@oh-my-pi/pi-mnemopi 18.12.6 → 18.13.0

package/CHANGELOG.md

@@ -2,6 +2,12 @@
 
 ## [Unreleased]
 
+## [18.13.0] - 2026-10-26
+
+### Fixed
+
+- Fixed polyphonic recall's graph voice finding no memories for non-English names (e.g. `Hà Nội`) or names of three or more words, and Vietnamese words like `tôi` being stored as the participant `i` ([#16723](https://github.com/can1357/oh-my-pi/issues/16723))
+
 ## [18.12.4] - 2026-10-20
 
 ### Fixed

package/dist/types/core/episodic-graph.d.ts

@@ -55,6 +55,15 @@ export interface EpisodicGraphOptions {
     readonly db?: Database;
     readonly dbPath?: DatabasePath;
 }
+/**
+ * Capitalized proper nouns of at most two words, in any cased script (Latin, Greek,
+ * Cyrillic, Vietnamese, …), deduplicated case-insensitively. Gist participants are
+ * stored with this extractor and the polyphonic graph voice looks gists up with it, so
+ * a query entity matches a stored participant exactly (`findGistsByParticipant`).
+ * Unicode property escapes and `\p{L}`-based lookaround boundaries replace `[A-Z][a-z]+`
+ * and ASCII `\b`, which matched no non-English name (issues #7918, #16723).
+ */
+export declare function extractProperNouns(text: string): string[];
 export declare class EpisodicGraph {
     readonly db: Database;
     readonly dbPath: DatabasePath;

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-mnemopi",
-	"version": "18.12.6",
+	"version": "18.13.0",
 	"description": "Local SQLite memory engine for omp agents",
 	"homepage": "https://omp.sh",
 	"author": {
@@ -42,10 +42,10 @@
 		"fmt": "oxfmt --no-error-on-unmatched-pattern 'src/**/*.{ts,tsx}' '{test,bench,examples,scripts}/**/*.ts' '*.ts'"
 	},
 	"dependencies": {
-		"@oh-my-pi/pi-ai": "18.12.6",
-		"@oh-my-pi/pi-catalog": "18.12.6",
-		"@oh-my-pi/pi-natives": "18.12.6",
-		"@oh-my-pi/pi-utils": "18.12.6"
+		"@oh-my-pi/pi-ai": "18.13.0",
+		"@oh-my-pi/pi-catalog": "18.13.0",
+		"@oh-my-pi/pi-natives": "18.13.0",
+		"@oh-my-pi/pi-utils": "18.13.0"
 	},
 	"peerDependencies": {
 		"fastembed": "3.0.0",

package/src/core/episodic-graph.ts

@@ -134,6 +134,23 @@ function unique(values: Iterable<string>, limit = Number.MAX_SAFE_INTEGER): stri
 	return out;
 }
 
+/**
+ * Capitalized proper nouns of at most two words, in any cased script (Latin, Greek,
+ * Cyrillic, Vietnamese, …), deduplicated case-insensitively. Gist participants are
+ * stored with this extractor and the polyphonic graph voice looks gists up with it, so
+ * a query entity matches a stored participant exactly (`findGistsByParticipant`).
+ * Unicode property escapes and `\p{L}`-based lookaround boundaries replace `[A-Z][a-z]+`
+ * and ASCII `\b`, which matched no non-English name (issues #7918, #16723).
+ */
+export function extractProperNouns(text: string): string[] {
+	return unique(
+		Array.from(
+			text.matchAll(/(?<![\p{L}\p{N}_])(\p{Lu}\p{Ll}+(?:\s+\p{Lu}\p{Ll}+)?)(?![\p{L}\p{N}_])/gu),
+			match => match[1] ?? "",
+		),
+	);
+}
+
 function parseJsonStringArray(value: string | null): string[] {
 	if (value === null || value === "") return [];
 	try {
@@ -604,19 +621,13 @@ export class EpisodicGraph {
 	}
 
 	private extractParticipants(content: string): string[] {
-		// Match capitalized proper nouns across cased scripts (Latin, Greek, Cyrillic, …).
-		// `[A-Z][a-z]+` plus ASCII `\b` matched no non-Latin token, so any non-English
-		// memory produced zero participants (issue #7918). Unicode property escapes with
-		// the `u` flag, and `\p{L}`-based lookaround boundaries in place of `\b`, fix that.
-		const names = Array.from(
-			content.matchAll(/(?<![\p{L}\p{N}_])(\p{Lu}\p{Ll}+(?:\s+\p{Lu}\p{Ll}+)?)(?![\p{L}\p{N}_])/gu),
-			match => match[1] ?? "",
-		);
+		// `\p{L}` boundaries, not ASCII `\b`: with `i`, `\b` matched the trailing `i` of
+		// words like Vietnamese `tôi` as the pronoun `I` (issue #16723).
 		const pronouns = Array.from(
-			content.matchAll(/\b(I|you|we|they|he|she|it|me|us|them|him|her)\b/gi),
+			content.matchAll(/(?<![\p{L}\p{N}_])(I|you|we|they|he|she|it|me|us|them|him|her)(?![\p{L}\p{N}_])/giu),
 			match => match[1] ?? "",
 		);
-		return unique([...names, ...pronouns], 5);
+		return unique([...extractProperNouns(content), ...pronouns], 5);
 	}
 
 	private extractTemporalScope(content: string): string | null {

package/src/core/polyphonic-recall.ts

@@ -4,7 +4,7 @@ import { type Env, polyphonicRecallEnabled } from "../config";
 import { closeQuietly, type DatabasePath, openDatabase } from "../db";
 import { backfillConsolidatedFacts, ensureVeracityConsolidator } from "./beam/consolidate";
 import type { BeamMemoryState, JsonValue, Metadata, RecallResult, RecallTierLabel } from "./beam/types";
-import { EpisodicGraph } from "./episodic-graph";
+import { EpisodicGraph, extractProperNouns } from "./episodic-graph";
 import { STOP_WORDS } from "./synonyms";
 import { VeracityConsolidator } from "./veracity-consolidation";
 
@@ -214,16 +214,6 @@ function cosineAgainstUnit(unit: Float32Array, raw: unknown): number | null {
 	return dot / Math.sqrt(normSq);
 }
 
-function extractEntities(text: string): string[] {
-	const seen = new Set<string>();
-	const matches = text.matchAll(/\b[A-Z][a-z]+(?:\s+[A-Z][a-z]+)*\b/g);
-	for (const match of matches) {
-		const entity = match[0];
-		if (entity.length > 0) seen.add(entity);
-	}
-	return [...seen];
-}
-
 function queryTokens(query: string): string[] {
 	const tokens: string[] = [];
 	for (const match of query.toLowerCase().matchAll(/[\p{L}\p{N}_-]+/gu)) tokens.push(match[0]);
@@ -420,7 +410,7 @@ export class PolyphonicRecallEngine {
 			const existing = seeds.get(hit.memoryId);
 			if (existing === undefined || hit.score > existing.score) seeds.set(hit.memoryId, hit);
 		};
-		for (const entity of extractEntities(query)) {
+		for (const entity of extractProperNouns(query)) {
 			for (const gist of this.graph.findGistsByParticipant(entity)) {
 				const memoryId = gist.id.startsWith("gist_") ? gist.id.slice(5) : gist.id;
 				addSeed({ memoryId, score: 0.6, voice: "graph", metadata: { entity, gist: gist.text } });