AI overview
NetlyDB does not run AI models. It keeps the data AI features depend on - documents, their embeddings and full-text indexes - next to your application data, and gives AI agents a governed way to read it.
The pieces
| Piece | What it gives an AI feature | Status |
|---|---|---|
| Vector search | Retrieve documents by meaning, from embeddings you store. | Available |
| Full-text search | Retrieve documents by keywords and exact phrases, ranked. | Available |
IEmbeddingProvider | Plug your embedding model in, so NetlyDB can embed a text query itself. | Available |
| MCP server | Read-only, authenticated tools an AI agent can call to explore and query allowed collections. | Available Enterprise |
| Management UI Chat | Ask questions about your data through a language model that uses the MCP tools (OpenAI API key required). | Available Enterprise |
| Managed embedding generation | Indexes that embed documents automatically as they change. | Not yet published |
Access patterns
Retrieval in your own code
Your application embeds the user's question, calls VectorSearch and/or FullTextSearch, filters
the candidates by whatever rules apply to the user, and passes the chosen documents to the model it uses. NetlyDB is the
store and the index; prompt construction, model calls and access rules stay in your code, where you can test them.
An agent with tools
An AI agent connects to the MCP server and calls its tools directly. This suits assistants for operators and analysts who need to explore data. Control exposure with the collection allow-list, per-client keys and result caps, and read what MCP does not do for you.
Roadmap
- Planned Measuring the recall of approximate vector search, and an explicit exact-search option.
- Planned Recording which embedding model or text analyzer built an index, and rebuilding when it changes.
- Planned Subject to demand: filtering inside vector search, hybrid keyword-plus-vector ranking, vector quantization, language-aware analyzers, and managed embedding pipelines.