', 'title', 'archived', 'tags', 'status',
+  'description', 'context', 'oneLiner', 'artist', 'notes'
+]);
+
 const PROJECTED_FIELDS = Object.freeze({
   inbox: Object.freeze(['capturedText']),
-  people: Object.freeze(['name', 'title', 'context', 'archived']),
-  projects: Object.freeze(['name', 'title', 'notes', 'archived']),
-  ideas: Object.freeze(['name', 'title', 'oneLiner', 'notes', 'archived']),
-  admin: Object.freeze(['name', 'title', 'notes', 'nextAction', 'archived']),
-  memories: Object.freeze(['name', 'title', 'content', 'mood', 'archived']),
+  people: GRAPH_PROJECTION_FIELDS,
+  projects: GRAPH_PROJECTION_FIELDS,
+  ideas: GRAPH_PROJECTION_FIELDS,
+  admin: Object.freeze([...GRAPH_PROJECTION_FIELDS, 'nextAction']),
+  memories: Object.freeze([...GRAPH_PROJECTION_FIELDS, 'mood']),
   links: Object.freeze(['title', 'url', 'description']),
   journals: Object.freeze(['date']),
-  songs: Object.freeze(['title', 'archived']),
+  songs: GRAPH_PROJECTION_FIELDS,
 });
 
 /**
  * Is this Daily Log entry non-empty? The graph only asks the question — it
  * never renders the answer's source — and a journal body is the largest record
  * in the brain, so the predicate is projected and `content`/`segments` are not.
- * Exported so `brainGraph`'s full-record path (which still loads bodies, for
- * summaries and tags) applies the identical rule.
+ * Exported so `brainGraph`'s journal path applies the identical rule.
  */
 export const journalHasBody = (record) => !!(record?.content || record?.segments?.length);
 
+const stringContent = (record) => typeof record?.content === 'string' ? record.content : undefined;
+const GRAPH_DERIVED_FIELDS = Object.freeze({ content: stringContent });
+
 // Fields computed from the record rather than copied off it, per type.
 const DERIVED_FIELDS = Object.freeze({
+  people: GRAPH_DERIVED_FIELDS,
+  projects: GRAPH_DERIVED_FIELDS,
+  ideas: GRAPH_DERIVED_FIELDS,
+  admin: GRAPH_DERIVED_FIELDS,
+  memories: GRAPH_DERIVED_FIELDS,
+  songs: GRAPH_DERIVED_FIELDS,
   journals: Object.freeze({ hasBody: journalHasBody }),
 });
 
diff --git a/server/services/brainSearchIndex.test.js b/server/services/brainSearchIndex.test.js
index 79d3437a33..c2aef31d7f 100644
--- a/server/services/brainSearchIndex.test.js
+++ b/server/services/brainSearchIndex.test.js
@@ -53,15 +53,47 @@ describe('brainSearchIndex', () => {
 
   it('projects only the fields its consumers read', async () => {
     getAll.mockResolvedValue([
-      { id: 'p1', name: 'Ada Placeholder', context: 'colleague', avatarBlob: 'x'.repeat(50), embedding: [1, 2, 3] }
+      {
+        id: 'p1',
+        name: 'Ada Placeholder',
+        context: 'colleague',
+        tags: ['example'],
+        status: 'active',
+        avatarBlob: 'x'.repeat(50),
+        embedding: [1, 2, 3]
+      }
     ])
 
     const [projection] = await getBrainProjections('people')
-    expect(projection).toEqual({ id: 'p1', name: 'Ada Placeholder', context: 'colleague' })
+    expect(projection).toMatchObject({
+      id: 'p1',
+      name: 'Ada Placeholder',
+      context: 'colleague',
+      tags: ['example'],
+      status: 'active'
+    })
     expect(projection).not.toHaveProperty('avatarBlob')
     expect(projection).not.toHaveProperty('embedding')
   })
 
+  it('keeps string content but drops object-shaped content from graph projections', async () => {
+    getAll.mockImplementation(async (type) => {
+      if (type === 'memories') return [{ id: 'm1', title: 'Example memory', content: 'searchable text' }]
+      if (type === 'songs') return [{
+        id: 's1',
+        title: 'Example song',
+        content: { format: 'chordpro', text: 'large sheet body' }
+      }]
+      return []
+    })
+
+    const [memory] = await getBrainProjections('memories', { ranked: false })
+    const [song] = await getBrainProjections('songs', { ranked: false })
+
+    expect(memory.content).toBe('searchable text')
+    expect(song.content).toBeUndefined()
+  })
+
   it('projects a journal entry as a body predicate, never the body', async () => {
     getAll.mockResolvedValue([
       { id: '2026-01-01', content: 'x'.repeat(500), segments: [{ text: 'x' }] },