Private preview · By invitation

The AI-native embedded database for Flutter.

Vector search, graph traversal, agent memory, offline-first sync — all on-device. Build private, fast, AI-powered mobile apps without renting a cloud database.

9 platforms · 32 Rust crates · 2,630+ tests

main.dart
import 'package:edgegraphbox/edgegraphbox.dart';

final store = await Store.open(StoreConfig.auto('kb'));
final docs  = store.box<Document>();

// Hybrid search: BM25 + HNSW with Reciprocal Rank Fusion
final hits = await docs.hybridSearch(
  textQuery: 'transformers attention',
  embedding: queryVector,
  textWeight:   0.3,
  vectorWeight: 0.7,
);

// Graph traversal — citation network, depth-limited
final related = await docs.traverse(
  from: paperId, edge: 'cites', maxDepth: 3,
);

// Agent memory — episodic recall with importance decay
await agent.remember('user prefers concise answers');
final ctx = await agent.recall(query, k: 8);

Why now

On-device AI is here. The database hasn't caught up.

SQLite is 24 years old

It wasn't designed for vectors, graphs, or AI. Bolting on extensions means slow queries, fragile builds, and no semantic search at all.

Pinecone & Co. need a cloud

Mobile RAG over a network round-trip is dead on arrival. Privacy, latency, offline — three reasons your AI features can't depend on a cloud vector DB.

Nobody else covers all four

Embedded + vector + graph + agent memory for Flutter — that's the gap. Realm, Hive, Isar, ObjectBox: none of them ship this stack.

Everything an AI-native mobile app needs — in one library.

HNSW Vector Search

Hierarchical Navigable Small World index with SIMD acceleration (AVX2/NEON). Sub-millisecond similarity search on millions of embeddings — fully on-device.

BM25 Full-Text Search

Production-grade ranking with Snowball stemmers in 14+ languages. Combine with vector search via Reciprocal Rank Fusion for hybrid retrieval.

Graph Traversal

Native entity relationships, BFS/DFS with depth limits and cycle detection. Build citation networks, knowledge graphs, recommendation paths — without an extra service.

AI Agent Memory

Episodic + semantic memory with importance decay, consolidation, and context-aware retrieval. Drop-in persistence for chat agents, copilots, RAG pipelines.

Offline-First Sync

HLC-based hybrid logical clocks, configurable conflict resolution, CRDT support. Optional self-hosted server for multi-device sync — never required.

Real-Time Subscriptions

Reactive queries, live subscriptions, push notifications. Update the UI when the data changes — without polling, without rebuilding lists.

9 Platforms, One Codebase

iOS, Android, macOS, Windows, Linux, Web (WASM partial). Built on flutter_rust_bridge — write Dart, ship Rust performance.

Architecture

A thin Dart API over a battle-tested Rust core.

32 specialized crates, 2,630+ tests. Every layer designed to be small, replaceable, and verifiable.

Dart / Flutter

Store · Box<T> · Query · Entity · AgentMemory

egb-ffi · flutter_rust_bridge

Generated bindings — type-safe, zero-copy where it counts.

Rust core (32 crates)

egb-core · egb-storage · egb-vector · egb-query · egb-graph · egb-agent · egb-rag · egb-embedding · egb-inference · egb-sync · egb-crdt · egb-server · egb-auth · egb-permission · egb-notification · egb-cypher · egb-codegen · …

Built quietly. Tested obsessively.

32
Rust crates
2,630+
tests passing
246
features complete
9
platforms

NeuralPaper, PKB, and PubSeek showcase apps validate every layer. Currently in private development.

Interest

Tell us why you're interested.

We're deciding whether to invest the next 6 months full-time. If you'd use, build on, contribute to, or fund EdgeGraphBox, we want to hear from you.

I'm interested as a…

We'll only use your email to follow up. Stored in EU (Cloudflare). No newsletters, no third parties.