Memory Operations

The SDK provides full CRUD operations for memories, conversations, facts, topics, and entities. Every example below assumes an initialized client:

import { Memnexus } from "@memnexus-ai/typescript-sdk";

const client = new Memnexus({
  token: process.env.MX_API_KEY,
});

Service methods return an HttpResponse wrapper — the payload is at response.data?.data.

Creating memories

Every memory belongs to a conversation, so conversationId is required. Create the conversation first, then attach memories to it.

// Create a new conversation
const convResponse = await client.conversations.createConversation({
  title: "API Rate Limiting Implementation",
});
const conversationId = convResponse.data?.data?.id;
if (!conversationId) throw new Error("conversation creation failed");

// Create a memory in that conversation
const response = await client.memories.createMemory({
  conversationId,
  content: "Starting rate limiting feature. Will use token bucket algorithm.",
  topics: ["rate-limiting", "architecture"],
});

console.log("Created:", response.data?.data?.id);

// Add more memories to the same conversation
await client.memories.createMemory({
  conversationId,
  content: "Rate limiting middleware complete. 100 req/min default.",
  topics: ["rate-limiting", "implementation"],
});

Retrieving memories

Get a single memory

const response = await client.memories.getMemoryById("mem_abc123");
console.log(response.data?.data?.content);
console.log(response.data?.data?.topics);

List memories

const response = await client.memories.listMemories({
  page: 0,
  limit: 20,
});

for (const memory of response.data?.data ?? []) {
  console.log(memory.content);
}

Batch get

const response = await client.memories.getMemoriesBatch({
  ids: ["mem_abc123", "mem_def456", "mem_ghi789"],
});

Updating memories

await client.memories.updateMemory("mem_abc123", {
  content: "Updated: Deployed v1.26.3 with additional fixes.",
});

Deleting memories

await client.memories.deleteMemory("mem_abc123");

Working with facts

Facts are structured knowledge as subject-predicate-object triples:

// Create a fact
const response = await client.facts.createFact({
  subject: "MemNexus",
  predicate: "built_with",
  object: "Node.js",
});

// Search facts
const facts = await client.facts.searchFacts({
  query: "Node.js",
});

// List facts
const allFacts = await client.facts.listFacts({ limit: 50, offset: 0 });

Working with conversations

// List recent conversations
const conversations = await client.conversations.listConversations({
  limit: 10,
});

// Get conversation details
const conv = await client.conversations.getConversationSummary("conv_abc123");

// Get conversation timeline
const timeline = await client.conversations.getConversationTimeline("conv_abc123");

Working with topics

// List topics
const topics = await client.topics.listTopics();

// Discover related topics
const related = await client.topics.discoverRelatedTopics({
  topicId: "topic_abc123",
});

// Cluster topics
const clusters = await client.topics.clusterTopics({
  body: { minClusterSize: 5 },
});

Pagination pattern

All list operations return paginated results:

async function getAllMemories() {
  const memories = [];
  let page = 0;
  let hasMore = true;

  while (hasMore) {
    const response = await client.memories.listMemories({ page, limit: 100 });
    const batch = response.data?.data ?? [];
    memories.push(...batch);
    hasMore = batch.length === 100;
    page++;
  }

  return memories;
}

Next steps

• Search — Semantic and hybrid search
• Core Concepts — Understand the data model