rag-memory-epf-mcp 3.3.0 → 3.3.2
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/index.js +55 -14
- package/dist/rag-memory.db +0 -0
- package/dist/rag-memory.db-shm +0 -0
- package/dist/rag-memory.db-wal +0 -0
- package/package.json +1 -1
package/dist/index.js
CHANGED
|
@@ -39,6 +39,40 @@ const DB_FILE_PATH = process.env.DB_FILE_PATH
|
|
|
39
39
|
: defaultDbPath;
|
|
40
40
|
const EMBEDDING_MODEL = process.env.EMBEDDING_MODEL || 'Xenova/bge-m3';
|
|
41
41
|
// Safe rowid for vec0 virtual tables (require literal integer, not parameterized)
|
|
42
|
+
// Trim incomplete UTF-8 multi-byte sequences at chunk boundaries.
|
|
43
|
+
// Continuation bytes match 10xxxxxx (0x80-0xBF); lead bytes indicate how many
|
|
44
|
+
// bytes the sequence needs (0xxxxxxx=1, 110xxxxx=2, 1110xxxx=3, 11110xxx=4).
|
|
45
|
+
// When a chunk is not at the document head/tail, any partial sequence at that
|
|
46
|
+
// edge belongs to an adjacent chunk and must be removed so TextDecoder does
|
|
47
|
+
// not emit U+FFFD. Pass trimHead/trimTail=false to preserve head/tail bytes.
|
|
48
|
+
function trimIncompleteUtf8(bytes, trimHead, trimTail) {
|
|
49
|
+
let start = 0;
|
|
50
|
+
let end = bytes.length;
|
|
51
|
+
if (trimHead) {
|
|
52
|
+
while (start < end && (bytes[start] & 0xC0) === 0x80)
|
|
53
|
+
start++;
|
|
54
|
+
}
|
|
55
|
+
if (trimTail) {
|
|
56
|
+
let i = end - 1;
|
|
57
|
+
while (i >= start && (bytes[i] & 0xC0) === 0x80)
|
|
58
|
+
i--;
|
|
59
|
+
if (i >= start) {
|
|
60
|
+
const lead = bytes[i];
|
|
61
|
+
let needed = 1;
|
|
62
|
+
if ((lead & 0x80) === 0)
|
|
63
|
+
needed = 1;
|
|
64
|
+
else if ((lead & 0xE0) === 0xC0)
|
|
65
|
+
needed = 2;
|
|
66
|
+
else if ((lead & 0xF0) === 0xE0)
|
|
67
|
+
needed = 3;
|
|
68
|
+
else if ((lead & 0xF8) === 0xF0)
|
|
69
|
+
needed = 4;
|
|
70
|
+
if (end - i < needed)
|
|
71
|
+
end = i;
|
|
72
|
+
}
|
|
73
|
+
}
|
|
74
|
+
return bytes.subarray(start, end);
|
|
75
|
+
}
|
|
42
76
|
function safeRowid(value) {
|
|
43
77
|
const n = Number(value);
|
|
44
78
|
if (!Number.isInteger(n) || n < 0) {
|
|
@@ -154,7 +188,7 @@ class RAGKnowledgeGraphManager {
|
|
|
154
188
|
VALUES (?, ?, ?, ?, ?)
|
|
155
189
|
`);
|
|
156
190
|
for (const entity of entities) {
|
|
157
|
-
const entityId = `entity_${entity.name.toLowerCase().replace(/[
|
|
191
|
+
const entityId = `entity_${entity.name.toLowerCase().replace(/[^\p{L}\p{N}]/gu, '_')}`;
|
|
158
192
|
const timestamped = (entity.observations || []).map(o => this._timestampObservation(o));
|
|
159
193
|
// Try insert first
|
|
160
194
|
const insertResult = insertStmt.run(entityId, entity.name, entity.entityType, JSON.stringify(timestamped), '{}');
|
|
@@ -199,8 +233,8 @@ class RAGKnowledgeGraphManager {
|
|
|
199
233
|
{ name: relation.from, entityType: 'CONCEPT', observations: [] },
|
|
200
234
|
{ name: relation.to, entityType: 'CONCEPT', observations: [] }
|
|
201
235
|
]);
|
|
202
|
-
const sourceId = `entity_${relation.from.toLowerCase().replace(/[
|
|
203
|
-
const targetId = `entity_${relation.to.toLowerCase().replace(/[
|
|
236
|
+
const sourceId = `entity_${relation.from.toLowerCase().replace(/[^\p{L}\p{N}]/gu, '_')}`;
|
|
237
|
+
const targetId = `entity_${relation.to.toLowerCase().replace(/[^\p{L}\p{N}]/gu, '_')}`;
|
|
204
238
|
const relationId = `rel_${sourceId}_${relation.relationType}_${targetId}`.toLowerCase();
|
|
205
239
|
const stmt = this.db.prepare(`
|
|
206
240
|
INSERT OR IGNORE INTO relationships
|
|
@@ -219,7 +253,7 @@ class RAGKnowledgeGraphManager {
|
|
|
219
253
|
throw new Error('Database not initialized');
|
|
220
254
|
const results = [];
|
|
221
255
|
for (const obs of observations) {
|
|
222
|
-
const entityId = `entity_${obs.entityName.toLowerCase().replace(/[
|
|
256
|
+
const entityId = `entity_${obs.entityName.toLowerCase().replace(/[^\p{L}\p{N}]/gu, '_')}`;
|
|
223
257
|
// Get current observations
|
|
224
258
|
const entity = this.db.prepare(`
|
|
225
259
|
SELECT observations FROM entities WHERE id = ?
|
|
@@ -250,7 +284,7 @@ class RAGKnowledgeGraphManager {
|
|
|
250
284
|
throw new Error('Database not initialized');
|
|
251
285
|
console.error(`🗑️ Deleting entities: ${entityNames.join(', ')}`);
|
|
252
286
|
for (const name of entityNames) {
|
|
253
|
-
const entityId = `entity_${name.toLowerCase().replace(/[
|
|
287
|
+
const entityId = `entity_${name.toLowerCase().replace(/[^\p{L}\p{N}]/gu, '_')}`;
|
|
254
288
|
try {
|
|
255
289
|
// Check if entity exists first
|
|
256
290
|
const entityExists = this.db.prepare(`
|
|
@@ -312,7 +346,7 @@ class RAGKnowledgeGraphManager {
|
|
|
312
346
|
if (!this.db)
|
|
313
347
|
throw new Error('Database not initialized');
|
|
314
348
|
for (const deletion of deletions) {
|
|
315
|
-
const entityId = `entity_${deletion.entityName.toLowerCase().replace(/[
|
|
349
|
+
const entityId = `entity_${deletion.entityName.toLowerCase().replace(/[^\p{L}\p{N}]/gu, '_')}`;
|
|
316
350
|
const entity = this.db.prepare(`
|
|
317
351
|
SELECT observations FROM entities WHERE id = ?
|
|
318
352
|
`).get(entityId);
|
|
@@ -329,8 +363,8 @@ class RAGKnowledgeGraphManager {
|
|
|
329
363
|
if (!this.db)
|
|
330
364
|
throw new Error('Database not initialized');
|
|
331
365
|
for (const relation of relations) {
|
|
332
|
-
const sourceId = `entity_${relation.from.toLowerCase().replace(/[
|
|
333
|
-
const targetId = `entity_${relation.to.toLowerCase().replace(/[
|
|
366
|
+
const sourceId = `entity_${relation.from.toLowerCase().replace(/[^\p{L}\p{N}]/gu, '_')}`;
|
|
367
|
+
const targetId = `entity_${relation.to.toLowerCase().replace(/[^\p{L}\p{N}]/gu, '_')}`;
|
|
334
368
|
this.db.prepare(`
|
|
335
369
|
DELETE FROM relationships
|
|
336
370
|
WHERE source_entity = ? AND target_entity = ? AND relationType = ?
|
|
@@ -343,8 +377,8 @@ class RAGKnowledgeGraphManager {
|
|
|
343
377
|
let updated = 0;
|
|
344
378
|
let notFound = 0;
|
|
345
379
|
for (const update of updates) {
|
|
346
|
-
const sourceId = `entity_${update.from.toLowerCase().replace(/[
|
|
347
|
-
const targetId = `entity_${update.to.toLowerCase().replace(/[
|
|
380
|
+
const sourceId = `entity_${update.from.toLowerCase().replace(/[^\p{L}\p{N}]/gu, '_')}`;
|
|
381
|
+
const targetId = `entity_${update.to.toLowerCase().replace(/[^\p{L}\p{N}]/gu, '_')}`;
|
|
348
382
|
const relationId = `rel_${sourceId}_${update.relationType}_${targetId}`.toLowerCase();
|
|
349
383
|
// Check if relation exists
|
|
350
384
|
const existing = this.db.prepare(`
|
|
@@ -407,7 +441,7 @@ class RAGKnowledgeGraphManager {
|
|
|
407
441
|
// Cap depth at 5 to prevent runaway queries
|
|
408
442
|
const effectiveDepth = Math.min(Math.max(depth, 1), 5);
|
|
409
443
|
// Convert entity names to IDs
|
|
410
|
-
const seedIds = entityNames.map(name => `entity_${name.toLowerCase().replace(/[
|
|
444
|
+
const seedIds = entityNames.map(name => `entity_${name.toLowerCase().replace(/[^\p{L}\p{N}]/gu, '_')}`);
|
|
411
445
|
// Build dynamic placeholders for the seed IDs
|
|
412
446
|
const seedPlaceholders = seedIds.map(() => '?').join(',');
|
|
413
447
|
// Build the recursive CTE query
|
|
@@ -1098,6 +1132,10 @@ class RAGKnowledgeGraphManager {
|
|
|
1098
1132
|
return Array.from(terms);
|
|
1099
1133
|
}
|
|
1100
1134
|
// Tokenize and chunk text
|
|
1135
|
+
// BPE tokenizers (cl100k_base) split multi-byte UTF-8 sequences across tokens.
|
|
1136
|
+
// Slicing token arrays at arbitrary boundaries can leave incomplete UTF-8
|
|
1137
|
+
// prefix/suffix bytes, which TextDecoder replaces with U+FFFD (�). Trim the
|
|
1138
|
+
// incomplete sequences at chunk boundaries; overlap covers the removed bytes.
|
|
1101
1139
|
chunkText(text, maxTokens = 800, overlap = 160) {
|
|
1102
1140
|
if (!this.encoding)
|
|
1103
1141
|
throw new Error('Tokenizer not initialized');
|
|
@@ -1106,7 +1144,10 @@ class RAGKnowledgeGraphManager {
|
|
|
1106
1144
|
for (let i = 0; i < tokens.length; i += maxTokens - overlap) {
|
|
1107
1145
|
const chunkTokens = tokens.slice(i, i + maxTokens);
|
|
1108
1146
|
const decodedBytes = this.encoding.decode(chunkTokens);
|
|
1109
|
-
const
|
|
1147
|
+
const isFirst = i === 0;
|
|
1148
|
+
const isLast = i + chunkTokens.length >= tokens.length;
|
|
1149
|
+
const safeBytes = trimIncompleteUtf8(decodedBytes, !isFirst, !isLast);
|
|
1150
|
+
const chunkText = new TextDecoder('utf-8').decode(safeBytes);
|
|
1110
1151
|
chunks.push({
|
|
1111
1152
|
id: '',
|
|
1112
1153
|
document_id: '',
|
|
@@ -1364,7 +1405,7 @@ class RAGKnowledgeGraphManager {
|
|
|
1364
1405
|
INSERT OR IGNORE INTO chunk_entities (chunk_rowid, entity_id) VALUES (?, ?)
|
|
1365
1406
|
`);
|
|
1366
1407
|
for (const entityName of entityNames) {
|
|
1367
|
-
const entityId = `entity_${entityName.toLowerCase().replace(/[
|
|
1408
|
+
const entityId = `entity_${entityName.toLowerCase().replace(/[^\p{L}\p{N}]/gu, '_')}`;
|
|
1368
1409
|
// Verify entity exists
|
|
1369
1410
|
const entity = this.db.prepare(`
|
|
1370
1411
|
SELECT id FROM entities WHERE id = ?
|
|
@@ -2144,7 +2185,7 @@ class RAGKnowledgeGraphManager {
|
|
|
2144
2185
|
if (entityNames && entityNames.length > 0) {
|
|
2145
2186
|
const targetIds = new Set();
|
|
2146
2187
|
for (const name of entityNames) {
|
|
2147
|
-
const id = `entity_${name.toLowerCase().replace(/[
|
|
2188
|
+
const id = `entity_${name.toLowerCase().replace(/[^\p{L}\p{N}]/gu, '_')}`;
|
|
2148
2189
|
if (graph.hasNode(id))
|
|
2149
2190
|
targetIds.add(id);
|
|
2150
2191
|
}
|
|
Binary file
|
|
Binary file
|
|
Binary file
|