LatticeDB
An embedded, single-file database that combines graph relationships, vector similarity search, and full-text search — built for local AI, RAG, and agent-memory apps.
🔗 Visit LatticeDB
Description
Building an app that needs to understand relationships between things (like a knowledge base or an AI agent's memory), search by meaning (semantic/vector search), and search by keyword (full-text) usually means running three different databases side by side and stitching the results together yourself. LatticeDB packages all three into a single embedded database that lives in one portable file on disk, so a local app or an AI agent's memory system can query connected, semantic, and textual data through one interface without standing up any servers.
It supports graph traversal alongside HNSW-based vector similarity search and BM25 full-text search in the same query layer, using a Cypher-based query language extended with vector (<=>) and full-text (@@) operators. It's ACID-compliant with write-ahead log durability, supports nodes/edges/labels/arbitrary properties like a typical graph database, and ships with Python, TypeScript, and Go bindings — all under the MIT license, with no external server or cluster to manage.
💬 Our review
The short version: a genuinely useful piece of infrastructure if you're building a local-first RAG system, knowledge tool, or agent-memory layer and don't want to run separate graph, vector, and search-index servers — the honest trade-off is you're picking a young, single-file engine over mature, battle-tested standalone databases.
Combining graph traversal, vector search, and full-text search in one embedded file is the real differentiator versus running Neo4j, a vector database (Qdrant/Weaviate/LanceDB), and a search engine separately — for local or single-machine apps, that's a meaningful reduction in operational complexity. What you give up is the depth and community of each specialized alternative: Neo4j's graph tooling, a dedicated vector database's scaling story, or a mature full-text engine's query features are each individually more capable than LatticeDB's combined implementation, and as a newer project it hasn't been proven at the scale or under the workloads that those established tools have.
💰 Pricing
📊 Global score
🤖 AI-enriched data
Open source sous licence MIT, fichier unique, aucun serveur requis.
Pros
Graphe + recherche vectorielle (HNSW) + plein texte (BM25) dans un seul fichier
Aucun serveur à gérer — base de données embarquée portable
Transactions ACID avec journal de write-ahead
Bindings Python, TypeScript et Go
Cons
Moins mature que des bases spécialisées établies (Neo4j, Qdrant...)
Pas conçu pour des déploiements distribués à grande échelle
Communauté et écosystème encore restreints