# Endee Documentation

> Endee is a high-performance vector database for approximate nearest neighbor (ANN) search. It supports dense vectors, sparse (BM25) vectors, and multi-vectors — enabling semantic search, keyword search, and hybrid search in a single collection. Available as a fully managed serverless platform (Endee Cloud) and self-hosted via Docker (v1).

Docs site: https://docs.endee.io
Dashboard: https://app.endee.io

---

## V2 (Serverless — Current)

V2 is the recommended version. It uses a serverless token for authentication and a multi-field collection model.

### Getting Started

- [Overview](https://docs.endee.io/v2/overview): What Endee is and how it works
- [Quick Start](https://docs.endee.io/v2/quick-start): Create a database, install the SDK, upsert objects, and run your first search

### Concepts

- [Collections](https://docs.endee.io/v2/concepts/indexes): Creating and managing collections with typed fields (vector, sparse, multi_vector)
- [Objects](https://docs.endee.io/v2/concepts/vectors): Object structure, upsert, get, delete, precision levels (float32 → binary)
- [Search](https://docs.endee.io/v2/concepts/search): Single-field and multi-field search, hybrid search with RRF fusion, reranking
- [Filtering](https://docs.endee.io/v2/concepts/filtering): Metadata filters ($eq, $in, $range, $gt, $gte, $lt, $lte)
- [Authentication](https://docs.endee.io/v2/concepts/authentication): Serverless token usage and security best practices
- [Backups](https://docs.endee.io/v2/concepts/backups): Creating and restoring collection backups

### Endee Tools

- [Sparse Vectors (BM25)](https://docs.endee.io/v2/endee-tools/sparse-vectors): Using endee_bm25 for keyword-aware hybrid search

### Tutorials

- [Hybrid Search with BM25](https://docs.endee.io/v2/tutorials/hybrid-search): Combine BM25 keyword matching with dense semantic search using rerank()
- [Search with Filters](https://docs.endee.io/v2/tutorials/search-with-filters): Use metadata filters to narrow search results with all seven filter operators
- [Precision Guide](https://docs.endee.io/v2/tutorials/precision-benchmark): Benchmark float32, float16, int16, int8e, int8, and binary to find the best speed-accuracy trade-off
- [Filter Tuning Guide](https://docs.endee.io/v2/tutorials/filter-tuning): Tune filter parameters for optimal performance

---

## V1 (Self-Hosted via Docker)

V1 runs locally or on your own infrastructure. Authentication uses a base URL and optional auth token.

### Getting Started

- [Overview](https://docs.endee.io/overview): Introduction to the self-hosted deployment model
- [Quick Start](https://docs.endee.io/quick-start): Docker setup, SDK install, and first search

### Concepts

- [Collections](https://docs.endee.io/concepts/indexes): Collection creation and management
- [Objects](https://docs.endee.io/concepts/vectors): Object structure and upsert
- [Search](https://docs.endee.io/concepts/search): Search parameters and hybrid search
- [Filtering](https://docs.endee.io/concepts/filtering): Metadata filters
- [Authentication](https://docs.endee.io/concepts/authentication): Auth token configuration
- [Backups](https://docs.endee.io/concepts/backups): Backup and restore

### Endee Tools

- [Sparse Vectors (BM25)](https://docs.endee.io/endee-tools/sparse-vectors): BM25 sparse embeddings with endee-model

### Tutorials

- [Hybrid Search with BM25](https://docs.endee.io/tutorials/hybrid-search): End-to-end hybrid search notebook
- [Search with Filters](https://docs.endee.io/tutorials/search-with-filters): Filtered search with all operators
- [Precision Guide](https://docs.endee.io/tutorials/precision-benchmark): Vector precision benchmarking
- [Filter Tuning Guide](https://docs.endee.io/tutorials/filter-tuning): Filter parameter tuning

### Integrations

- [LangChain RAG](https://docs.endee.io/integrations/langchain-rag): Use Endee as a vector store in LangChain RAG pipelines
- [LlamaIndex](https://docs.endee.io/integrations/llamaindex): Endee as a vector store in LlamaIndex
- [CrewAI](https://docs.endee.io/integrations/crewai): Endee memory store for CrewAI agents

---

## Key Concepts for LLMs

- A **collection** holds **objects**. Each object has an `id`, optional `meta` (returned in results), optional `filter` (used for filtering), and `fields` (the actual vector data).
- **Fields** are typed: `vector` (dense ANN), `sparse` (BM25 keyword), `multi_vector` (ColBERT-style).
- **Search** takes a `fields` map — each entry is `{ query, limit, ef_search }`. Results are always keyed by field name.
- **Hybrid search** runs multi-field search in one call then calls `rerank()` to fuse results via Reciprocal Rank Fusion (RRF).
- **Precision** is set per-field: `float32` (default), `float16`, `int16`, `int8e` (enhanced, recommended), `int8`, `binary`.
- **V2 SDK** (Python): `pip install endee` — client is `Endee("serverless-token")`, search is `collection.search(fields={...})`.
- **V2 SDK** (TypeScript/Node): `npm install endee` — client is `new Endee("serverless-token")`, search is `await collection.search({ fields: {...} })`.