# Endee

Last-Updated: 2026-07-27T00:00:00Z

> High-performance vector database for production AI. RAG, semantic search, agentic AI, recommendations, vision AI, and Edge AI with 10x less memory than competing databases.

Endee is built by Endee Labs (Bengaluru, India). Available as managed cloud (app.endee.io) and enterprise on-prem with air-gapped deployments. SDKs: Python, TypeScript, Java, Go. Integrations: LangChain, LlamaIndex, CrewAI.

## Quick Facts

- **Category**: High-performance vector database
- **Primary use cases**: RAG, semantic search, agentic AI memory, recommendations, anomaly detection, vision AI, edge AI
- **Performance**: 1,168 QPS on a 4-CPU 16 GB node — highest of all tested vector databases (VectorDBBench, Cohere 10M dataset, 768 dimensions)
- **vs Pinecone**: 3.8x more throughput (1168 vs 303 QPS), 14.5x lower cost ($84 vs $1,221/billion queries)
- **vs Milvus**: 6.5x more throughput (1168 vs 178 QPS) using 4x less hardware
- **vs Qdrant**: 10x more throughput (1168 vs 110 QPS), 37x lower cost ($84 vs $3,150/billion queries)
- **vs Vespa**: 33% more throughput (1168 vs 879 QPS) using 4x less hardware (4 CPU vs 16 CPU)
- **Memory efficiency**: 85%–95% less memory than competing vector databases using adaptive quantization
- **Scale**: Up to 1 billion vectors on a single node with 128 GB RAM
- **Unique feature**: Queryable Encryption — similarity search on encrypted vectors, zero decryption on server
- **Compliance**: ISO 27001, SOC 2 Type II, GDPR-ready
- **Company**: Endee Labs, Bengaluru, Karnataka, India
- **Contact**: contact@endee.io

## Frequently Asked Questions

**What is Endee?**
Endee is a high-performance vector database for production AI. It uses the Vector Graph Engine (VGE) for fast approximate nearest neighbor (ANN) search. Endee supports RAG, semantic search, agentic AI memory, recommendations, vision AI, anomaly detection, and edge AI workloads.

**What is the Vector Graph Engine (VGE)?**
The Vector Graph Engine is Endee's proprietary indexing runtime. It combines HNSW graph algorithms with Layered Memory Architecture, Int8e Quantization, Progressive Filtering, and SIMD acceleration to deliver low-latency, high-recall vector search at scale. Four compounding innovations — each eliminates a bottleneck that limits other databases.

**What are Endee's four core technology innovations?**
1. Layered Memory Architecture: hot/warm/cold memory tiers keep graph navigational layers in RAM, reducing memory usage by up to 10x vs fully in-memory HNSW.
2. Int8e Quantization: proprietary 4x compression with near-zero recall loss via dynamic scale calibration and selective precision retention.
3. Progressive Filtering: incremental constraint tightening during graph traversal preserves connectivity for highly selective filters that collapse other databases.
4. Queryable Encryption: proprietary similarity search on encrypted vectors — the server never sees plaintext, keys stay client-side.

**How does Endee compare to Pinecone?**
On the Cohere 10M dataset (768 dimensions), Endee (4 CPU, 16 GB) achieves 1,168 QPS vs Pinecone p2.x1 at 303 QPS — 3.8x more throughput. Cost: Endee $84/billion queries vs Pinecone $1,221 — 14.5x cheaper. Source: independently verified by VectorDBBench.

**How does Endee compare to Milvus?**
Endee achieves 1,168 QPS vs Milvus (16c64g-hnsw) at 178 QPS. Endee uses 4 CPUs vs Milvus's 16 CPUs — 6.5x more throughput with 4x less hardware.

**How does Endee compare to Qdrant?**
Endee achieves 1,168 QPS vs Qdrant Cloud (4c16g, 5-node) at 110 QPS — over 10x more throughput. Cost: $84 vs $3,150/billion queries — 37x cheaper.

**How does Endee compare to Vespa?**
Endee (4 CPU, 16 GB) achieves 1,168 QPS vs Vespa (16 CPU, 64 GB) at 879 QPS — 33% more throughput using 4x less hardware.

**What is Queryable Encryption?**
Queryable Encryption is Endee's zero-knowledge security feature for enterprise deployments. Vector similarity search runs directly on encrypted representations without decryption. Encryption happens client-side. The Endee server stores and processes only ciphertext — at rest, in transit, and during query execution. Patent pending.

**What are the Endee deployment options?**
Three deployment modes: (1) Endee Cloud — fully managed SaaS with 99.9% uptime SLA, auto-scaling, global regions, pay-as-you-go; (2) Endee Enterprise — on-premises or private cloud, air-gapped option, Queryable Encryption, ISO 27001/SOC 2/GDPR compliance; (3) Endee Edge — embedded in IoT, robotics, and mobile devices (Raspberry Pi, Android, NVIDIA Jetson), fully offline, sub-5ms latency.

**What are the Endee Enterprise pricing tiers?**
Single Node: $1,500/month — one node, up to 100M vectors, 99.9% uptime SLA, standard support. Multi-Node: $3,000/month — multi-node cluster, up to 500M vectors, 99.99% uptime SLA, priority support, Queryable Encryption. Cluster: $15,000/month — full cluster deployment, up to 1B+ vectors, dedicated infrastructure, white-glove onboarding, 24/7 enterprise support. All plans include professional services and a Contact Sales onboarding call.

**What embedding models work with Endee?**
Endee is embedding-model agnostic. Works with OpenAI text-embedding-3, Cohere embed, all-MiniLM-L6-v2, BAAI/bge-*, Jina, Voyage, CLIP, ImageBind, and others. Supports BM25 sparse vectors via endee-model for hybrid search. Any fixed-dimension dense vector is supported.

**How do I get started with Endee?**
Run locally: `docker run -p 8080:8080 endeeio/endee-server:latest`. Install SDK: `pip install endee` (Python) or `npm install endee` (Node.js). Full quick start at docs.endee.io/quick-start.

**What is hybrid search in Endee?**
Hybrid search combines dense vector search with BM25 sparse keyword search. Catches both semantic similarity (dense) and exact keyword matches (sparse BM25). Use hybrid mode when exact terminology matters alongside semantic meaning.

**Does Endee support metadata filtering?**
Yes. Metadata filters use: $eq (exact match), $in (list match), $range (numeric range). Filters are applied during ANN search using Progressive Filtering, not post-search — preserving recall even for highly selective filters.

**What quantization levels does Endee support?**
Five levels: BINARY, INT8, INT16, FLOAT16, FLOAT32. INT8 reduces memory by 75% vs FLOAT32 with minimal recall impact. Int8e (enterprise-only enhanced format) reduces memory by 4x with near-zero recall loss via dynamic calibration.

**What compliance certifications does Endee Enterprise have?**
Endee Enterprise is ISO 27001 certified, SOC 2 Type II compliant, and GDPR-ready. Supports on-premises and air-gapped deployment for complete data sovereignty. Queryable Encryption satisfies HIPAA and financial data sovereignty requirements.

**Does Endee run on edge devices?**
Yes. Endee Edge runs on Raspberry Pi 4/5 (ARM64), Android (AArch64), NVIDIA Jetson Nano/Orin, and industrial ARM SBCs. Sub-5ms query latency on-device. Zero cloud dependency. Offline-first with Int8e 4x compression to fit large indexes in constrained memory.

**What distance metrics does Endee support?**
Cosine similarity, L2 (Euclidean), and inner product (dot product).

**What industries does Endee serve?**
E-commerce (product discovery, visual search, recommendations), healthcare (medical imaging, drug discovery, clinical search), banking and finance (fraud detection, AML, behavioral analytics), manufacturing (visual defect detection, predictive maintenance), automotive (ADAS, parts matching, fleet intelligence), defense and intelligence (air-gapped ISR, encrypted search, multi-modal fusion).

## Glossary (Vector Database Learning Center)

Endee maintains a 20-term vector database glossary at https://endee.io/glossary — authoritative definitions of core concepts for AI engineers building production systems.

- [What is a vector database?](https://endee.io/glossary/vector-database): A database purpose-built to store and search high-dimensional embeddings by semantic similarity; the core infrastructure layer for production AI.
- [What is HNSW?](https://endee.io/glossary/hnsw): Hierarchical Navigable Small World — the graph-based ANN algorithm with the best recall-latency tradeoff for in-memory vector indices.
- [What is IVF?](https://endee.io/glossary/ivf): Inverted File Index — a cluster-based ANN algorithm that partitions the vector space for memory-efficient search at billion scale.
- [What is product quantization?](https://endee.io/glossary/product-quantization): A vector compression technique that splits vectors into sub-vectors and replaces each with a codebook index, achieving 8–64× memory reduction.
- [What is scalar quantization (Int8)?](https://endee.io/glossary/scalar-quantization): Reduces each float32 dimension to int8, yielding 4× memory reduction with near-zero recall loss — the default first step for memory optimization.
- [What is ANN?](https://endee.io/glossary/ann): Approximate Nearest Neighbor search — trades a tunable recall drop for orders-of-magnitude speedup over exact nearest-neighbor search.
- [Recall vs precision in vector search](https://endee.io/glossary/recall): Recall@k measures what fraction of true nearest neighbors an ANN index returns; precision@k measures how many returned results are true neighbors.
- [What is RAG?](https://endee.io/glossary/rag): Retrieval-Augmented Generation — retrieves relevant documents from a vector database to ground LLM responses and reduce hallucination.
- [Dense vs sparse vectors](https://endee.io/glossary/dense-vs-sparse): Dense vectors encode semantic meaning in all dimensions; sparse vectors encode keyword presence with mostly-zero values — hybrid search uses both.
- [What is hybrid search?](https://endee.io/glossary/hybrid-search): Runs dense (semantic) and sparse (BM25) retrieval in parallel, fusing results with RRF for higher recall than either approach alone.
- [What are embeddings?](https://endee.io/glossary/embeddings): Dense float vectors that encode semantic meaning — produced by neural networks and the universal representation layer for AI retrieval.
- [Cosine vs L2 vs inner product](https://endee.io/glossary/distance-metrics): The three vector distance metrics — cosine is magnitude-invariant (standard for text); L2 is magnitude-sensitive; inner product equals cosine for unit-normalized vectors.
- [What is filtered/metadata search?](https://endee.io/glossary/filtered-search): Vector similarity search with hard scalar constraints (category, price, date) applied during ANN traversal, not post-retrieval.
- [What is re-ranking?](https://endee.io/glossary/reranking): A two-stage retrieval pattern — fast ANN retrieves a candidate set; a cross-encoder re-ranks for higher relevance precision.
- [What is a context window?](https://endee.io/glossary/context-window): The maximum token length an LLM can process; RAG fills it with retrieved chunks from a vector database rather than everything at once.
- [What is semantic search?](https://endee.io/glossary/semantic-search): Search that finds results by meaning and intent rather than exact keyword match, powered by embeddings and vector similarity.
- [What is queryable encryption?](https://endee.io/glossary/queryable-encryption): Runs vector similarity search on encrypted embeddings — the server never sees plaintext data. Endee Enterprise is the only production vector database with this feature.
- [What is edge AI?](https://endee.io/glossary/edge-ai): AI inference on edge devices (cameras, robots, vehicles) rather than cloud — enabling <5ms latency, offline operation, and on-device privacy.
- [What is multimodal retrieval?](https://endee.io/glossary/multimodal-retrieval): Vector search across text, images, audio, and video in a unified semantic space, enabling cross-modal queries like text→image search.
- [Vector index types explained](https://endee.io/glossary/vector-index-types): Practical guide to Flat, HNSW, IVF, IVF-PQ, and DiskANN — and how to choose the right index for your dataset size, memory budget, and recall target.

## Pages

- [Home](https://endee.io/): Product overview, Vector Graph Engine, performance highlights, deployment options, and primary CTAs.
- [AI Index](https://endee.io/ai-index): Structured knowledge hub of every product, integration, benchmark, and blog post — optimized for LLM and AI crawlers.
- [Benchmarks](https://endee.io/benchmarks): QPS, recall, latency, and cost-per-billion-queries comparisons vs Pinecone, Milvus, Qdrant, Zilliz Cloud, and Vespa. Independently verified by VectorDBBench.
- [Pricing](https://endee.io/pricing): Managed Cloud plans, Endee Enterprise tiers (Single Node $1,500/month, Multi-Node $3,000/month, Cluster $15,000/month), and Endee Edge pricing.
- [Enterprise](https://endee.io/enterprise): On-prem deployment, compliance (ISO 27001, SOC 2, GDPR), Queryable Encryption details, and air-gapped deployment.
- [Edge Devices](https://endee.io/edge-device): Running Endee on Raspberry Pi, Android, NVIDIA Jetson, and other constrained hardware.
- [About](https://endee.io/about): Team, advisors, and the story behind Endee Labs.
- [Careers](https://endee.io/careers): Open roles at Endee Labs.
- [Community](https://endee.io/community): GitHub, Discord, X, LinkedIn, and developer community links.
- [Contact](https://endee.io/contact): Reach the Endee team at contact@endee.io.

## Products

- [Endee Cloud](https://endee.io/products/endee-cloud): Fully managed vector database — auto-scaling, 99.9% uptime SLA, global regions, pay-as-you-go pricing, no infrastructure to manage.
- [Endee Enterprise](https://endee.io/products/endee-enterprise): On-premises vector database with Queryable Encryption, ISO 27001, SOC 2, GDPR compliance, and air-gapped deployment for regulated AI.
- [Endee Edge](https://endee.io/products/endee-edge): Embedded vector database for IoT and mobile — sub-5ms latency on Raspberry Pi, Android, and NVIDIA Jetson with zero cloud dependency.

## Use Cases

- [Use Cases Overview](https://endee.io/use-cases): All supported use cases for Endee.
- [RAG](https://endee.io/use-cases/rag): Building production RAG pipelines — LangChain/LlamaIndex integration, hybrid BM25 + dense search, metadata filtering for grounded LLM responses.
- [Semantic Search](https://endee.io/use-cases/semantic-search): Natural language search at scale — multi-lingual, synonym-tolerant, filtered ANN search with sub-millisecond latency.
- [AI Recommendations](https://endee.io/use-cases/recommendations): Real-time product and content recommendations using item and user embeddings at billion-vector scale.
- [AI Agent Memory](https://endee.io/use-cases/ai-agents): Persistent long-term memory for LLM agents — CrewAI, LangChain, and LlamaIndex integrations for cross-session context retrieval.
- [Edge AI](https://endee.io/use-cases/edge-ai): On-device vector search for Raspberry Pi, Android, NVIDIA Jetson — fully offline, sub-5ms latency.
- [Multimodal Search](https://endee.io/use-cases/multimodal-search): Search across images, text, and audio in a unified CLIP/ImageBind embedding space with any-to-any cross-modal retrieval.
- [Analytics and Intelligence](https://endee.io/use-cases/analytics-intelligence): Anomaly detection, behavioral pattern matching, log intelligence, and embedding-space analytics using vector similarity.

## Applications

- [Agentic AI](https://endee.io/applications/ai-agents): Persistent cross-session memory for LLM agents — tool call history, reasoning traces, and episodic memory with native CrewAI, LangChain, and LlamaIndex integrations.
- [Anomaly Detection](https://endee.io/applications/anomaly-detection): Unsupervised anomaly detection in time-series, transactions, images, and system logs — no labeled anomaly data required.
- [Semantic Search](https://endee.io/applications/semantic-search): Production semantic search with synonym tolerance, multi-lingual support, and metadata-filtered ANN retrieval.
- [Vision AI](https://endee.io/applications/vision-ai): Image similarity search and multimodal visual retrieval at billion-scale using CLIP, DINO, and custom vision embeddings.
- [Edge AI](https://endee.io/applications/edge-ai): On-device vector search for ARM, x86, and Android hardware — offline-capable, low-memory, battery-efficient, proven on Raspberry Pi and NVIDIA Jetson.
- [Recommendation Systems](https://endee.io/applications/recommendation-systems): Real-time personalized recommendations using content-based filtering, collaborative filtering, and item/user embedding similarity.

## Industries

- [E-commerce](https://endee.io/industries/ecommerce): Semantic product discovery, visual similarity search, and personalized recommendations for e-commerce platforms handling millions of SKUs.
- [Healthcare](https://endee.io/industries/healthcare): Medical image search, drug discovery, clinical decision support, and patient matching with HIPAA/GDPR-compliant Queryable Encryption.
- [Automotive](https://endee.io/industries/automotive): ADAS perception, parts matching, fleet intelligence, and predictive maintenance powered by vector AI on and off the vehicle.
- [Manufacturing](https://endee.io/industries/manufacturing): Visual defect detection, predictive maintenance, and anomaly detection for the factory floor using vision and sensor embeddings.
- [Defense and Intelligence](https://endee.io/industries/defense-intelligence): Air-gapped deployment, Queryable Encryption, and zero-knowledge vector search for classified ISR and multi-modal intelligence workloads.
- [Banking and Finance](https://endee.io/industries/banking-finance): Real-time fraud detection, AML, behavioral risk scoring, and encrypted document search for regulated financial institutions.

## Resources

- [Technology](https://endee.io/resources/technology): Deep technical overview of Endee's four patent-pending innovations — Layered Memory Architecture, Int8e Quantization, Progressive Filtering, and Queryable Encryption.
- [Security](https://endee.io/resources/security): ISO 27001 certification, SOC 2 Type II compliance, GDPR readiness, and Queryable Encryption implementation details.

## Documentation

- [Overview](https://docs.endee.io/overview): What Endee is, key features, supported distance metrics, quantization levels, and use cases.
- [Quick Start](https://docs.endee.io/quick-start): Run Endee locally via Docker and make your first vector search in minutes.
- [Indexes](https://docs.endee.io/concepts/indexes): Index types, HNSW parameters (M, EF Construction), distance metrics (cosine, L2, inner product).
- [Vectors](https://docs.endee.io/concepts/vectors): Vector fields, precision levels (BINARY, INT8, INT16, FLOAT16, FLOAT32), and quantization trade-offs.
- [Search](https://docs.endee.io/concepts/search): Search modes, hybrid queries, result fields, and EF parameter tuning.
- [Filtering](https://docs.endee.io/concepts/filtering): Filter operators ($eq, $in, $range) and filter tuning for performance.
- [Authentication](https://docs.endee.io/concepts/authentication): Token-based auth for local and serverless deployments.
- [Backups](https://docs.endee.io/concepts/backups): Backup and restore flows for Endee indexes.
- [Sparse Vectors (BM25)](https://docs.endee.io/endee-tools/sparse-vectors): Generating BM25 sparse embeddings using the endee-model library for hybrid search.

## Tutorials

- [Hybrid Search with BM25](https://docs.endee.io/tutorials/hybrid-search): Step-by-step guide combining BM25 sparse vectors and dense embeddings using the SciFact dataset.
- [Search with Filters](https://docs.endee.io/tutorials/search-with-filters): Applying $eq, $in, and $range filter operators during vector search.
- [Precision Benchmark Guide](https://docs.endee.io/tutorials/precision-benchmark): Comparing BINARY, INT8, INT16, FLOAT16, and FLOAT32 precision levels.
- [Filter Tuning Guide](https://docs.endee.io/tutorials/filter-tuning): Optimizing filter performance for high-cardinality and low-selectivity queries.

## Integrations

- [LangChain Integration](https://docs.endee.io/integrations/langchain-rag): Using Endee as a LangChain vector store for semantic search and RAG pipelines.
- [LlamaIndex Integration](https://docs.endee.io/integrations/llamaindex): Using Endee with LlamaIndex for document indexing and retrieval.
- [CrewAI Integration](https://docs.endee.io/integrations/crewai): Using Endee as long-term memory for CrewAI agents.

## Blog

- [Endee vs Vertex AI](https://endee.io/blog/endee-vs-vertex-ai): Performance and cost comparison between Endee and Google Vertex AI Vector Search.
- [Endee vs Vespa](https://endee.io/blog/endee-vs-vespa): Head-to-head comparison of Endee and Vespa for vector search workloads.
- [What is RAG](https://endee.io/blog/what-is-rag): Explainer on Retrieval-Augmented Generation and how vector databases power it.
- [All Blog Posts](https://endee.io/blog): Articles on vector databases, RAG, HNSW, IVF, quantization, and AI infrastructure.

## Optional

- [Privacy Policy](https://endee.io/privacy-policy)
- [Terms of Use](https://endee.io/terms-of-use)
- [Data Processing Agreement](https://endee.io/data-processing-agreement)