hhblaze/DBreeze

DBreeze – an embeddable, multi‑paradigm database for .NET

What it is – DBreeze is an open‑source, pure‑C# data‑store that can be dropped into any .NET 5+ (or older .NET Framework, .NET Core, Xamarin, Unity, etc.) application. It works on Windows, Linux, macOS and even mobile platforms via Xamarin.

Why it matters for AI/ML – Besides classic key‑value and NoSQL capabilities, DBreeze ships with an integrated vector database that supports HNSW‑based similarity search and TurboQuant compression. This makes it suitable for storing and querying embeddings produced by large language models (LLMs) or other neural networks, enabling fast nearest‑neighbor look‑ups for retrieval‑augmented generation (RAG), recommendation, clustering, etc.

Core capabilities

• ACID‑compliant, multi‑threaded: transactional guarantees with dead‑lock handling, parallel reads and synchronized writes.
• Schema‑free tables: tables are created on the fly, can live in different folders or in‑memory, and support nested/fractal tables.
• Liana‑Trie indexing: indexes never need defragmentation, keeping insert/update speed stable over time.
• High‑throughput CRUD: benchmarks claim ~500 k inserts or 260 k updates per second per core on a regular PC (as of 2012).
• Rich query features: range scans, prefix searches, multi‑parameter queries, full‑text search, and JSON/Biser serialization.
• Vector layer: HNSW graph for approximate nearest‑neighbor search, plus compression for large embedding collections.
• HA/RAID: optional Raft‑based replication for high availability.

Typical use‑cases

• Storing user profiles, session data, or configuration in embedded devices.
• Building a lightweight embedding store for LLM‑driven applications (semantic search, RAG, similarity clustering).
• Serving as the persistence layer for custom object/document stores, graph‑like structures, or full‑text indexes.
• Offline or edge scenarios where a full‑blown server DB is overkill.

How to get started

1. Add the DBreeze NuGet package to your project (available since 2014).
2. Follow the quick‑start guides on the GitHub wiki to create a DBreezeEngine, open a database folder, and begin reading/writing tables.
3. For vector work, consult the LLM Agents Documentation (linked in the repo) which shows how to insert embeddings and run similarity queries.
4. Use the provided PDF/HTML documentation for deeper configuration (tuning, backup/restore, Raft replication, etc.).

Where to find more – The project’s homepage (http://dbreeze.tiesky.com) and the GitHub wiki contain quick‑start tutorials, release notes, benchmark results, and a PDF manual. The source is BSD‑3‑Clause licensed and all binaries are signed with anti‑tampering protection.

Bottom line – DBreeze is a versatile, embeddable .NET database that goes beyond simple key‑value storage by offering built‑in vector similarity search, making it a practical building block for AI‑centric .NET applications that need fast, on‑device or low‑latency embedding retrieval.