New memory system
This commit is contained in:
515
src/memory.ts
515
src/memory.ts
@@ -1,177 +1,420 @@
|
||||
// memory.ts
|
||||
import {LLMRequest, LLMMessage} from './llm.ts';
|
||||
import {AiTool} from './tools.ts';
|
||||
import {KDTree, KDPoint} from './kd-tree.ts';
|
||||
|
||||
/** Background information the AI will be fed as a knowledge document */
|
||||
export type Memory = {
|
||||
/** Memory subject */
|
||||
name: string;
|
||||
/** Short description of what this document contains - used for RAG retrieval */
|
||||
description: string;
|
||||
/** Full markdown content of the document */
|
||||
content: string;
|
||||
/** Embedding vector of the description - used for similarity search */
|
||||
embedding: number[];
|
||||
links: string[];
|
||||
backlinks: string[];
|
||||
}
|
||||
|
||||
export type MemoryCollection = {
|
||||
/** Memory subject */
|
||||
type MemoryRef = {
|
||||
name: string;
|
||||
/** Short description - required if isNew */
|
||||
description?: string;
|
||||
/** Extracted facts to merge */
|
||||
description: string;
|
||||
}
|
||||
|
||||
type FactBucket = {
|
||||
subject: string;
|
||||
facts: string[];
|
||||
}
|
||||
// In memory.ts - replace findGhostNodes with this:
|
||||
|
||||
export type MemoryNode = {
|
||||
name: string;
|
||||
missing: boolean;
|
||||
links: string[];
|
||||
backlinks: string[];
|
||||
}
|
||||
|
||||
export function buildMemoryGraph(memories: Memory[] | MemoryCache): MemoryNode[] {
|
||||
const mems = memories instanceof MemoryCache ? memories.memories : memories;
|
||||
const nameSet = new Set(mems.map(m => m.name));
|
||||
const ghosts = new Set<string>();
|
||||
|
||||
// Collect all ghost references
|
||||
for (const m of mems) {
|
||||
for (const link of m.links) {
|
||||
if (!nameSet.has(link)) ghosts.add(link);
|
||||
}
|
||||
}
|
||||
|
||||
// Build node list: real nodes + ghost nodes
|
||||
return [
|
||||
...mems.map(m => ({
|
||||
name: m.name,
|
||||
missing: false,
|
||||
links: m.links,
|
||||
backlinks: m.backlinks,
|
||||
})),
|
||||
...[...ghosts].map(name => ({
|
||||
name,
|
||||
missing: true,
|
||||
links: [],
|
||||
backlinks: mems
|
||||
.filter(m => m.links.includes(name))
|
||||
.map(m => m.name),
|
||||
}))
|
||||
];
|
||||
}
|
||||
|
||||
function extractLinks(content: string): string[] {
|
||||
const matches = content.matchAll(/\[\[([^\]]+)\]\]/g);
|
||||
return [...new Set([...matches].map(m => m[1].trim()))];
|
||||
}
|
||||
|
||||
function rebuildBacklinks(memories: Memory[]): void {
|
||||
for (const m of memories) m.backlinks = [];
|
||||
for (const m of memories) {
|
||||
for (const link of m.links) {
|
||||
const target = memories.find(t => t.name === link);
|
||||
if (target) target.backlinks.push(m.name);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
function cosineDistance(a: number[], b: number[]): number {
|
||||
let dot = 0, normA = 0, normB = 0;
|
||||
for (let i = 0; i < a.length; i++) {
|
||||
dot += a[i] * b[i];
|
||||
normA += a[i] * a[i];
|
||||
normB += b[i] * b[i];
|
||||
}
|
||||
const denom = Math.sqrt(normA) * Math.sqrt(normB);
|
||||
return denom === 0 ? 1 : 1 - dot / denom;
|
||||
}
|
||||
|
||||
export class MemoryCache {
|
||||
private tree: KDTree<MemoryRef>;
|
||||
public memories: Memory[];
|
||||
|
||||
constructor(memories: Memory[]) {
|
||||
this.memories = memories;
|
||||
this.tree = this.buildTree();
|
||||
}
|
||||
|
||||
private buildTree(): KDTree<MemoryRef> {
|
||||
const embedded = this.memories.filter(m => m.embedding?.length);
|
||||
if(!embedded.length) return new KDTree<MemoryRef>(0);
|
||||
|
||||
const dims = embedded[0].embedding.length;
|
||||
const points: KDPoint<MemoryRef>[] = embedded.map(m => ({
|
||||
vector: m.embedding,
|
||||
payload: {name: m.name, description: m.description},
|
||||
}));
|
||||
|
||||
return new KDTree<MemoryRef>(dims, 'cosine', points);
|
||||
}
|
||||
|
||||
search(query: number[], limit: number): MemoryRef[] {
|
||||
const results = this.tree.knn(query, limit);
|
||||
return results.map(r => r.point.payload);
|
||||
}
|
||||
|
||||
add(memory: Memory): void {
|
||||
this.memories.push(memory);
|
||||
this.rebuild();
|
||||
}
|
||||
|
||||
update(memory: Memory): void {
|
||||
const idx = this.memories.findIndex(m => m.name === memory.name);
|
||||
if (idx !== -1) {
|
||||
this.memories[idx] = memory;
|
||||
} else {
|
||||
this.memories.push(memory);
|
||||
}
|
||||
this.rebuild();
|
||||
}
|
||||
|
||||
rebuild(): void {
|
||||
this.tree = this.buildTree();
|
||||
}
|
||||
|
||||
rebuildLinks(): void {
|
||||
rebuildBacklinks(this.memories);
|
||||
}
|
||||
}
|
||||
|
||||
export class MemoryManager {
|
||||
|
||||
tools = {
|
||||
edit: (memory: Memory): AiTool => ({
|
||||
name: 'edit_memory',
|
||||
description: 'Edit a memory. Omit start/end to append. Pass start only to replace from that line on (Note line 0 = first line of content / line AFTER description). Pass start+end to replace a specific range. start=0 replaces the whole document. Returns updated document',
|
||||
args: {
|
||||
content: {type: 'string', description: 'New content', required: true},
|
||||
start: {type: 'number', description: 'First line to replace (0-indexed, inclusive). Omit to append.'},
|
||||
end: {type: 'number', description: 'Last line to replace (0-indexed, inclusive). Omit to replace from start to end of doc.'},
|
||||
},
|
||||
fn: (args: any) => {
|
||||
const lines = memory.content ? memory.content.split('\n') : [];
|
||||
const newLines = args.content.split('\n');
|
||||
if(args.start === undefined) lines.push(...newLines);
|
||||
else if(args.end === undefined) lines.splice(args.start, lines.length - args.start, ...newLines);
|
||||
else lines.splice(args.start, args.end - args.start + 1, ...newLines);
|
||||
memory.content = lines.join('\n');
|
||||
return memory.content;
|
||||
}
|
||||
}),
|
||||
extract: (pools: MemoryCollection[]): AiTool => ({
|
||||
name: 'extract_facts',
|
||||
description: 'Extract a list of facts to group into a single memory',
|
||||
args: {
|
||||
name: {type: 'string', description: 'Exact name of an existing memory, or a new name if none fits ([pro]nouns only)', required: true},
|
||||
description: {type: 'string', description: 'One sentence description of the memory subject', required: true},
|
||||
facts: {type: 'string', description: 'Comma separated list of extracted facts', required: true},
|
||||
},
|
||||
fn: (args: any) => {
|
||||
pools.push({
|
||||
name: args.name,
|
||||
description: args.description,
|
||||
facts: args.facts.split(',').map((f: string) => f.trim()).filter(Boolean),
|
||||
});
|
||||
return 'Success';
|
||||
}}),
|
||||
read: (memories: Memory[]): AiTool => ({
|
||||
read: (memories: Memory[] | MemoryCache): AiTool => ({
|
||||
name: 'read_memory',
|
||||
description: 'Read entire memory',
|
||||
description: 'Read the full content of a memory document',
|
||||
args: {
|
||||
name: {type: 'string', description: 'Exact memory name', required: true},
|
||||
},
|
||||
fn: (args: any) => {
|
||||
const mem = memories.find(m => m.name === args.name);
|
||||
fn:(args: any) => {
|
||||
const mems = memories instanceof MemoryCache ? memories.memories : memories;
|
||||
const mem = mems.find(m => m.name === args.name);
|
||||
if(!mem) return 'Document not found';
|
||||
return `Name: ${mem.name}\nDescription: ${mem.description}\n\n${mem.content}`;
|
||||
return this.formatMemory(mem);
|
||||
}
|
||||
}),
|
||||
};
|
||||
|
||||
constructor(private llm: any) {}
|
||||
|
||||
private cosineSearch(query: number[], memories: Memory[], limit: number): MemoryRef[] {
|
||||
const scored = memories
|
||||
.filter(m => m.embedding?.length)
|
||||
.map(m => ({
|
||||
ref: {name: m.name, description: m.description},
|
||||
distance: cosineDistance(query, m.embedding)
|
||||
}))
|
||||
.sort((a, b) => a.distance - b.distance)
|
||||
.slice(0, limit);
|
||||
return scored.map(s => s.ref);
|
||||
}
|
||||
|
||||
constructor(private llm: any, private model?: string) {}
|
||||
|
||||
/**
|
||||
* Extracts facts from conversation and groups them into individual memories
|
||||
* @param {string} conversation Full conversation formatted as [role]: content
|
||||
* @param {Memory[]} memories The user's memory documents
|
||||
* @param {LLMRequest} options LLM options
|
||||
* @returns {Promise<MemoryCollection[]>} Fact pools grouped by target document
|
||||
*/
|
||||
private async extract(conversation: string, memories: Memory[], options: LLMRequest): Promise<MemoryCollection[]> {
|
||||
const existingDocs = memories.map(m => `Name: ${m.name}\nDescription: ${m.description}`).join('\n\n');
|
||||
const pools: MemoryCollection[] = [];
|
||||
await this.llm.ask(conversation, {
|
||||
model: this.model || options.model,
|
||||
temperature: 0.2,
|
||||
system: `You are a fact extractor. Analyze this conversation and extract facts worth remembering long term.
|
||||
Rules:
|
||||
- ONLY extract facts the USER explicitly stated about themselves or their business
|
||||
- ONLY extract decisions that were MADE during this conversation
|
||||
- DO NOT extract anything the AI said, its name, capabilities, or how it introduced itself
|
||||
- DO NOT extract greetings, pleasantries or generic exchanges
|
||||
- If nothing worth remembering was said, dont do anything, skip calling tools
|
||||
|
||||
For each fact decide whether it belongs in an existing document or needs a new one, then call the \`extract_facts\` tool.
|
||||
|
||||
Existing documents:\n${existingDocs || 'None yet.'}`,
|
||||
tools: [this.tools.extract(pools)]
|
||||
});
|
||||
return pools;
|
||||
private createNode(name: string, memories: Memory[]): Memory {
|
||||
const existing = memories.find(m => m.name === name);
|
||||
if(existing) return existing;
|
||||
return {
|
||||
name,
|
||||
description: '',
|
||||
content: '',
|
||||
embedding: [],
|
||||
links: [],
|
||||
backlinks: [],
|
||||
};
|
||||
}
|
||||
|
||||
/**
|
||||
* Bot 2 - Editor: merges a pool of facts into a specific document using surgical line-based edits.
|
||||
* Receives full document content and uses read + amend tools to make precise edits.
|
||||
* @param {MemoryCollection} newMem The fact pool to merge
|
||||
* @param {Memory[]} memories The user's memory documents
|
||||
* @param {LLMRequest} options LLM options
|
||||
*/
|
||||
private async edit(newMem: MemoryCollection, memories: Memory[], options: LLMRequest): Promise<void> {
|
||||
const existing = memories.find(m => m.name === newMem.name);
|
||||
const mem: Memory = existing || {name: newMem.name, description: newMem.description || '', content: '', embedding: []};
|
||||
const isNew = !existing;
|
||||
private formatMemory(mem: Memory): string {
|
||||
return [
|
||||
`# ${mem.name}`,
|
||||
mem.description ? `> ${mem.description}` : '',
|
||||
mem.links.length ? `**Links:** ${mem.links.map(l => `[[${l}]]`).join(', ')}` : '',
|
||||
mem.backlinks.length ? `**Referenced by:** ${mem.backlinks.map(l => `[[${l}]]`).join(', ')}` : '',
|
||||
'',
|
||||
mem.content,
|
||||
].filter(l => l !== undefined).join('\n');
|
||||
}
|
||||
|
||||
await this.llm.ask(newMem.facts.map(f => `- ${f}`).join('\n'),
|
||||
private listNodes(memories: Memory[]): MemoryRef[] {
|
||||
return memories.map(m => ({name: m.name, description: m.description}));
|
||||
}
|
||||
|
||||
async recollect(query: string, memories: Memory[] | MemoryCache, limit = 5, graphDepth = 1): Promise<Memory[]> {
|
||||
const mem: Memory[] = memories instanceof MemoryCache ? memories.memories : memories;
|
||||
if(!mem.length) return [];
|
||||
const [e] = await this.llm.embedding(query);
|
||||
if(!e) return [];
|
||||
|
||||
let vectorResults: MemoryRef[];
|
||||
if(memories instanceof MemoryCache) vectorResults = memories.search(e.embedding, limit);
|
||||
else vectorResults = this.cosineSearch(e.embedding, mem, limit);
|
||||
const found = new Set<string>(vectorResults.map(r => r.name));
|
||||
|
||||
// Graph expansion
|
||||
if(graphDepth > 0) {
|
||||
const frontier = [...found];
|
||||
for(let depth = 0; depth < graphDepth; depth++) {
|
||||
const next: string[] = [];
|
||||
for(const name of frontier) {
|
||||
const node = mem.find(m => m.name === name);
|
||||
if(!node) continue;
|
||||
for(const link of node.links) {
|
||||
if(!found.has(link) && mem.find(m => m.name === link)) {
|
||||
found.add(link);
|
||||
next.push(link);
|
||||
}
|
||||
}
|
||||
}
|
||||
frontier.splice(0, frontier.length, ...next);
|
||||
if(!frontier.length) break;
|
||||
}
|
||||
}
|
||||
|
||||
const vectorOrder = vectorResults.map(r => r.name);
|
||||
const graphExpansions = [...found].filter(n => !vectorOrder.includes(n));
|
||||
const ordered = [...vectorOrder, ...graphExpansions];
|
||||
return ordered.map(n => mem.find(m => m.name === n)!).filter(Boolean);
|
||||
}
|
||||
|
||||
async memorize(history: LLMMessage[], memories: Memory[] | MemoryCache, options: LLMRequest): Promise<void> {
|
||||
const mem = memories instanceof MemoryCache ? memories.memories : memories;
|
||||
const conversation = history
|
||||
.filter(h => h.role === 'user' || h.role === 'assistant')
|
||||
.map(h => `[${h.role}]: ${h.content}`).join('\n\n').trim();
|
||||
if(!conversation) return;
|
||||
|
||||
const buckets = await this.factAgent(conversation, mem, options);
|
||||
if(!buckets.length) return;
|
||||
await Promise.all(buckets.map(async bucket => {
|
||||
const node = await this.organizingAgent(bucket, mem, options);
|
||||
if(!mem.find(m => m.name === node.name)) mem.push(node);
|
||||
await this.docAgent(node, bucket, mem, options);
|
||||
}));
|
||||
|
||||
// Rebuild indexes
|
||||
if (memories instanceof MemoryCache) {
|
||||
memories.rebuildLinks();
|
||||
memories.rebuild();
|
||||
} else {
|
||||
rebuildBacklinks(mem);
|
||||
}
|
||||
}
|
||||
|
||||
private async docAgent(node: Memory, bucket: FactBucket, memories: Memory[], options: LLMRequest): Promise<void> {
|
||||
let finalContent = node.content;
|
||||
|
||||
await this.llm.ask(
|
||||
`New facts to integrate:\n${bucket.facts.map(f => `- ${f}`).join('\n')}`,
|
||||
{
|
||||
model: this.model || options.model,
|
||||
temperature: 0.2,
|
||||
system: `You are a document editor. Merge the users list of facts into the following document using the \`edit_memory\` tool; call it as many times as necessary:
|
||||
\`\`\`
|
||||
${mem.content}
|
||||
model: options.model,
|
||||
temperature: 0.3,
|
||||
system: `You are a knowledge base editor. Integrate the provided facts into the document below.
|
||||
|
||||
Formatting rules:
|
||||
- Use Obsidian-style markdown: # headings, **bold** for key terms, bullet lists for facts
|
||||
- Link related concepts with [[WikiLink]] notation — only link things that are genuinely related
|
||||
- You may create links to nodes that don't exist yet if the concept is important
|
||||
- Keep the document concise, factual, and human-readable
|
||||
- Resolve any contradictions between old content and new facts (new facts win)
|
||||
- Do not add filler, preamble, or AI commentary — just clean knowledge documents
|
||||
|
||||
All nodes:
|
||||
${this.listNodes(memories).map(n => n.name).join(', ') || 'none'}
|
||||
|
||||
Current document:
|
||||
\`\`\`markdown
|
||||
${node.content || '(empty — this is a new document)'}
|
||||
\`\`\``,
|
||||
tools: [this.tools.edit(mem)]
|
||||
tools: [{
|
||||
name: 'update_document',
|
||||
description: 'Write the complete updated document content',
|
||||
args: {
|
||||
description: {type: 'string', description: 'One-line description of what this document covers, no formatting or emojis', required: true},
|
||||
content: {type: 'string', description: 'Fully updated document in markdown', required: true},
|
||||
},
|
||||
fn:(args: any) => {
|
||||
node.description = args.description;
|
||||
finalContent = args.content;
|
||||
return 'Saved';
|
||||
}
|
||||
}]
|
||||
}
|
||||
);
|
||||
|
||||
if(isNew || mem.description !== existing?.description) {
|
||||
const e = await this.llm.embedding(mem.description);
|
||||
mem.embedding = e?.[0]?.embedding;
|
||||
}
|
||||
|
||||
if(isNew) memories.push(mem);
|
||||
else {
|
||||
const idx = memories.findIndex(m => m.name === newMem.name);
|
||||
if(idx >= 0) memories[idx] = mem;
|
||||
node.content = finalContent;
|
||||
node.links = extractLinks(finalContent);
|
||||
const needsEmbed = !node.embedding?.length || node.description !== memories.find(m => m.name === node.name)?.description;
|
||||
if (needsEmbed) {
|
||||
const [e] = await this.llm.embedding(node.description);
|
||||
if (e) node.embedding = e.embedding;
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Find relevant memory documents for a query using description embeddings
|
||||
* @param {string} query The query to search against
|
||||
* @param {Memory[]} memories The user's memory documents
|
||||
* @param {number} limit Max number of results to return
|
||||
* @returns {Promise<Memory[]>} The most relevant memory documents
|
||||
*/
|
||||
async recollect(query: string, memories: Memory[], limit = 5): Promise<Memory[]> {
|
||||
const [e] = await this.llm.embedding(query);
|
||||
return memories
|
||||
.filter(m => m.embedding?.length)
|
||||
.map(m => ({...m, score: this.llm.cosineSimilarity(m.embedding, e.embedding)}))
|
||||
.toSorted((a: any, b: any) => b.score - a.score)
|
||||
.slice(0, limit);
|
||||
private async factAgent(conversation: string, memories: Memory[], options: LLMRequest): Promise<FactBucket[]> {
|
||||
const buckets: FactBucket[] = [];
|
||||
|
||||
await this.llm.ask(conversation, {
|
||||
model: options.model,
|
||||
temperature: 0.2,
|
||||
system: `You are a fact extractor. Analyze this conversation and extract facts worth remembering long-term.
|
||||
|
||||
Rules:
|
||||
- ONLY extract facts the USER explicitly stated about themselves, their work, or their projects
|
||||
- ONLY extract decisions that were MADE during this conversation
|
||||
- DO NOT extract anything the AI said, its capabilities, or meta-conversation about the AI
|
||||
- DO NOT extract greetings, pleasantries, or generic exchanges
|
||||
- If nothing worth remembering was said, do not call any tools
|
||||
|
||||
Group facts by subject. For each group call \`extract_facts\` once.
|
||||
|
||||
Known nodes (name: description):
|
||||
${this.listNodes(memories).map(n => `- ${n.name}: ${n.description}`).join('\n') || 'None yet.'}`,
|
||||
tools: [{
|
||||
name: 'extract_facts',
|
||||
description: 'Submit a group of related facts for a specific subject',
|
||||
args: {
|
||||
subject: {type: 'string', description: 'Subject matter facts regard', required: true},
|
||||
facts: {type: 'string', description: 'Comma-separated list of extracted facts', required: true},
|
||||
},
|
||||
fn: (args: any) => {
|
||||
buckets.push({
|
||||
subject: args.subject,
|
||||
facts: args.facts.split(',').map((f: string) => f.trim()).filter(Boolean),
|
||||
});
|
||||
return 'Recorded';
|
||||
}
|
||||
}]
|
||||
});
|
||||
|
||||
return buckets;
|
||||
}
|
||||
|
||||
/**
|
||||
* Two-stage memory pipeline: classify facts from conversation history then surgically merge them into documents.
|
||||
* Bot 1 (classify) extracts and groups facts cheaply. Bot 2 (edit) runs per-document in parallel with full content access.
|
||||
* @param {LLMMessage[]} history Full conversation history to digest
|
||||
* @param {Memory[]} memories The user's memory documents — mutated in place
|
||||
* @param {LLMRequest} options LLM options
|
||||
*/
|
||||
async memorize(history: LLMMessage[], memories: Memory[], options: LLMRequest): Promise<void> {
|
||||
const conversation = history
|
||||
.filter(h => h.role === 'user' || h.role === 'assistant')
|
||||
.map(h => `[${h.role}]: ${h.content}`)
|
||||
.join('\n\n');
|
||||
if(!conversation.trim()) return;
|
||||
const pools = await this.extract(conversation, memories, options);
|
||||
if(!pools.length) return;
|
||||
await Promise.all(pools.map(pool => this.edit(pool, memories, options)));
|
||||
private async organizingAgent(bucket: FactBucket, memories: Memory[], options: LLMRequest): Promise<Memory> {
|
||||
let candidates = this.listNodes(memories);
|
||||
let attempts = 0;
|
||||
const maxAttempts = 3;
|
||||
|
||||
while (attempts++ < maxAttempts) {
|
||||
let home = '', mode: string | null = null;
|
||||
|
||||
const resp = await this.llm.ask(`Subject: ${bucket.subject}\n\nFacts:\n${bucket.facts.map(f => `- ${f}`).join('\n')}`, {
|
||||
model: options.model,
|
||||
temperature: 0.1,
|
||||
system: `You are a knowledge organizer. Your job is to find the correct home for the supplied facts.
|
||||
|
||||
1. Review the facts and the node list below. Pick the most likely match or decide if a new node is needed.
|
||||
2. If you picked an existing node, use \`read\` to verify it's the right place.
|
||||
- After reading, call either \`confirm\` (correct node) or \`mismatched\` (wrong node).
|
||||
3. If none of the nodes match, call \`create\` to make a new node.
|
||||
|
||||
Available nodes:
|
||||
${candidates.map(n => `- ${n.name}: ${n.description}`).join('\n') || 'None — create a new node.'}`,
|
||||
tools: [{
|
||||
name: 'read',
|
||||
description: 'Read a node file to verify it is the right home for these facts',
|
||||
args: {name: {type: 'string', description: 'Exact node name', required: true}},
|
||||
fn: ({name}) => {
|
||||
const mem = memories.find(m => m.name === name);
|
||||
if (!mem) return 'Node not found';
|
||||
home = name;
|
||||
return this.formatMemory(mem);
|
||||
}
|
||||
}, {
|
||||
name: 'confirm',
|
||||
description: 'Confirm this is the correct node for the facts',
|
||||
args: {},
|
||||
fn: () => {
|
||||
mode = 'success';
|
||||
resp.abort();
|
||||
}
|
||||
}, {
|
||||
name: 'mismatched',
|
||||
description: 'This is not the node you are looking for',
|
||||
args: {},
|
||||
fn: () => {
|
||||
mode = 'failed';
|
||||
resp.abort();
|
||||
}
|
||||
}, {
|
||||
name: 'create',
|
||||
description: 'No existing node fits — create a new one',
|
||||
args: {name: {type: 'string', description: 'Canonical name for the new node', required: true}},
|
||||
fn: ({name}) => {
|
||||
home = name;
|
||||
mode = 'create';
|
||||
resp.abort();
|
||||
}
|
||||
}]
|
||||
});
|
||||
|
||||
if(mode === 'create') {
|
||||
return this.createNode(home, memories);
|
||||
} else if (mode === 'failed') {
|
||||
candidates = candidates.filter(c => c.name !== home);
|
||||
if(!candidates.length) return this.createNode(bucket.subject, memories);
|
||||
} else if (mode === 'success') {
|
||||
const existing = memories.find(m => m.name === home);
|
||||
return existing || this.createNode(home, memories);
|
||||
}
|
||||
}
|
||||
return this.createNode(bucket.subject, memories);
|
||||
}
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user