@cubememory/sdk

Official JavaScript/TypeScript SDK for Cube Memory — semantic search with 100% exact recall, isolated multi-tenant islands, and native MCP support.

Install

npm install @cubememory/sdk

Quick Start

import { createCubeClient } from '@cubememory/sdk';

const cube = createCubeClient('https://cubememory.com.br/gateway', 'cm_live_YOUR_KEY');

// Semantic search — 100% exact cosine recall (no ANN approximation)
const results = await cube.index('my-docs').search('wireless headphones', { limit: 10 });

// Index a document
await cube.index('my-docs').insert({
  id: 'doc-001',
  text: 'Noise cancelling Sony WH-1000XM5',
  metadata: { price: 349, brand: 'Sony' },
});

// LLM Memory (agent remembers between sessions)
await cube.memory.store('User prefers dark mode and speaks Portuguese');
const ctx = await cube.memory.search('user preferences');

// PKM Notes
await cube.notes.upsert({ title: 'Meeting Q3', text: 'Discussed the roadmap...', tags: ['meeting'] });
const notes = await cube.notes.search('roadmap');

CDN (no bundler)

<script src="https://cubememory.com.br/sdk.js"></script>
<script>
  const cube = CubeMemory.createCubeClient('https://cubememory.com.br/gateway', 'cm_live_YOUR_KEY');
  cube.index('docs').search('hello world').then(console.log);
</script>

API Reference

createCubeClient(baseUrl, apiKey, options?)

Returns a CubeClient instance.

| Param | Type | Description | |---|---|---| | baseUrl | string | Gateway URL, e.g. https://cubememory.com.br/gateway | | apiKey | string | Your API key (cm_live_xxx) | | options.projectId | string | Project ID (proj_xxx) |

cube.index(name)

| Method | Description | |---|---| | .insert({ id, text, metadata? }) | Index a document | | .insertMany(docs[]) | Batch index | | .search(query, { limit?, threshold?, filter? }) | Semantic search | | .get(id) | Fetch by ID | | .delete(id) | Remove document | | .stats() | Index statistics |

cube.memory

Agent memory — persists context between LLM sessions.

| Method | Description | |---|---| | .store(text, layer?) | Save a memory (long-term or short-term) | | .search(query, limit?) | Semantic search over memories | | .list(limit?) | List all memories | | .delete(id) | Delete a memory |

cube.notes

PKM Notes — Obsidian-style knowledge base with AI auto-connections.

| Method | Description | |---|---| | .upsert({ title, text, tags?, id? }) | Create or update a note | | .list() | List all notes | | .get(id) | Get note with full content and related notes | | .search(query, limit?) | Semantic search across notes | | .delete(id) | Delete note and its chunks |

MCP Integration

Connect any LLM (Claude, GPT, Cursor, Windsurf) directly to your Cube Memory:

{
  "mcpServers": {
    "cube-memory": {
      "type": "http",
      "url": "https://cubememory.com.br/gateway/v1/mcp",
      "headers": {
        "Authorization": "Bearer cm_live_YOUR_KEY",
        "X-Project-Id": "proj_YOUR_PROJECT"
      }
    }
  }
}

Available MCP tools: search_memory, store_memory, list_memories, search_notes, get_note.

Python SDK

pip install cubememory

from cubememory import CubeClient

cube = CubeClient('https://cubememory.com.br/gateway', 'cm_live_YOUR_KEY', 'proj_YOUR_PROJECT')
results = cube.index('my-docs').search('wireless headphones', limit=10)

REST API (raw requests)

pip install requests

import requests

H = {
    'Authorization': 'Bearer cm_live_YOUR_KEY',
    'X-Project-Id': 'proj_YOUR_PROJECT',
}

# save a memory
requests.post('https://cubememory.com.br/gateway/v1/memory',
    headers=H, json={'text': 'user prefers short answers'})

# recall before LLM call (semantic search)
ctx = requests.post('https://cubememory.com.br/gateway/v1/memory/search',
    headers=H, json={'query': user_input, 'limit': 5}).json()
memories = [r['text_preview'] for r in ctx['results']]

Why Cube Memory?

| | Cube Memory | Bancos vetoriais baseados em ANN/HNSW | |---|:---:|:---:| | 100% exact recall | ✅ cosine float32 | ❌ ANN approximation | | Multi-tenant isolation | ✅ island per project | ⚠️ shared index with filters | | MCP native | ✅ built-in | ❌ | | Self-host footprint | ✅ minimal, single-host friendly | ❌ heavy multi-service clusters | | LGPD/GDPR | ✅ data never leaves your server | ❌ cloud by default |

Links

• Dashboard: cubememory.com.br
• PyPI: pypi.org/project/cubememory
• Issues: github.com/cubememory/sdk