DeltaMemory
The Cognitive Memory Layer for AI Applications
DeltaMemory provides production-grade long-term memory infrastructure for AI agents, assistants, and applications. Unlike simple vector stores, DeltaMemory implements a cognitive architecture that mirrors human memory systems—automatically extracting structured knowledge, maintaining temporal context, and enabling multi-hop reasoning across your data.
Why DeltaMemory?
Modern AI applications require more than keyword search or basic RAG. They need memory that:
- Understands context — Extracts structured profiles, events, and relationships automatically
- Reasons across knowledge — Multi-hop graph traversal for complex queries
- Prioritizes relevance — Cognitive scoring combines similarity, recency, and salience
- Scales efficiently — Scalable architecture for production workloads
Key Features
Cognitive Processing
Every piece of content ingested through DeltaMemory undergoes intelligent extraction:
- User Profiles — Structured facts organized by topic (preferences, work, interests)
- Event Timeline — Activities, plans, and milestones with temporal awareness
- Knowledge Graph — Concepts and relationships for multi-hop reasoning
- Fact Extraction — Discrete, queryable facts with confidence scores
Hybrid Recall
DeltaMemory's recall system combines multiple signals for optimal retrieval:
cognitive_score = (similarity × w₁) + (recency × w₂) + (salience × w₃)- Semantic Similarity — Vector-based relevance matching
- Temporal Recency — Recent memories weighted appropriately
- Salience Scoring — Important memories surface naturally
Pre-formatted Context
Every recall returns a ready-to-use context string optimized for LLM consumption:
const { context } = await db.recall('user preferences');
// Returns formatted markdown with profiles, events, and relevant memoriesEnterprise-Grade Infrastructure
DeltaMemory is built from the ground up for production workloads:
- High Performance — Optimized storage engine designed for sub-millisecond recall
- Scalable — Handles millions of memories with consistent performance
- Reliable — Production-tested with enterprise customers
- Secure — Self-hosted deployment with full data control
Quick Example
import { DeltaMemory } from 'deltamemory';
const db = new DeltaMemory({
apiKey: process.env.DELTAMEMORY_API_KEY,
defaultCollection: 'my-agent'
});
// Ingest with automatic cognitive processing
await db.ingest('I work as a senior engineer at Acme Corp. I prefer TypeScript.');
// Recall with structured context
const { profiles, events, context } = await db.recall('user background');
console.log(profiles);
// [
// { topic: 'work', sub_topic: 'title', content: 'senior engineer' },
// { topic: 'work', sub_topic: 'company', content: 'Acme Corp' },
// { topic: 'interest', sub_topic: 'language', content: 'TypeScript' }
// ]Getting Started
Choose your preferred SDK to begin:
Integration Ecosystem
DeltaMemory integrates seamlessly with popular AI frameworks:
Licensing
DeltaMemory is available as a managed service at app.deltamemory.com (opens in a new tab). Sign up to get your API key and endpoint.
Getting Started
- Sign up at app.deltamemory.com (opens in a new tab)
- Create a project and get your API key
- Install the SDK and start building
import { DeltaMemory } from 'deltamemory';
const db = new DeltaMemory({
apiKey: process.env.DELTAMEMORY_API_KEY, // From dashboard
baseUrl: process.env.DELTAMEMORY_URL, // Your assigned endpoint
});Multi-Tenant Architecture
DeltaMemory is designed for multi-tenant SaaS applications. Each API key is bound to a specific tenant, providing complete data isolation:
// Each API key routes to an isolated tenant
const tenant1 = new DeltaMemory({ apiKey: 'dm_tenant1_key' });
const tenant2 = new DeltaMemory({ apiKey: 'dm_tenant2_key' });
// Data is completely isolated between tenants
await tenant1.ingest('Tenant 1 data'); // Stored in tenant1's isolated storage
await tenant2.ingest('Tenant 2 data'); // Stored in tenant2's isolated storageKey features:
- API key determines tenant routing automatically
- Complete storage isolation (WAL, SSTables, vector index, graph)
- No cross-tenant data leakage
- Per-tenant rate limiting and quotas
For multi-user isolation within a single tenant, use collections:
const db = new DeltaMemory({ apiKey: process.env.DELTAMEMORY_API_KEY });
// Isolate users within your tenant using collections
await db.ingest('User 1 preference', { collection: 'user-123' });
await db.ingest('User 2 preference', { collection: 'user-456' });