The Ultimate Local-First Agentic Memory Framework for Apple Silicon.
Uniting the best architectural innovations of Mem0, Supermemory, Letta/MemGPT, and Zep into a unified, lightweight native Swift library. 100% on-device execution with zero cloud hosting costs, seamlessly synced across user Apple devices via Apple CloudKit private database.
import SynapseMemory
// 1. Initialize SynapseClient (Defaults to Apple Foundation Models + CloudKit Sync)
let synapse = try await SynapseClient(config: SynapseConfig())
// 2. Add conversational turns (Extracts memories, knowledge graph triples & recall logs)
let messages = [
Message(role: .user, content: "Hi! I am Alex, I live in Bangkok, and I build native Swift apps."),
Message(role: .assistant, content: "Great to meet you Alex! I will remember you live in Bangkok.")
]
try await synapse.add(messages: messages, userId: "alex_123")
// 3. Ingest documents and web bookmarks with auto-tagging (Supermemory feature)
try await synapse.ingest(
content: "Apple Intelligence combines generative models with on-device personal context...",
title: "Apple Intelligence Notes",
userId: "alex_123",
tags: ["apple", "ai", "privacy"]
)
// 4. Hybrid Search (Accelerate SIMD + SQLite FTS5 + Time-Decay)
let results = try await synapse.search(query: "Where does Alex live?", userId: "alex_123")
for item in results {
print("Found Memory: \(item.item.memory) (Score: \(item.score))")
}
// 5. Query Knowledge Graph Triples (Mem0 & Graphiti feature)
let triples = try await synapse.getRelations(userId: "alex_123")
// Prints: Alex -> [lives_in] -> Bangkok
// 6. RAG — Ingest a file and retrieve LLM-ready context with citations
let pdfURL = Bundle.main.url(forResource: "Q3Report", withExtension: "pdf")!
let chunks = try await synapse.ingestDocument(fileURL: pdfURL, tags: ["finance", "Q3"])
let ragContext = try await synapse.retrieveContext(query: "What was the revenue growth?", limit: 4)
print(ragContext.contextPrompt) // Ready-to-paste LLM prompt with [1][2] citations
print(ragContext.bibliography()) // [1] "Q3Report.pdf" - Revenue Section, Page 4Why pay recurring server bills or manage Docker containers when Apple Silicon can run everything on-device for free?
| Capability / Architecture Dimension | SynapseMemory | Mem0 (Cloud / OSS) | Supermemory | Letta (MemGPT) | Zep |
|---|---|---|---|---|---|
| Zero Hosting Cost ($0.00 / Zero Server Bills) | ✅ Yes ($0.00) | ❌ Paid / Docker | ❌ Subscription | ||
| 100% On-Device Local Privacy | ✅ Complete | ❌ Cloud Only | ❌ Cloud Only | ||
| Hardware-Accelerated SIMD Vector Search | ✅ Apple Accelerate (vDSP) |
❌ Qdrant / Chroma | ❌ Cloud Index | ❌ Postgres / Chroma | ❌ Cloud Index |
| Full-Text BM25 Keyword Search | ✅ SQLite FTS5 Virtual Tables | ||||
| Hybrid Search Fusion (Dense + Sparse + Decay) | ✅ Vector + BM25 + Time Decay | ❌ Vector Only | ❌ Vector Only | ||
| Knowledge Graph Memory (Entity Triples) | ✅ Local SQLite Triples | ✅ Graphiti / NetworkX | ❌ Not Supported | ✅ Temporal Graphiti | |
Bi-Temporal Fact Superseding (validFrom/validTo) |
✅ Built-in | ✅ Built-in | ❌ Not Supported | ❌ Not Supported | |
| Document & Bookmark Ingestion (Chunk & Auto-Tag) | ✅ Built-in (ingest()) |
✅ Built-in | ❌ Not Supported | ||
| Hierarchical Working Memory Blocks (Persona State) | ✅ Built-in (coreBlock) |
❌ Not Supported | ❌ Not Supported | ✅ Core Architecture | ❌ Not Supported |
| Chronological Recall Memory Log & Playback | ✅ Built-in (recall()) |
❌ Not Supported | ❌ Not Supported | ✅ Built-in | |
| Rolling Dialogue Summarization & Context Compression | ✅ Built-in (summarize()) |
❌ Not Supported | ❌ Not Supported | ✅ Core Architecture | |
| Multi-Device Private Sync (Zero Third-Party DB) | ✅ Apple CloudKit iCloud | ❌ Cloud DB Cluster | ❌ Server Sync | ❌ Cloud Backend | |
| On-Device Apple Foundation Models & Guided Generation | ✅ Native Zero-Config | ❌ Cloud API / Ollama | ❌ Cloud API | ❌ Cloud API / Ollama | ❌ Cloud API |
| Local Offline LLM Support (Ollama on Mac) | ✅ Built-in (OllamaProvider) |
❌ Cloud Only | ❌ Cloud Only | ||
| Apple Native Ecosystem (CoreSpotlight & Siri AppIntents) | ✅ Native iOS/macOS | ❌ Not Supported | ❌ Not Supported | ❌ Not Supported | ❌ Not Supported |
| Multi-Format RAG Document Loaders (PDF, MD, Code, CSV, JSON) | ✅ Built-in (6 loaders) | ❌ Not Supported | ❌ Not Supported | ❌ Not Supported | ❌ Not Supported |
| Recursive & Syntax-Aware Chunking Strategies | ✅ 3 built-in strategies | ❌ Not Supported | ❌ Not Supported | ❌ Not Supported | ❌ Not Supported |
| On-Device Vision OCR (Apple Vision Framework) | ✅ Zero-cost on-device | ❌ Not Supported | ❌ Not Supported | ❌ Not Supported | ❌ Not Supported |
| Citation Tracking & RAG Context with Numbered References | ✅ Built-in RAGContextBuilder |
❌ Not Supported | ❌ Not Supported | ❌ Not Supported | ❌ Not Supported |
For a detailed architectural deep dive, see Competitor Comparison & Analysis.
Explore complete, granular documentation inside the docs/ directory:
| Section | Topic & Documentation Link | Description |
|---|---|---|
| 🚀 Getting Started | Quick Start Guide | Installation via SPM, requirements, and basic usage. |
| Configuration Reference | SynapseConfig options, vector dimensions, and storage settings. |
|
| 🏛️ Architecture | System Architecture | Multi-tiered memory engine, actor isolation, and subsystem design. |
| Competitor Matrix | Architectural comparison vs Mem0, Supermemory, Letta, and Zep. | |
| 📖 Guides | Agent & LLM Integration | Standard OpenAPI/JSON schema tools for SynapseAgent & Foundation Models. |
| Conversational Memory | State machine (ADD, UPDATE, DELETE), bi-temporal fact superseding. |
|
| Knowledge Graph | Entity triple extraction, SQLite graph store, and graph traversal. | |
| Document & RAG Ingestion | PDF, Markdown, Code, CSV/JSON loaders; RecursiveCharacter/Code/PDF chunkers; Vision OCR; ingest() & ingestDocument(). |
|
| RAG Pipeline Architecture | Full pipeline: Loaders → Chunkers → KnowledgeBaseIndex → RAGRetriever → RAGContext with citations & bibliography. | |
| Hybrid Search Engine | Fusion of SIMD vector search (vDSP), SQLite FTS5 BM25, and time-decay. |
|
| CloudKit Private Sync | Encrypted multi-device sync, delta change tokens, and offline queues. | |
| Apple Integrations | Spotlight system indexing, Siri AppIntents, and BGTaskScheduler. |
|
| 🛈 API Reference | SynapseClient API | Complete Swift API reference for SynapseClient actor methods, including RAG APIs. |
| Intelligence Providers | Apple Foundation Models, Ollama, OpenAI, and Mock providers. |
┌─────────────────────────────────────────────────────────────────────────────┐
│ Apple Intelligence Client App │
└──────────────────────────────────────┬──────────────────────────────────────┘
│
┌──────────────────────────────────────▼──────────────────────────────────────┐
│ SynapseClient (Actor) │
│ ┌──────────────────────┐ ┌──────────────────────┐ ┌──────────────────────┐ │
│ │ Apple Foundation │ │ SIMD Vector │ │ Knowledge Graph │ │
│ │ Model Extractor │ │ Search Engine │ │ Engine (Triples) │ │
│ ├──────────────────────┤ ├──────────────────────┤ ├──────────────────────┤ │
│ │ Supermemory Ingest │ │ Letta Hierarchical │ │ Zep Rolling Dialogue │ │
│ │ & Document Chunker │ │ Recall Memory Log │ │ Summarizer Engine │ │
│ └──────────┬───────────┘ └──────────┬───────────┘ └──────────┬───────────┘ │
└────────────┼────────────────────────┼────────────────────────┼──────────────┘
│ │ │
┌────────────▼────────────────────────▼────────────────────────▼──────────────┐
│ Core Storage Subsystem │
│ ┌─────────────────────────────────────────────────────────────────────────┐ │
│ │ Local Hybrid SQLite FTS5 + Accelerate SIMD Vector Store + Graph Store │ │
│ └────────────────────────────────────┬────────────────────────────────────┘ │
└──────────────────────────────────────┼──────────────────────────────────────┘
│ CloudKit Delta Sync
┌──────────────────────────────────────▼──────────────────────────────────────┐
│ Apple CloudKit Private iCloud Database │
│ ┌─────────────────────────────────────────────────────────────────────────┐ │
│ │ CKRecordZone: "SynapsePrivateZone" │ │
│ └─────────────────────────────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────────────────┘
- Mem0: State machine & bi-temporal fact invalidation.
- Supermemory: Document chunking & tag extraction.
- Letta / MemGPT: Tiered working memory & recall logs.
- Zep: Rolling dialogue summarization & recency decay.
SynapseMemory is licensed under the MIT License.