Endee Serverless Overview
Endee is a high-performance C++ vector database for AI search, RAG, and hybrid retrieval. Endee Serverless is the fully-managed version: no Docker, no servers, and no infrastructure to maintain. You spin up a database from the dashboard, grab an auth token, and start upserting and searching from your application in minutes.
In Endee, a collection holds objects, and each object can carry values for several named, typed fields at once (dense vectors, sparse (keyword) vectors, and multi-vectors), so a single object can be searched many ways.
Want to run Endee yourself? Endee Serverless is the managed platform. To run Endee locally with Docker, follow v1 setup.
Key Features
- Multi-field objects: One object, many vector fields. Combine dense embeddings, sparse keywords, and multi-vectors under a single id.
- Hybrid search: Query any subset of fields in one request;
search()returns per-field results, which you can fuse withrerank()(Reciprocal Rank Fusion). - Fast ANN search: Efficient approximate nearest-neighbour search over embeddings using the HNSW algorithm.
- Multiple distance metrics:
cosine(default, normalized client-side),l2, andip. - Precision trade-offs: Six quantization levels (
float32,float16,int16,int8,int8e,binary) to balance memory, speed, and recall. - Advanced filtering: Server-side metadata filters with operators like
$eq,$in,$range,$gt, and$lt. - Fully managed: Serverless tokens encode the region, so the SDK targets the right endpoint automatically. No base URL setup required.
Use Cases
- Semantic search: Build search systems that understand meaning, not just keywords.
- High-performance RAG: Fast multi-field retrieval and multi-vector reranking for low-latency context lookup.
- E-commerce: Semantic + keyword hybrid search with metadata filters (price, category, rating, …).
- Recommendation systems: Power personalized recommendations based on similarity matching.
Workflow
Create a database
Sign in at app.endee.io , create a database, and grab its auth token.
Create a collection
Declare your named, typed fields: a dense vector, a sparse keyword field, a multi_vector field, or any combination.
Upsert objects
Insert objects with a unique id, optional meta and filter tags, and one entry per field you populate. upsert is insert-or-replace by id.
Search and rerank
Query one or more fields in a single request. search() returns one ranked list per field; fuse them into a single list with rerank() when you need hybrid results.