An SDK for contextual memory

Deeply-indexed graph memory for the agents you're building. Index everything once, then recall grounded, cited context in a single call — so the agents you ship think with real knowledge, not flat embeddings.

import { Constella } from '@constella/sdk'

const constella = new Constella({ apiKey: process.env.CONSTELLA_API_KEY })

// Recall grounded context from your graph memory
const { context, sources } = await constella.recall({
  query: 'surface-code thresholds',
  limit: 8,
})

agent.system(context)  // grounded in `sources`

Index once → one recall() → grounded, cited context your agent can act on.

• Graph memoryNotes, papers, and chats indexed into a connected graph — not flat embeddings.
• One recall callSemantic + graph search returns grounded context with citations.
• Bring your sourcesSlack, Notion, Drive, Gmail, GitHub, Obsidian and more feed the same graph.
• Drop into any agentUse the SDK, or reach the same memory over MCP at fastfind.app/mcp.