diff --git a/packages/palace/package.json b/packages/palace/package.json
new file mode 100644
index 00000000..44110e34
--- /dev/null
+++ b/packages/palace/package.json
@@ -0,0 +1,31 @@
+{
+  "name": "@bytelyst/palace",
+  "version": "0.1.0",
+  "description": "Shared MemPalace primitives — types, cosine similarity, dedup, relevance decay, extraction prompts, KG helpers, wake-up context builder",
+  "type": "module",
+  "exports": {
+    ".": {
+      "import": "./dist/index.js",
+      "types": "./dist/index.d.ts"
+    }
+  },
+  "main": "./dist/index.js",
+  "types": "./dist/index.d.ts",
+  "files": [
+    "dist"
+  ],
+  "scripts": {
+    "build": "tsc",
+    "test": "vitest run --pool forks"
+  },
+  "devDependencies": {
+    "@types/node": "^22.12.0",
+    "vitest": "^3.0.5"
+  },
+  "peerDependencies": {
+    "zod": "^3.0.0"
+  },
+  "publishConfig": {
+    "registry": "https://gitea.bytelyst.com/api/packages/ByteLyst/npm/"
+  }
+}
diff --git a/packages/palace/src/__tests__/cosine.test.ts b/packages/palace/src/__tests__/cosine.test.ts
new file mode 100644
index 00000000..4d8a4495
--- /dev/null
+++ b/packages/palace/src/__tests__/cosine.test.ts
@@ -0,0 +1,86 @@
+import { describe, it, expect } from 'vitest';
+import { cosineSimilarity, normalizeVector, topKByCosine } from '../cosine.js';
+
+describe('cosineSimilarity', () => {
+  it('returns 1.0 for identical vectors', () => {
+    expect(cosineSimilarity([1, 0, 0], [1, 0, 0])).toBeCloseTo(1.0);
+  });
+
+  it('returns 0.0 for orthogonal vectors', () => {
+    expect(cosineSimilarity([1, 0], [0, 1])).toBeCloseTo(0.0);
+  });
+
+  it('returns -1.0 for opposite vectors', () => {
+    expect(cosineSimilarity([1, 0], [-1, 0])).toBeCloseTo(-1.0);
+  });
+
+  it('returns 0 for empty vectors', () => {
+    expect(cosineSimilarity([], [])).toBe(0);
+  });
+
+  it('returns 0 for mismatched dimensions', () => {
+    expect(cosineSimilarity([1, 2], [1, 2, 3])).toBe(0);
+  });
+
+  it('returns 0 for zero vector', () => {
+    expect(cosineSimilarity([0, 0, 0], [1, 2, 3])).toBe(0);
+  });
+
+  it('computes correct similarity for arbitrary vectors', () => {
+    const a = [1, 2, 3];
+    const b = [4, 5, 6];
+    // Known value: (4+10+18) / (sqrt(14)*sqrt(77)) ≈ 0.9746
+    expect(cosineSimilarity(a, b)).toBeCloseTo(0.9746, 3);
+  });
+});
+
+describe('normalizeVector', () => {
+  it('normalizes to unit length', () => {
+    const v = normalizeVector([3, 4]);
+    const magnitude = Math.sqrt(v[0] ** 2 + v[1] ** 2);
+    expect(magnitude).toBeCloseTo(1.0);
+  });
+
+  it('returns zero vector for zero input', () => {
+    expect(normalizeVector([0, 0, 0])).toEqual([0, 0, 0]);
+  });
+
+  it('preserves direction', () => {
+    const v = normalizeVector([2, 0, 0]);
+    expect(v[0]).toBeCloseTo(1);
+    expect(v[1]).toBeCloseTo(0);
+    expect(v[2]).toBeCloseTo(0);
+  });
+});
+
+describe('topKByCosine', () => {
+  const items = [
+    { id: 'a', emb: [1, 0, 0] },
+    { id: 'b', emb: [0, 1, 0] },
+    { id: 'c', emb: [0.9, 0.1, 0] },
+    { id: 'd', emb: undefined as number[] | undefined },
+  ];
+
+  it('returns top-K sorted by score', () => {
+    const results = topKByCosine([1, 0, 0], items, i => i.emb, 2);
+    expect(results).toHaveLength(2);
+    expect(results[0].item.id).toBe('a');
+    expect(results[0].score).toBeCloseTo(1.0);
+    expect(results[1].item.id).toBe('c');
+  });
+
+  it('skips items with missing embeddings', () => {
+    const results = topKByCosine([1, 0, 0], items, i => i.emb, 10);
+    expect(results).toHaveLength(3); // 'd' skipped
+  });
+
+  it('respects minScore filter', () => {
+    const results = topKByCosine([1, 0, 0], items, i => i.emb, 10, 0.5);
+    expect(results.every(r => r.score >= 0.5)).toBe(true);
+  });
+
+  it('returns empty array for empty items', () => {
+    const results = topKByCosine([1, 0, 0], [], () => undefined, 5);
+    expect(results).toEqual([]);
+  });
+});
diff --git a/packages/palace/src/__tests__/decay.test.ts b/packages/palace/src/__tests__/decay.test.ts
new file mode 100644
index 00000000..dff95bdf
--- /dev/null
+++ b/packages/palace/src/__tests__/decay.test.ts
@@ -0,0 +1,57 @@
+import { describe, it, expect } from 'vitest';
+import { computeDecayedRelevance, boostRelevance } from '../decay.js';
+
+describe('computeDecayedRelevance', () => {
+  it('returns original relevance for just-created memory', () => {
+    const now = new Date();
+    expect(computeDecayedRelevance(0.8, now.toISOString(), 30, now)).toBeCloseTo(0.8);
+  });
+
+  it('halves relevance after exactly one half-life', () => {
+    const now = new Date();
+    const thirtyDaysAgo = new Date(now.getTime() - 30 * 86_400_000);
+    expect(computeDecayedRelevance(1.0, thirtyDaysAgo.toISOString(), 30, now)).toBeCloseTo(0.5, 2);
+  });
+
+  it('quarters relevance after two half-lives', () => {
+    const now = new Date();
+    const sixtyDaysAgo = new Date(now.getTime() - 60 * 86_400_000);
+    expect(computeDecayedRelevance(1.0, sixtyDaysAgo.toISOString(), 30, now)).toBeCloseTo(0.25, 2);
+  });
+
+  it('returns 0 for zero relevance', () => {
+    expect(computeDecayedRelevance(0, '2024-01-01')).toBe(0);
+  });
+
+  it('clamps to [0, 1]', () => {
+    const now = new Date();
+    expect(computeDecayedRelevance(1.5, now.toISOString(), 30, now)).toBeLessThanOrEqual(1);
+  });
+
+  it('returns original relevance for zero half-life', () => {
+    expect(computeDecayedRelevance(0.8, '2024-01-01', 0)).toBe(0.8);
+  });
+
+  it('accepts Date objects', () => {
+    const now = new Date();
+    expect(computeDecayedRelevance(0.8, now, 30, now)).toBeCloseTo(0.8);
+  });
+});
+
+describe('boostRelevance', () => {
+  it('boosts relevance toward 1.0', () => {
+    expect(boostRelevance(0.5, 0.3)).toBeCloseTo(0.65);
+  });
+
+  it('does not exceed 1.0', () => {
+    expect(boostRelevance(0.99, 0.5)).toBeLessThanOrEqual(1.0);
+  });
+
+  it('returns 0 for zero relevance with zero boost', () => {
+    expect(boostRelevance(0, 0)).toBe(0);
+  });
+
+  it('applies default boost factor of 0.3', () => {
+    expect(boostRelevance(0.5)).toBeCloseTo(0.65);
+  });
+});
diff --git a/packages/palace/src/__tests__/dedup.test.ts b/packages/palace/src/__tests__/dedup.test.ts
new file mode 100644
index 00000000..680cc322
--- /dev/null
+++ b/packages/palace/src/__tests__/dedup.test.ts
@@ -0,0 +1,70 @@
+import { describe, it, expect } from 'vitest';
+import { isContentDuplicate, isExactDuplicate, findClosestMatch } from '../dedup.js';
+
+describe('isContentDuplicate', () => {
+  const baseEmbedding = [1, 0, 0];
+  const similarEmbedding = [0.99, 0.1, 0]; // very similar to base
+  const differentEmbedding = [0, 1, 0]; // orthogonal
+
+  it('detects near-duplicate above threshold', () => {
+    expect(isContentDuplicate(baseEmbedding, [similarEmbedding], 0.9)).toBe(true);
+  });
+
+  it('rejects non-duplicate below threshold', () => {
+    expect(isContentDuplicate(baseEmbedding, [differentEmbedding], 0.9)).toBe(false);
+  });
+
+  it('returns false for empty existing embeddings', () => {
+    expect(isContentDuplicate(baseEmbedding, [], 0.9)).toBe(false);
+  });
+
+  it('skips embeddings with mismatched dimensions', () => {
+    expect(isContentDuplicate([1, 0], [[1, 0, 0]], 0.9)).toBe(false);
+  });
+
+  it('uses default threshold of 0.90', () => {
+    expect(isContentDuplicate(baseEmbedding, [similarEmbedding])).toBe(true);
+  });
+});
+
+describe('isExactDuplicate', () => {
+  it('detects exact match', () => {
+    expect(isExactDuplicate('Hello World', 'Hello World')).toBe(true);
+  });
+
+  it('is case-insensitive', () => {
+    expect(isExactDuplicate('Hello', 'hello')).toBe(true);
+  });
+
+  it('trims whitespace', () => {
+    expect(isExactDuplicate('  hello  ', 'hello')).toBe(true);
+  });
+
+  it('rejects different content', () => {
+    expect(isExactDuplicate('Hello', 'World')).toBe(false);
+  });
+});
+
+describe('findClosestMatch', () => {
+  it('finds the closest embedding', () => {
+    const result = findClosestMatch(
+      [1, 0, 0],
+      [
+        [0, 1, 0],
+        [0.9, 0.1, 0],
+        [0, 0, 1],
+      ]
+    );
+    expect(result).not.toBeNull();
+    expect(result!.index).toBe(1);
+    expect(result!.score).toBeGreaterThan(0.9);
+  });
+
+  it('returns null for empty embeddings', () => {
+    expect(findClosestMatch([1, 0], [])).toBeNull();
+  });
+
+  it('returns null when no embedding exceeds minScore', () => {
+    expect(findClosestMatch([1, 0, 0], [[0, 1, 0]], 0.99)).toBeNull();
+  });
+});
diff --git a/packages/palace/src/__tests__/extraction.test.ts b/packages/palace/src/__tests__/extraction.test.ts
new file mode 100644
index 00000000..a787dc7a
--- /dev/null
+++ b/packages/palace/src/__tests__/extraction.test.ts
@@ -0,0 +1,131 @@
+import { describe, it, expect } from 'vitest';
+import {
+  buildExtractionPrompt,
+  parseExtractionResponse,
+  regexFallbackExtraction,
+} from '../extraction.js';
+
+describe('buildExtractionPrompt', () => {
+  it('includes hall types in the prompt', () => {
+    const prompt = buildExtractionPrompt('some content', {
+      hallTypes: ['decisions', 'events', 'discoveries'],
+    });
+    expect(prompt).toContain('decisions, events, discoveries');
+    expect(prompt).toContain('some content');
+  });
+
+  it('includes title when provided', () => {
+    const prompt = buildExtractionPrompt('content', {
+      title: 'My Note',
+      hallTypes: ['decisions'],
+    });
+    expect(prompt).toContain('Title: My Note');
+  });
+
+  it('includes context when provided', () => {
+    const prompt = buildExtractionPrompt('content', {
+      context: 'Work brain',
+      hallTypes: ['decisions'],
+    });
+    expect(prompt).toContain('Context: Work brain');
+  });
+
+  it('omits title/context when not provided', () => {
+    const prompt = buildExtractionPrompt('content', {
+      hallTypes: ['decisions'],
+    });
+    expect(prompt).not.toContain('Title:');
+    expect(prompt).not.toContain('Context:');
+  });
+});
+
+describe('parseExtractionResponse', () => {
+  it('parses valid JSON array', () => {
+    const input = JSON.stringify([
+      { hall: 'decisions', content: 'Use TypeScript', roomSlug: 'tech', entities: ['TypeScript'] },
+    ]);
+    const result = parseExtractionResponse(input);
+    expect(result).toHaveLength(1);
+    expect(result[0].hall).toBe('decisions');
+    expect(result[0].content).toBe('Use TypeScript');
+    expect(result[0].roomSlug).toBe('tech');
+    expect(result[0].entities).toEqual(['TypeScript']);
+  });
+
+  it('handles JSON wrapped in code fences', () => {
+    const input =
+      '```json\n[{"hall":"events","content":"Released v2","roomSlug":"releases","entities":[]}]\n```';
+    const result = parseExtractionResponse(input);
+    expect(result).toHaveLength(1);
+    expect(result[0].hall).toBe('events');
+  });
+
+  it('returns empty array for malformed JSON', () => {
+    expect(parseExtractionResponse('not json at all')).toEqual([]);
+  });
+
+  it('returns empty array for empty string', () => {
+    expect(parseExtractionResponse('')).toEqual([]);
+  });
+
+  it('filters out items missing required fields', () => {
+    const input = JSON.stringify([{ hall: 'decisions', content: 'valid' }, { noHall: true }]);
+    const result = parseExtractionResponse(input);
+    expect(result).toHaveLength(1);
+  });
+
+  it('defaults roomSlug to general', () => {
+    const input = JSON.stringify([{ hall: 'events', content: 'something' }]);
+    const result = parseExtractionResponse(input);
+    expect(result[0].roomSlug).toBe('general');
+  });
+
+  it('handles room_slug snake_case variant', () => {
+    const input = JSON.stringify([{ hall: 'events', content: 'x', room_slug: 'my-room' }]);
+    const result = parseExtractionResponse(input);
+    expect(result[0].roomSlug).toBe('my-room');
+  });
+});
+
+describe('regexFallbackExtraction', () => {
+  it('extracts Decision: lines', () => {
+    const result = regexFallbackExtraction('Decision: Use Cosmos DB for storage');
+    expect(result).toHaveLength(1);
+    expect(result[0].hall).toBe('decisions');
+    expect(result[0].content).toContain('Use Cosmos DB');
+  });
+
+  it('extracts TODO: lines', () => {
+    const result = regexFallbackExtraction('TODO: Fix the auth bug');
+    expect(result).toHaveLength(1);
+    expect(result[0].hall).toBe('decisions');
+  });
+
+  it('extracts Found: lines as discoveries', () => {
+    const result = regexFallbackExtraction('Found: New API endpoint for search');
+    expect(result).toHaveLength(1);
+    expect(result[0].hall).toBe('discoveries');
+  });
+
+  it('extracts @mentions as entities', () => {
+    const result = regexFallbackExtraction('Decision: @alice proposed the new schema');
+    expect(result[0].entities).toContain('alice');
+  });
+
+  it('extracts #tags as entities', () => {
+    const result = regexFallbackExtraction('Found: #typescript has better types');
+    expect(result[0].entities).toContain('typescript');
+  });
+
+  it('returns empty array for plain text without patterns', () => {
+    expect(regexFallbackExtraction('Just some random text here')).toEqual([]);
+  });
+
+  it('handles multiple patterns in multi-line content', () => {
+    const content = `Decision: Adopt ESM
+Found: Vitest is faster
+TODO: Migrate tests`;
+    const result = regexFallbackExtraction(content);
+    expect(result).toHaveLength(3);
+  });
+});
diff --git a/packages/palace/src/__tests__/halls.test.ts b/packages/palace/src/__tests__/halls.test.ts
new file mode 100644
index 00000000..dd07b0c4
--- /dev/null
+++ b/packages/palace/src/__tests__/halls.test.ts
@@ -0,0 +1,75 @@
+import { describe, it, expect } from 'vitest';
+import { HALL_PRESETS, getHallPreset, hallFromLabel, ALL_HALL_TYPES } from '../halls.js';
+
+describe('HALL_PRESETS', () => {
+  it('notelett has 6 halls', () => {
+    expect(HALL_PRESETS.notelett.halls).toHaveLength(6);
+    expect(HALL_PRESETS.notelett.halls).toContain('insights');
+    expect(HALL_PRESETS.notelett.halls).not.toContain('errors');
+  });
+
+  it('mindlyst has 6 halls with patterns and emotions', () => {
+    expect(HALL_PRESETS.mindlyst.halls).toHaveLength(6);
+    expect(HALL_PRESETS.mindlyst.halls).toContain('patterns');
+    expect(HALL_PRESETS.mindlyst.halls).toContain('emotions');
+  });
+
+  it('coding has 6 halls with errors and advice', () => {
+    expect(HALL_PRESETS.coding.halls).toHaveLength(6);
+    expect(HALL_PRESETS.coding.halls).toContain('errors');
+    expect(HALL_PRESETS.coding.halls).toContain('advice');
+  });
+});
+
+describe('getHallPreset', () => {
+  it('returns preset by name', () => {
+    expect(getHallPreset('notelett')?.name).toBe('NoteLett');
+  });
+
+  it('returns undefined for unknown preset', () => {
+    expect(getHallPreset('nonexistent')).toBeUndefined();
+  });
+});
+
+describe('hallFromLabel', () => {
+  it('matches exact hall type', () => {
+    expect(hallFromLabel('decisions')).toBe('decisions');
+  });
+
+  it('is case-insensitive', () => {
+    expect(hallFromLabel('DECISIONS')).toBe('decisions');
+  });
+
+  it('matches singular form via synonym', () => {
+    expect(hallFromLabel('decision')).toBe('decisions');
+  });
+
+  it('maps synonyms to halls', () => {
+    expect(hallFromLabel('bug')).toBe('errors');
+    expect(hallFromLabel('feeling')).toBe('emotions');
+    expect(hallFromLabel('tip')).toBe('advice');
+    expect(hallFromLabel('trend')).toBe('patterns');
+    expect(hallFromLabel('fact')).toBe('discoveries');
+  });
+
+  it('respects allowedHalls filter', () => {
+    // 'errors' is not in notelett preset
+    expect(hallFromLabel('bug', HALL_PRESETS.notelett.halls)).toBeUndefined();
+    // 'errors' is in coding preset
+    expect(hallFromLabel('bug', HALL_PRESETS.coding.halls)).toBe('errors');
+  });
+
+  it('returns undefined for unrecognized label', () => {
+    expect(hallFromLabel('xyzzy')).toBeUndefined();
+  });
+
+  it('trims whitespace', () => {
+    expect(hallFromLabel('  events  ')).toBe('events');
+  });
+});
+
+describe('ALL_HALL_TYPES', () => {
+  it('has 9 total hall types', () => {
+    expect(ALL_HALL_TYPES).toHaveLength(9);
+  });
+});
diff --git a/packages/palace/src/__tests__/kg.test.ts b/packages/palace/src/__tests__/kg.test.ts
new file mode 100644
index 00000000..40590f1d
--- /dev/null
+++ b/packages/palace/src/__tests__/kg.test.ts
@@ -0,0 +1,113 @@
+import { describe, it, expect } from 'vitest';
+import { findContradictions, mergeTriples, isTripleCurrent } from '../kg.js';
+import type { TripleInput } from '../kg.js';
+
+const now = new Date('2025-06-01T00:00:00Z');
+
+describe('isTripleCurrent', () => {
+  it('returns true for triple with no validTo', () => {
+    expect(isTripleCurrent({ validTo: undefined }, now)).toBe(true);
+  });
+
+  it('returns true for triple with future validTo', () => {
+    expect(isTripleCurrent({ validTo: '2026-01-01T00:00:00Z' }, now)).toBe(true);
+  });
+
+  it('returns false for triple with past validTo', () => {
+    expect(isTripleCurrent({ validTo: '2024-01-01T00:00:00Z' }, now)).toBe(false);
+  });
+});
+
+describe('findContradictions', () => {
+  it('detects contradiction (same subject+predicate, different object)', () => {
+    const existing: TripleInput[] = [
+      { subject: 'App', predicate: 'uses', object: 'PostgreSQL', validFrom: '2025-01-01' },
+    ];
+    const incoming: TripleInput[] = [
+      { subject: 'App', predicate: 'uses', object: 'Cosmos DB', validFrom: '2025-06-01' },
+    ];
+    const result = findContradictions(existing, incoming, now);
+    expect(result).toHaveLength(1);
+    expect(result[0].existing.object).toBe('PostgreSQL');
+    expect(result[0].incoming.object).toBe('Cosmos DB');
+  });
+
+  it('does not flag same triple as contradiction', () => {
+    const existing: TripleInput[] = [
+      { subject: 'App', predicate: 'uses', object: 'Cosmos DB', validFrom: '2025-01-01' },
+    ];
+    const incoming: TripleInput[] = [
+      { subject: 'App', predicate: 'uses', object: 'Cosmos DB', validFrom: '2025-06-01' },
+    ];
+    expect(findContradictions(existing, incoming, now)).toHaveLength(0);
+  });
+
+  it('ignores expired triples', () => {
+    const existing: TripleInput[] = [
+      {
+        subject: 'App',
+        predicate: 'uses',
+        object: 'PostgreSQL',
+        validFrom: '2024-01-01',
+        validTo: '2024-12-31',
+      },
+    ];
+    const incoming: TripleInput[] = [
+      { subject: 'App', predicate: 'uses', object: 'Cosmos DB', validFrom: '2025-06-01' },
+    ];
+    expect(findContradictions(existing, incoming, now)).toHaveLength(0);
+  });
+
+  it('is case-insensitive', () => {
+    const existing: TripleInput[] = [
+      { subject: 'app', predicate: 'Uses', object: 'PostgreSQL', validFrom: '2025-01-01' },
+    ];
+    const incoming: TripleInput[] = [
+      { subject: 'App', predicate: 'uses', object: 'Cosmos DB', validFrom: '2025-06-01' },
+    ];
+    expect(findContradictions(existing, incoming, now)).toHaveLength(1);
+  });
+});
+
+describe('mergeTriples', () => {
+  it('adds non-conflicting triples', () => {
+    const existing: TripleInput[] = [
+      { subject: 'App', predicate: 'uses', object: 'TypeScript', validFrom: '2025-01-01' },
+    ];
+    const incoming: TripleInput[] = [
+      { subject: 'App', predicate: 'runs-on', object: 'Node.js', validFrom: '2025-06-01' },
+    ];
+    const result = mergeTriples(existing, incoming, now);
+    expect(result.added).toHaveLength(1);
+    expect(result.skipped).toHaveLength(0);
+    expect(result.invalidated).toHaveLength(0);
+    expect(result.merged).toHaveLength(2);
+  });
+
+  it('skips duplicate triples', () => {
+    const existing: TripleInput[] = [
+      { subject: 'App', predicate: 'uses', object: 'TypeScript', validFrom: '2025-01-01' },
+    ];
+    const incoming: TripleInput[] = [
+      { subject: 'App', predicate: 'uses', object: 'TypeScript', validFrom: '2025-06-01' },
+    ];
+    const result = mergeTriples(existing, incoming, now);
+    expect(result.skipped).toHaveLength(1);
+    expect(result.added).toHaveLength(0);
+  });
+
+  it('invalidates contradicted triples', () => {
+    const existing: TripleInput[] = [
+      { subject: 'App', predicate: 'uses', object: 'PostgreSQL', validFrom: '2025-01-01' },
+    ];
+    const incoming: TripleInput[] = [
+      { subject: 'App', predicate: 'uses', object: 'Cosmos DB', validFrom: '2025-06-01' },
+    ];
+    const result = mergeTriples(existing, incoming, now);
+    expect(result.invalidated).toHaveLength(1);
+    expect(result.added).toHaveLength(1);
+    // The invalidated triple should have validTo set
+    const invalidated = result.merged.find(t => t.object === 'PostgreSQL');
+    expect(invalidated?.validTo).toBeDefined();
+  });
+});
diff --git a/packages/palace/src/__tests__/wakeup.test.ts b/packages/palace/src/__tests__/wakeup.test.ts
new file mode 100644
index 00000000..a4e58b4d
--- /dev/null
+++ b/packages/palace/src/__tests__/wakeup.test.ts
@@ -0,0 +1,96 @@
+import { describe, it, expect } from 'vitest';
+import {
+  buildWakeUpLayers,
+  truncateToTokenBudget,
+  estimateTokens,
+  WAKEUP_PRESETS,
+} from '../wakeup.js';
+
+describe('estimateTokens', () => {
+  it('estimates ~4 chars per token', () => {
+    expect(estimateTokens('abcd')).toBe(1);
+    expect(estimateTokens('abcdefgh')).toBe(2);
+  });
+
+  it('returns 0 for empty string', () => {
+    expect(estimateTokens('')).toBe(0);
+  });
+});
+
+describe('truncateToTokenBudget', () => {
+  it('returns unchanged text if within budget', () => {
+    expect(truncateToTokenBudget('short', 100)).toBe('short');
+  });
+
+  it('truncates long text with ellipsis', () => {
+    const longText = 'a'.repeat(1000);
+    const result = truncateToTokenBudget(longText, 10); // 40 chars max
+    expect(result.length).toBeLessThanOrEqual(43); // 40 + '...'
+    expect(result.endsWith('...')).toBe(true);
+  });
+
+  it('returns empty string for empty input', () => {
+    expect(truncateToTokenBudget('', 100)).toBe('');
+  });
+});
+
+describe('buildWakeUpLayers', () => {
+  const config = WAKEUP_PRESETS.notelett; // 600 total, 50/150/400
+
+  it('assembles all three layers', () => {
+    const result = buildWakeUpLayers(
+      'Project: NoteLett',
+      'Key fact: Uses Cosmos DB',
+      'Related: User asked about search',
+      config
+    );
+    expect(result.text).toContain('[Identity]');
+    expect(result.text).toContain('[Critical Facts]');
+    expect(result.text).toContain('[Relevant Memories]');
+    expect(result.layers).toHaveLength(3);
+  });
+
+  it('handles empty L0', () => {
+    const result = buildWakeUpLayers('', 'facts', 'memories', config);
+    expect(result.layers.find(l => l.label === 'L0:identity')).toBeUndefined();
+    expect(result.layers).toHaveLength(2);
+  });
+
+  it('handles all empty layers gracefully', () => {
+    const result = buildWakeUpLayers('', '', '', config);
+    expect(result.text).toBe('');
+    expect(result.layers).toHaveLength(0);
+    expect(result.truncated).toBe(false);
+  });
+
+  it('marks truncated when content exceeds budget', () => {
+    const longL2 = 'x '.repeat(5000);
+    const result = buildWakeUpLayers('id', 'fact', longL2, config);
+    expect(result.truncated).toBe(true);
+  });
+
+  it('respects total budget', () => {
+    const result = buildWakeUpLayers(
+      'identity context here',
+      'critical facts here',
+      'semantic memories here',
+      config
+    );
+    const tokens = estimateTokens(result.text);
+    expect(tokens).toBeLessThanOrEqual(config.totalBudget + 10); // small margin for headers
+  });
+});
+
+describe('WAKEUP_PRESETS', () => {
+  it('has notelett preset with 600 budget', () => {
+    expect(WAKEUP_PRESETS.notelett.totalBudget).toBe(600);
+  });
+
+  it('has mindlyst preset with 800 budget', () => {
+    expect(WAKEUP_PRESETS.mindlyst.totalBudget).toBe(800);
+  });
+
+  it('has coding preset with 800 budget', () => {
+    expect(WAKEUP_PRESETS.coding.totalBudget).toBe(800);
+  });
+});
diff --git a/packages/palace/src/config.ts b/packages/palace/src/config.ts
new file mode 100644
index 00000000..569c9b5e
--- /dev/null
+++ b/packages/palace/src/config.ts
@@ -0,0 +1,36 @@
+/**
+ * Palace configuration schema — Zod-based env var validation.
+ *
+ * Products extend their backend config with these palace-specific vars.
+ */
+
+import { z } from 'zod';
+
+export const palaceConfigSchema = z.object({
+  PALACE_ENABLED: z
+    .enum(['true', 'false'])
+    .default('true')
+    .transform(v => v === 'true'),
+
+  PALACE_WAKE_UP_BUDGET: z.coerce.number().default(800),
+
+  PALACE_DEDUP_THRESHOLD: z.coerce.number().min(0).max(1).default(0.9),
+
+  PALACE_RELEVANCE_HALF_LIFE_DAYS: z.coerce.number().min(1).default(30),
+
+  PALACE_MAX_MEMORIES_PER_SEARCH: z.coerce.number().min(1).default(20),
+
+  PALACE_EXTRACTION_MAX_CHARS: z.coerce.number().min(100).default(6000),
+});
+
+export type PalaceConfig = z.infer<typeof palaceConfigSchema>;
+
+/**
+ * Parse palace config from environment.
+ * Products typically merge this with their own config schema.
+ */
+export function parsePalaceConfig(
+  env: Record<string, string | undefined> = process.env as Record<string, string | undefined>
+): PalaceConfig {
+  return palaceConfigSchema.parse(env);
+}
diff --git a/packages/palace/src/cosine.ts b/packages/palace/src/cosine.ts
new file mode 100644
index 00000000..c6b5ae2f
--- /dev/null
+++ b/packages/palace/src/cosine.ts
@@ -0,0 +1,70 @@
+/**
+ * Vector similarity utilities for semantic search and deduplication.
+ */
+
+/**
+ * Compute cosine similarity between two vectors.
+ * Returns a value between -1 and 1 (1 = identical direction).
+ * Returns 0 if either vector is zero-length or dimensions don't match.
+ */
+export function cosineSimilarity(a: number[], b: number[]): number {
+  if (a.length !== b.length || a.length === 0) return 0;
+
+  let dotProduct = 0;
+  let normA = 0;
+  let normB = 0;
+
+  for (let i = 0; i < a.length; i++) {
+    dotProduct += a[i] * b[i];
+    normA += a[i] * a[i];
+    normB += b[i] * b[i];
+  }
+
+  const denominator = Math.sqrt(normA) * Math.sqrt(normB);
+  if (denominator === 0) return 0;
+
+  return dotProduct / denominator;
+}
+
+/**
+ * Normalize a vector to unit length (magnitude = 1).
+ * Returns a zero vector if input is zero-length.
+ */
+export function normalizeVector(v: number[]): number[] {
+  const magnitude = Math.sqrt(v.reduce((sum, val) => sum + val * val, 0));
+  if (magnitude === 0) return v.map(() => 0);
+  return v.map(val => val / magnitude);
+}
+
+/**
+ * Find the top-K most similar items to a query vector.
+ *
+ * @param query - The query embedding vector
+ * @param items - Array of items to search
+ * @param getEmbedding - Function to extract embedding from an item (returns undefined if missing)
+ * @param k - Maximum number of results to return
+ * @param minScore - Minimum cosine similarity score (default: 0)
+ * @returns Sorted array of { item, score } pairs, highest score first
+ */
+export function topKByCosine<T>(
+  query: number[],
+  items: T[],
+  getEmbedding: (item: T) => number[] | undefined,
+  k: number,
+  minScore = 0
+): Array<{ item: T; score: number }> {
+  const scored: Array<{ item: T; score: number }> = [];
+
+  for (const item of items) {
+    const embedding = getEmbedding(item);
+    if (!embedding || embedding.length === 0) continue;
+
+    const score = cosineSimilarity(query, embedding);
+    if (score >= minScore) {
+      scored.push({ item, score });
+    }
+  }
+
+  scored.sort((a, b) => b.score - a.score);
+  return scored.slice(0, k);
+}
diff --git a/packages/palace/src/decay.ts b/packages/palace/src/decay.ts
new file mode 100644
index 00000000..122e85f1
--- /dev/null
+++ b/packages/palace/src/decay.ts
@@ -0,0 +1,52 @@
+/**
+ * Relevance decay for palace memories.
+ *
+ * Uses exponential half-life decay: relevance halves every N days.
+ * Memories that are accessed/referenced get their relevance boosted.
+ */
+
+const MS_PER_DAY = 86_400_000;
+
+/**
+ * Compute decayed relevance using exponential half-life.
+ *
+ * @param originalRelevance - Initial relevance score (0-1)
+ * @param createdAt - ISO date string or Date when the memory was created/last accessed
+ * @param halfLifeDays - Number of days for relevance to halve (default: 30)
+ * @param asOf - Reference time for decay calculation (default: now)
+ * @returns Decayed relevance score (0-1)
+ */
+export function computeDecayedRelevance(
+  originalRelevance: number,
+  createdAt: string | Date,
+  halfLifeDays = 30,
+  asOf: Date = new Date()
+): number {
+  if (halfLifeDays <= 0) return originalRelevance;
+  if (originalRelevance <= 0) return 0;
+
+  const created = typeof createdAt === 'string' ? new Date(createdAt) : createdAt;
+  const elapsedMs = asOf.getTime() - created.getTime();
+
+  if (elapsedMs <= 0) return Math.min(originalRelevance, 1);
+
+  const elapsedDays = elapsedMs / MS_PER_DAY;
+  const decayFactor = Math.pow(0.5, elapsedDays / halfLifeDays);
+
+  return Math.max(0, Math.min(1, originalRelevance * decayFactor));
+}
+
+/**
+ * Compute a boosted relevance for a memory that was accessed/referenced.
+ *
+ * Boost formula: new = old + (1 - old) * boostFactor
+ * This asymptotically approaches 1.0 without exceeding it.
+ *
+ * @param currentRelevance - Current relevance score (0-1)
+ * @param boostFactor - How much of the remaining gap to close (default: 0.3)
+ * @returns Boosted relevance score (0-1)
+ */
+export function boostRelevance(currentRelevance: number, boostFactor = 0.3): number {
+  const boosted = currentRelevance + (1 - currentRelevance) * boostFactor;
+  return Math.min(1, Math.max(0, boosted));
+}
diff --git a/packages/palace/src/dedup.ts b/packages/palace/src/dedup.ts
new file mode 100644
index 00000000..9fe604e6
--- /dev/null
+++ b/packages/palace/src/dedup.ts
@@ -0,0 +1,64 @@
+/**
+ * Deduplication utilities for palace memories.
+ *
+ * Detects near-duplicate content using cosine similarity over embeddings.
+ * Products handle the Cosmos/DB queries; this module operates on pure data.
+ */
+
+import { cosineSimilarity } from './cosine.js';
+
+/**
+ * Check if a candidate embedding is a near-duplicate of any existing embedding.
+ *
+ * @param candidate - Embedding of the new memory
+ * @param existingEmbeddings - Embeddings of existing memories in the same room/hall
+ * @param threshold - Cosine similarity threshold (default: 0.90)
+ * @returns true if any existing embedding exceeds the threshold
+ */
+export function isContentDuplicate(
+  candidate: number[],
+  existingEmbeddings: number[][],
+  threshold = 0.9
+): boolean {
+  for (const existing of existingEmbeddings) {
+    if (existing.length !== candidate.length) continue;
+    if (cosineSimilarity(candidate, existing) > threshold) {
+      return true;
+    }
+  }
+  return false;
+}
+
+/**
+ * Check if two text strings are exact duplicates after normalization.
+ * Trims whitespace and lowercases before comparison.
+ */
+export function isExactDuplicate(a: string, b: string): boolean {
+  return a.trim().toLowerCase() === b.trim().toLowerCase();
+}
+
+/**
+ * Find the most similar embedding and return its index + score.
+ * Returns null if no embeddings exist or none exceed minScore.
+ */
+export function findClosestMatch(
+  candidate: number[],
+  existingEmbeddings: number[][],
+  minScore = 0
+): { index: number; score: number } | null {
+  let bestIndex = -1;
+  let bestScore = minScore;
+
+  for (let i = 0; i < existingEmbeddings.length; i++) {
+    const existing = existingEmbeddings[i];
+    if (existing.length !== candidate.length) continue;
+
+    const score = cosineSimilarity(candidate, existing);
+    if (score > bestScore) {
+      bestScore = score;
+      bestIndex = i;
+    }
+  }
+
+  return bestIndex >= 0 ? { index: bestIndex, score: bestScore } : null;
+}
diff --git a/packages/palace/src/extraction.ts b/packages/palace/src/extraction.ts
new file mode 100644
index 00000000..bd3c3049
--- /dev/null
+++ b/packages/palace/src/extraction.ts
@@ -0,0 +1,154 @@
+/**
+ * Memory extraction utilities — prompt building, response parsing, regex fallback.
+ *
+ * Products call their own LLM provider with the prompt from buildExtractionPrompt(),
+ * then pass the response to parseExtractionResponse().
+ * If LLM is unavailable, regexFallbackExtraction() provides basic extraction.
+ */
+
+import type { ExtractedMemory } from './types.js';
+
+export interface ExtractionContext {
+  title?: string;
+  context?: string;
+  hallTypes: readonly string[];
+}
+
+/**
+ * Build a structured extraction prompt for an LLM.
+ *
+ * @param content - The text content to extract memories from
+ * @param ctx - Context including title, additional context, and allowed hall types
+ * @returns A system/user prompt string ready for LLM chat()
+ */
+export function buildExtractionPrompt(content: string, ctx: ExtractionContext): string {
+  const hallList = ctx.hallTypes.join(', ');
+  const titleLine = ctx.title ? `\nTitle: ${ctx.title}` : '';
+  const contextLine = ctx.context ? `\nContext: ${ctx.context}` : '';
+
+  return `Extract structured memories from the following content.
+
+For each distinct memory, return a JSON array where each element has:
+- "hall": one of [${hallList}]
+- "content": the memory summarized in 1-2 sentences
+- "roomSlug": a short kebab-case topic slug (e.g. "auth-migration", "api-design")
+- "entities": array of named entities mentioned (people, projects, technologies, places)
+
+Rules:
+- Only extract genuinely important or referenceable facts, decisions, or events
+- Skip trivial or obvious statements
+- Each memory should be self-contained (understandable without the original context)
+- Prefer specific details over vague summaries
+- Return valid JSON only — no markdown fences, no explanation${titleLine}${contextLine}
+
+Content:
+${content}`;
+}
+
+/**
+ * Parse an LLM extraction response into ExtractedMemory[].
+ *
+ * Handles:
+ * - Clean JSON arrays
+ * - JSON wrapped in markdown code fences
+ * - Malformed JSON (returns empty array)
+ */
+export function parseExtractionResponse(llmOutput: string): ExtractedMemory[] {
+  if (!llmOutput || llmOutput.trim().length === 0) return [];
+
+  let cleaned = llmOutput.trim();
+
+  // Strip markdown code fences if present
+  if (cleaned.startsWith('```')) {
+    cleaned = cleaned.replace(/^```(?:json)?\s*\n?/, '').replace(/\n?```\s*$/, '');
+  }
+
+  try {
+    const parsed = JSON.parse(cleaned);
+
+    if (!Array.isArray(parsed)) return [];
+
+    return parsed
+      .filter(
+        (item: unknown): item is Record<string, unknown> =>
+          typeof item === 'object' && item !== null && 'hall' in item && 'content' in item
+      )
+      .map(item => ({
+        hall: String(item.hall || ''),
+        content: String(item.content || ''),
+        roomSlug: String(item.roomSlug || item.room_slug || 'general'),
+        entities: Array.isArray(item.entities) ? item.entities.map(String) : [],
+      }));
+  } catch {
+    return [];
+  }
+}
+
+/**
+ * Regex-based fallback extraction when LLM is unavailable.
+ *
+ * Scans for common patterns:
+ * - "Decision:" / "Decided:" → decisions
+ * - "TODO:" / "Action:" → decisions
+ * - "Found:" / "Discovered:" / "Learned:" → discoveries
+ * - "Prefer:" / "Always:" / "Never:" → preferences
+ * - "Event:" / "Happened:" / date patterns → events
+ * - "Tip:" / "Note:" / "Remember:" → advice
+ *
+ * @param content - Raw text content
+ * @returns Array of extracted memories (best-effort)
+ */
+export function regexFallbackExtraction(content: string): ExtractedMemory[] {
+  const memories: ExtractedMemory[] = [];
+  const lines = content.split('\n');
+
+  const patterns: Array<{ regex: RegExp; hall: string }> = [
+    { regex: /^(?:decision|decided|resolve[ds]?):\s*(.+)/i, hall: 'decisions' },
+    { regex: /^(?:todo|action|task):\s*(.+)/i, hall: 'decisions' },
+    { regex: /^(?:found|discovered|learned|til):\s*(.+)/i, hall: 'discoveries' },
+    { regex: /^(?:prefer|always|never):\s*(.+)/i, hall: 'preferences' },
+    { regex: /^(?:event|happened|occurred):\s*(.+)/i, hall: 'events' },
+    { regex: /^(?:tip|note|remember|important):\s*(.+)/i, hall: 'advice' },
+    { regex: /^(?:error|bug|issue|broken):\s*(.+)/i, hall: 'errors' },
+    { regex: /^(?:pattern|recurring|trend):\s*(.+)/i, hall: 'patterns' },
+    { regex: /^(?:feeling|mood|emotion):\s*(.+)/i, hall: 'emotions' },
+    { regex: /^(?:insight|observation|noticed):\s*(.+)/i, hall: 'insights' },
+  ];
+
+  for (const line of lines) {
+    const trimmed = line.replace(/^[\s\-*>#]+/, '').trim();
+    if (!trimmed) continue;
+
+    for (const { regex, hall } of patterns) {
+      const match = trimmed.match(regex);
+      if (match && match[1]) {
+        memories.push({
+          hall,
+          content: match[1].trim(),
+          roomSlug: 'general',
+          entities: extractEntities(match[1]),
+        });
+        break;
+      }
+    }
+  }
+
+  return memories;
+}
+
+/**
+ * Extract simple entities from text (mentions, tags, capitalized phrases).
+ */
+function extractEntities(text: string): string[] {
+  const entities = new Set<string>();
+
+  // @mentions
+  const mentions = text.match(/@(\w+)/g);
+  if (mentions) mentions.forEach(m => entities.add(m.slice(1)));
+
+  // #tags
+  const tags = text.match(/#(\w+)/g);
+  if (tags) tags.forEach(t => entities.add(t.slice(1)));
+
+  return Array.from(entities);
+}
diff --git a/packages/palace/src/halls.ts b/packages/palace/src/halls.ts
new file mode 100644
index 00000000..3e4220f6
--- /dev/null
+++ b/packages/palace/src/halls.ts
@@ -0,0 +1,103 @@
+/**
+ * Hall types and presets for different products.
+ *
+ * Each product picks a preset (or defines custom halls).
+ * Halls categorize memories by type — decisions, events, discoveries, etc.
+ */
+
+export const ALL_HALL_TYPES = [
+  'decisions',
+  'events',
+  'discoveries',
+  'preferences',
+  'advice',
+  'insights',
+  'patterns',
+  'emotions',
+  'errors',
+] as const;
+
+export type HallType = (typeof ALL_HALL_TYPES)[number];
+
+export interface HallPreset {
+  name: string;
+  halls: HallType[];
+}
+
+/**
+ * Product-specific hall presets.
+ *
+ * - notelett: insights instead of errors (note-taking domain)
+ * - mindlyst: patterns + emotions (multimodal/emotional domain)
+ * - coding: errors + advice (developer/agent domain, e.g. Claw-Cowork)
+ */
+export const HALL_PRESETS: Record<string, HallPreset> = {
+  notelett: {
+    name: 'NoteLett',
+    halls: ['decisions', 'events', 'discoveries', 'preferences', 'advice', 'insights'],
+  },
+  mindlyst: {
+    name: 'MindLyst',
+    halls: ['decisions', 'events', 'discoveries', 'preferences', 'patterns', 'emotions'],
+  },
+  coding: {
+    name: 'Coding Agent',
+    halls: ['decisions', 'events', 'discoveries', 'preferences', 'advice', 'errors'],
+  },
+};
+
+/**
+ * Get a hall preset by name.
+ * Returns undefined if the preset does not exist.
+ */
+export function getHallPreset(presetName: string): HallPreset | undefined {
+  return HALL_PRESETS[presetName];
+}
+
+/**
+ * Classify a label string to the closest hall type.
+ * Case-insensitive, tries exact match first, then substring.
+ * Returns undefined if no match.
+ */
+export function hallFromLabel(label: string, allowedHalls?: HallType[]): HallType | undefined {
+  const normalized = label.toLowerCase().trim();
+  const candidates = allowedHalls ?? (ALL_HALL_TYPES as unknown as HallType[]);
+
+  // Exact match
+  const exact = candidates.find(h => h === normalized);
+  if (exact) return exact;
+
+  // Substring match (e.g. "decision" → "decisions")
+  const partial = candidates.find(h => h.startsWith(normalized) || normalized.startsWith(h));
+  if (partial) return partial;
+
+  // Common synonyms
+  const synonymMap: Record<string, HallType> = {
+    decision: 'decisions',
+    event: 'events',
+    discovery: 'discoveries',
+    preference: 'preferences',
+    insight: 'insights',
+    pattern: 'patterns',
+    emotion: 'emotions',
+    error: 'errors',
+    fact: 'discoveries',
+    finding: 'discoveries',
+    todo: 'decisions',
+    task: 'decisions',
+    bug: 'errors',
+    fix: 'decisions',
+    feeling: 'emotions',
+    mood: 'emotions',
+    trend: 'patterns',
+    recurring: 'patterns',
+    tip: 'advice',
+    recommendation: 'advice',
+    suggestion: 'advice',
+  };
+
+  const synonym = synonymMap[normalized];
+  if (synonym && candidates.includes(synonym)) return synonym;
+
+  return undefined;
+}
diff --git a/packages/palace/src/index.ts b/packages/palace/src/index.ts
new file mode 100644
index 00000000..4689d58b
--- /dev/null
+++ b/packages/palace/src/index.ts
@@ -0,0 +1,48 @@
+// ── Types ──────────────────────────────────────────────────────────
+export type {
+  BasePalaceWingDoc,
+  BasePalaceRoomDoc,
+  BasePalaceMemoryDoc,
+  BasePalaceTunnelDoc,
+  BasePalaceKGTripleDoc,
+  BasePalaceDiaryDoc,
+  ExtractedMemory,
+} from './types.js';
+
+// ── Halls ──────────────────────────────────────────────────────────
+export { ALL_HALL_TYPES, HALL_PRESETS, getHallPreset, hallFromLabel } from './halls.js';
+export type { HallType, HallPreset } from './halls.js';
+
+// ── Cosine Similarity ──────────────────────────────────────────────
+export { cosineSimilarity, normalizeVector, topKByCosine } from './cosine.js';
+
+// ── Deduplication ──────────────────────────────────────────────────
+export { isContentDuplicate, isExactDuplicate, findClosestMatch } from './dedup.js';
+
+// ── Relevance Decay ────────────────────────────────────────────────
+export { computeDecayedRelevance, boostRelevance } from './decay.js';
+
+// ── Extraction ─────────────────────────────────────────────────────
+export {
+  buildExtractionPrompt,
+  parseExtractionResponse,
+  regexFallbackExtraction,
+} from './extraction.js';
+export type { ExtractionContext } from './extraction.js';
+
+// ── Knowledge Graph ────────────────────────────────────────────────
+export { findContradictions, mergeTriples, isTripleCurrent } from './kg.js';
+export type { TripleInput } from './kg.js';
+
+// ── Wake-Up Context ────────────────────────────────────────────────
+export {
+  buildWakeUpLayers,
+  truncateToTokenBudget,
+  estimateTokens,
+  WAKEUP_PRESETS,
+} from './wakeup.js';
+export type { WakeUpConfig, WakeUpContext, WakeUpLayer } from './wakeup.js';
+
+// ── Config ─────────────────────────────────────────────────────────
+export { palaceConfigSchema, parsePalaceConfig } from './config.js';
+export type { PalaceConfig } from './config.js';
diff --git a/packages/palace/src/kg.ts b/packages/palace/src/kg.ts
new file mode 100644
index 00000000..def00520
--- /dev/null
+++ b/packages/palace/src/kg.ts
@@ -0,0 +1,138 @@
+/**
+ * Knowledge graph helpers for palace triple management.
+ *
+ * Triples are (subject, predicate, object) with temporal validity.
+ * This module provides pure functions for contradiction detection,
+ * merging, and currency checking.
+ */
+
+/**
+ * A lightweight triple for comparison (does not require full doc fields).
+ */
+export interface TripleInput {
+  subject: string;
+  predicate: string;
+  object: string;
+  validFrom: string;
+  validTo?: string;
+  confidence?: number;
+}
+
+/**
+ * Find contradictions between existing triples and incoming ones.
+ *
+ * A contradiction exists when:
+ * - Same subject + predicate, different object
+ * - Both are currently valid (no validTo or validTo in the future)
+ *
+ * @returns Array of { existing, incoming } contradiction pairs
+ */
+export function findContradictions(
+  existing: TripleInput[],
+  incoming: TripleInput[],
+  asOf: Date = new Date()
+): Array<{ existing: TripleInput; incoming: TripleInput }> {
+  const contradictions: Array<{ existing: TripleInput; incoming: TripleInput }> = [];
+
+  for (const inc of incoming) {
+    for (const ext of existing) {
+      if (
+        normalizeEntity(ext.subject) === normalizeEntity(inc.subject) &&
+        normalizeEntity(ext.predicate) === normalizeEntity(inc.predicate) &&
+        normalizeEntity(ext.object) !== normalizeEntity(inc.object) &&
+        isTripleCurrent(ext, asOf) &&
+        isTripleCurrent(inc, asOf)
+      ) {
+        contradictions.push({ existing: ext, incoming: inc });
+      }
+    }
+  }
+
+  return contradictions;
+}
+
+/**
+ * Merge incoming triples into existing set.
+ *
+ * - If an incoming triple contradicts an existing one, the existing triple
+ *   is invalidated (validTo set) and the incoming one is kept.
+ * - If an incoming triple is a duplicate (same S/P/O), it is skipped.
+ * - Otherwise, the incoming triple is added.
+ *
+ * @returns { merged, invalidated, added, skipped } counts
+ */
+export function mergeTriples(
+  existing: TripleInput[],
+  incoming: TripleInput[],
+  asOf: Date = new Date()
+): {
+  merged: TripleInput[];
+  invalidated: TripleInput[];
+  added: TripleInput[];
+  skipped: TripleInput[];
+} {
+  const invalidated: TripleInput[] = [];
+  const added: TripleInput[] = [];
+  const skipped: TripleInput[] = [];
+
+  const merged = [...existing];
+
+  for (const inc of incoming) {
+    // Check for exact duplicate
+    const isDuplicate = merged.some(
+      ext =>
+        normalizeEntity(ext.subject) === normalizeEntity(inc.subject) &&
+        normalizeEntity(ext.predicate) === normalizeEntity(inc.predicate) &&
+        normalizeEntity(ext.object) === normalizeEntity(inc.object) &&
+        isTripleCurrent(ext, asOf)
+    );
+
+    if (isDuplicate) {
+      skipped.push(inc);
+      continue;
+    }
+
+    // Check for contradiction
+    const contradictIdx = merged.findIndex(
+      ext =>
+        normalizeEntity(ext.subject) === normalizeEntity(inc.subject) &&
+        normalizeEntity(ext.predicate) === normalizeEntity(inc.predicate) &&
+        normalizeEntity(ext.object) !== normalizeEntity(inc.object) &&
+        isTripleCurrent(ext, asOf)
+    );
+
+    if (contradictIdx >= 0) {
+      // Invalidate the old triple
+      const old = merged[contradictIdx];
+      merged[contradictIdx] = { ...old, validTo: asOf.toISOString() };
+      invalidated.push(old);
+    }
+
+    merged.push(inc);
+    added.push(inc);
+  }
+
+  return { merged, invalidated, added, skipped };
+}
+
+/**
+ * Check if a triple is currently valid.
+ *
+ * @param triple - The triple to check
+ * @param asOf - Reference time (default: now)
+ * @returns true if the triple has no validTo or validTo is in the future
+ */
+export function isTripleCurrent(
+  triple: Pick<TripleInput, 'validTo'>,
+  asOf: Date = new Date()
+): boolean {
+  if (!triple.validTo) return true;
+  return new Date(triple.validTo).getTime() > asOf.getTime();
+}
+
+/**
+ * Normalize an entity string for comparison (lowercase, trim, collapse whitespace).
+ */
+function normalizeEntity(s: string): string {
+  return s.toLowerCase().trim().replace(/\s+/g, ' ');
+}
diff --git a/packages/palace/src/types.ts b/packages/palace/src/types.ts
new file mode 100644
index 00000000..d62f84b2
--- /dev/null
+++ b/packages/palace/src/types.ts
@@ -0,0 +1,105 @@
+/**
+ * Base palace document types.
+ *
+ * Products extend these with product-specific fields
+ * (e.g. sourceWorkspaceId for NoteLett, sourceBrainId for MindLyst).
+ */
+
+// ── Wing (top-level grouping — workspace/brain/project) ────────────
+
+export interface BasePalaceWingDoc {
+  id: string;
+  productId: string;
+  userId: string;
+  name: string;
+  description?: string;
+  memoryCount: number;
+  l1Cache?: string;
+  l1CacheUpdatedAt?: string;
+  createdAt: string;
+  updatedAt: string;
+}
+
+// ── Room (topic within a wing) ─────────────────────────────────────
+
+export interface BasePalaceRoomDoc {
+  id: string;
+  productId: string;
+  userId: string;
+  wingId: string;
+  name: string;
+  description?: string;
+  memoryCount: number;
+  createdAt: string;
+  updatedAt: string;
+}
+
+// ── Memory (core unit — one fact/decision/event/etc.) ──────────────
+
+export interface BasePalaceMemoryDoc {
+  id: string;
+  productId: string;
+  userId: string;
+  wingId: string;
+  roomId: string;
+  hall: string;
+  content: string;
+  relevance: number;
+  embedding?: number[];
+  sourceId?: string;
+  createdAt: string;
+  updatedAt: string;
+}
+
+// ── Tunnel (cross-room/cross-wing link) ────────────────────────────
+
+export interface BasePalaceTunnelDoc {
+  id: string;
+  productId: string;
+  userId: string;
+  fromMemoryId: string;
+  fromWingId: string;
+  toMemoryId: string;
+  toWingId: string;
+  relationship: string;
+  strength: number;
+  createdAt: string;
+}
+
+// ── Knowledge Graph Triple ─────────────────────────────────────────
+
+export interface BasePalaceKGTripleDoc {
+  id: string;
+  productId: string;
+  userId: string;
+  wingId: string;
+  subject: string;
+  predicate: string;
+  object: string;
+  confidence: number;
+  validFrom: string;
+  validTo?: string;
+  sourceMemoryId?: string;
+  createdAt: string;
+}
+
+// ── Diary Entry ────────────────────────────────────────────────────
+
+export interface BasePalaceDiaryDoc {
+  id: string;
+  productId: string;
+  userId: string;
+  roleId: string;
+  wingId?: string;
+  entry: string;
+  createdAt: string;
+}
+
+// ── Extracted memory (output of extraction pipeline) ───────────────
+
+export interface ExtractedMemory {
+  hall: string;
+  content: string;
+  roomSlug: string;
+  entities: string[];
+}
diff --git a/packages/palace/src/wakeup.ts b/packages/palace/src/wakeup.ts
new file mode 100644
index 00000000..22717b3f
--- /dev/null
+++ b/packages/palace/src/wakeup.ts
@@ -0,0 +1,126 @@
+/**
+ * Wake-up context builder for palace-augmented sessions.
+ *
+ * Builds a layered context string (L0/L1/L2) within a token budget.
+ * Products provide the raw data; this module assembles and truncates.
+ */
+
+export interface WakeUpLayer {
+  label: string;
+  content: string;
+  priority: number;
+}
+
+export interface WakeUpConfig {
+  totalBudget: number;
+  l0Budget: number;
+  l1Budget: number;
+  l2Budget: number;
+}
+
+export interface WakeUpContext {
+  text: string;
+  layers: { label: string; charCount: number }[];
+  totalChars: number;
+  truncated: boolean;
+}
+
+/**
+ * Approximate token count for a string.
+ * Uses the rough heuristic of ~4 characters per token.
+ */
+export function estimateTokens(text: string): number {
+  return Math.ceil(text.length / 4);
+}
+
+/**
+ * Truncate text to fit within a token budget.
+ *
+ * @param text - Input text
+ * @param maxTokens - Maximum tokens allowed
+ * @returns Truncated text (with "..." appended if truncated)
+ */
+export function truncateToTokenBudget(text: string, maxTokens: number): string {
+  if (!text) return '';
+
+  const maxChars = maxTokens * 4;
+  if (text.length <= maxChars) return text;
+
+  // Truncate at word boundary
+  const truncated = text.slice(0, maxChars);
+  const lastSpace = truncated.lastIndexOf(' ');
+  const cutPoint = lastSpace > maxChars * 0.8 ? lastSpace : maxChars;
+
+  return truncated.slice(0, cutPoint) + '...';
+}
+
+/**
+ * Build a wake-up context from L0/L1/L2 layers within a total token budget.
+ *
+ * Layer priority:
+ * - L0 (identity/project context) — always included, smallest budget
+ * - L1 (critical facts from recent memories) — high priority
+ * - L2 (semantically relevant memories) — fills remaining budget
+ *
+ * @param l0 - Identity/project context string
+ * @param l1 - Critical facts string
+ * @param l2 - Semantically relevant memories string
+ * @param config - Token budget configuration
+ * @returns Assembled wake-up context with metadata
+ */
+export function buildWakeUpLayers(
+  l0: string,
+  l1: string,
+  l2: string,
+  config: WakeUpConfig
+): WakeUpContext {
+  const layers: { label: string; charCount: number }[] = [];
+  const parts: string[] = [];
+  let truncated = false;
+
+  // L0: identity (always included)
+  const l0Truncated = truncateToTokenBudget(l0, config.l0Budget);
+  if (l0Truncated) {
+    parts.push(`[Identity]\n${l0Truncated}`);
+    layers.push({ label: 'L0:identity', charCount: l0Truncated.length });
+    if (l0Truncated.endsWith('...')) truncated = true;
+  }
+
+  // L1: critical facts
+  const l1Truncated = truncateToTokenBudget(l1, config.l1Budget);
+  if (l1Truncated) {
+    parts.push(`[Critical Facts]\n${l1Truncated}`);
+    layers.push({ label: 'L1:facts', charCount: l1Truncated.length });
+    if (l1Truncated.endsWith('...')) truncated = true;
+  }
+
+  // L2: semantic context (gets remaining budget)
+  const usedTokens = estimateTokens(parts.join('\n\n'));
+  const remainingBudget = Math.max(0, config.totalBudget - usedTokens);
+  const l2Budget = Math.min(config.l2Budget, remainingBudget);
+
+  const l2Truncated = truncateToTokenBudget(l2, l2Budget);
+  if (l2Truncated) {
+    parts.push(`[Relevant Memories]\n${l2Truncated}`);
+    layers.push({ label: 'L2:semantic', charCount: l2Truncated.length });
+    if (l2Truncated.endsWith('...')) truncated = true;
+  }
+
+  const text = parts.join('\n\n');
+
+  return {
+    text,
+    layers,
+    totalChars: text.length,
+    truncated,
+  };
+}
+
+/**
+ * Default wake-up configs for each product.
+ */
+export const WAKEUP_PRESETS: Record<string, WakeUpConfig> = {
+  notelett: { totalBudget: 600, l0Budget: 50, l1Budget: 150, l2Budget: 400 },
+  mindlyst: { totalBudget: 800, l0Budget: 80, l1Budget: 200, l2Budget: 500 },
+  coding: { totalBudget: 800, l0Budget: 80, l1Budget: 200, l2Budget: 500 },
+};
diff --git a/packages/palace/tsconfig.json b/packages/palace/tsconfig.json
new file mode 100644
index 00000000..5edad813
--- /dev/null
+++ b/packages/palace/tsconfig.json
@@ -0,0 +1,9 @@
+{
+  "extends": "../../tsconfig.base.json",
+  "compilerOptions": {
+    "outDir": "dist",
+    "rootDir": "src"
+  },
+  "include": ["src"],
+  "exclude": ["src/**/*.test.ts"]
+}