{ "meta": { "title": "D-Central — Self-Hosted Vector Databases for RAG", "description": "Self-hostable vector databases (14) by type, language, index types, native hybrid search, quantization, persistence and OSI-open-vs-source-available license — the storage layer of a self-hosted RAG stack.", "generated": "2026-07-17T18:21:40+00:00", "last_verified": "July 2026", "version": "1.0", "license": "https://creativecommons.org/licenses/by/4.0/", "license_name": "CC BY 4.0", "source": "https://d-central.tech/self-hosted-vector-databases/", "record_count": 14, "provenance": "Each project's GitHub LICENSE file + official docs, reconciled adversarially. license_osi_open distinguishes genuine OSI open-source (Apache/BSD/MIT/PostgreSQL/AGPL) from source-available (SSPL/BSL/Elastic License/RSAL).", "disclaimer": "Feature sets and licenses change fast (several products relicensed recently); confirm the current LICENSE and docs before committing. \"Hybrid search\" here means native dense+sparse/BM25 fusion, not metadata filtering." }, "rows": [ { "db_id": "chroma", "name": "Chroma (ChromaDB)", "developer": "Chroma (chroma-core)", "db_type": "embedded-library", "primary_language": "Rust core (1.x rewrite) with Python/JS clients", "deployment": "in-process embedded (PersistentClient / in-memory) or a single self-hosted server (chroma run / Docker)", "index_types": "HNSW (SPANN in the distributed/Cloud engine)", "hybrid_search": "partial", "metadata_filtering": "yes (pre-filter)", "quantization": "none (stores full-precision vectors)", "distance_metrics": "cosine, L2, inner product (dot)", "persistence": "both", "license": "Apache-2.0", "license_osi_open": true, "license_note": "Core is Apache-2.0 (cleanly OSI-open). Hybrid caveat, not license: stable self-hosted OSS does dense + metadata + document $contains/$regex full-text FILTER, not BM25 fusion; native sparse (BM25/SPLADE) + dense RRF 'Search() API' is newly announced and documented Cloud-first — OSS-GA unconfirmed.", "managed_cloud": true, "best_for": "fastest 'pip install and go' DX for a RAG prototype; embedded first, scales to a single server later", "source_url": "https://github.com/chroma-core/chroma", "confidence": "medium" }, { "db_id": "lancedb", "name": "LanceDB", "developer": "LanceDB Inc. (built on the Lance columnar format)", "db_type": "embedded-library", "primary_language": "Rust (with Python, TypeScript, Rust, Go SDKs)", "deployment": "in-process embedded / serverless (opens a directory of Lance files, incl. on object storage); optional LanceDB Cloud/Enterprise", "index_types": "IVF-PQ, IVF-HNSW variants (HNSW), flat/brute-force; disk-based columnar Lance format", "hybrid_search": "yes", "metadata_filtering": "yes (pre-filter)", "quantization": "product (PQ), scalar (SQ), binary", "distance_metrics": "cosine, L2, dot", "persistence": "on-disk", "license": "Apache-2.0", "license_osi_open": true, "license_note": "Embedded OSS core is Apache-2.0 and includes native FTS+vector hybrid; LanceDB Cloud/Enterprise are separate proprietary offerings (does not affect the OSS boolean).", "managed_cloud": true, "best_for": "embedded multimodal RAG at scale on a laptop or object storage — native Tantivy BM25 + vector hybrid with rerankers, no server to run", "source_url": "https://github.com/lancedb/lancedb", "confidence": "high" }, { "db_id": "sqlite-vec", "name": "sqlite-vec", "developer": "Alex Garcia (asg017); backed by Mozilla Builders", "db_type": "embedded-library", "primary_language": "C (zero dependencies)", "deployment": "in-process embedded (a single loadable SQLite extension; runs anywhere SQLite/libSQL runs)", "index_types": "flat/brute-force KNN in stable v0.1.9 (2026-03-31); IVF ('experimental, not enabled') + DiskANN only in v0.1.10-alpha pre-releases; no HNSW. vec0 virtual tables with partition & metadata columns", "hybrid_search": "no", "metadata_filtering": "yes (pre-filter)", "quantization": "binary (bit vectors) + scalar (int8); float32 default", "distance_metrics": "L2, cosine, hamming (for bit vectors)", "persistence": "both", "license": "MIT OR Apache-2.0 (dual)", "license_osi_open": true, "license_note": null, "managed_cloud": false, "best_for": "lowest-friction option — add vector search to an existing SQLite/libSQL app with one extension load, no server; brute-force only at GA so best for small/medium corpora", "source_url": "https://github.com/asg017/sqlite-vec", "confidence": "high" }, { "db_id": "txtai", "name": "txtai", "developer": "NeuML (David Mezzetti)", "db_type": "embedded-library", "primary_language": "Python", "deployment": "in-process embedded embeddings database (optional FastAPI service)", "index_types": "pluggable ANN backends: Faiss (default), HNSW (hnswlib), Annoy, NumPy brute-force, Torch, pgvector, sqlite-vec", "hybrid_search": "yes", "metadata_filtering": "yes (pre-filter)", "quantization": "scalar; product/scalar via the Faiss backend", "distance_metrics": "cosine (default, normalized), L2, inner product (backend-dependent)", "persistence": "both", "license": "Apache-2.0", "license_osi_open": true, "license_note": null, "managed_cloud": false, "best_for": "all-in-one embedded RAG/semantic-search framework — native sparse(BM25)+dense hybrid, SQLite content store, pick your ANN backend", "source_url": "https://github.com/neuml/txtai", "confidence": "high" }, { "db_id": "pgvector", "name": "pgvector", "developer": "Andrew Kane", "db_type": "database-extension", "primary_language": "C", "deployment": "Postgres extension (CREATE EXTENSION vector in any existing Postgres)", "index_types": "HNSW, IVFFlat, flat/exact", "hybrid_search": "no", "metadata_filtering": "yes (pre-filter)", "quantization": "scalar (halfvec/fp16), binary (bit), plus sparsevec type", "distance_metrics": "L2, cosine, inner product, L1, hamming, jaccard", "persistence": "on-disk", "license": "PostgreSQL License", "license_osi_open": true, "license_note": "PostgreSQL License is an OSI-approved permissive (MIT/BSD-style) license — cleanly open.", "managed_cloud": false, "best_for": "self-hosters who already run Postgres — keep vectors, metadata, and joins in one ACID database (native BM25 fusion is DIY via tsvector + manual RRF)", "source_url": "https://github.com/pgvector/pgvector", "confidence": "high" }, { "db_id": "vectorchord", "name": "VectorChord (vchord)", "developer": "TensorChord (successor to pgvecto.rs)", "db_type": "database-extension", "primary_language": "Rust (pgrx)", "deployment": "Postgres extension (builds on pgvector's vector type; disk-friendly)", "index_types": "RaBitQ-based IVF, disk-friendly ANN (DiskANN-style); pairs with the vchord_bm25 companion extension for keyword ranking", "hybrid_search": "partial", "metadata_filtering": "yes (pre-filter)", "quantization": "binary (RaBitQ), scalar", "distance_metrics": "L2, cosine, dot", "persistence": "on-disk", "license": "AGPL-3.0-only OR Elastic-License-2.0 (dual)", "license_osi_open": true, "license_note": "VERIFIED dual AGPL-3.0 OR ELv2 (both VectorChord and the vchord_bm25 companion). license_osi_open=true only because AGPLv3 is genuinely selectable and copyleft-open — NOT permissive, and a real openness downgrade from its pgvecto.rs predecessor (Apache-2.0). The ELv2 option is source-available/non-OSI. Hybrid is 'partial': requires the separate vchord_bm25 extension + manual fusion.", "managed_cloud": true, "best_for": "Postgres users who need pgvector-compatible search at larger scale / lower cost via RaBitQ disk indexing — but note AGPL/ELv2 dual licensing, not the permissive Apache-2.0 of pgvecto.rs", "source_url": "https://github.com/tensorchord/VectorChord", "confidence": "high" }, { "db_id": "qdrant", "name": "Qdrant", "developer": "Qdrant Solutions GmbH", "db_type": "single-binary-server", "primary_language": "Rust", "deployment": "single self-hosted binary / Docker; scales to a distributed cluster (Raft-based sharding + replication)", "index_types": "HNSW (payload-aware / filterable); flat/brute-force for exact search", "hybrid_search": "yes", "metadata_filtering": "yes (pre-filter)", "quantization": "multiple (scalar, product, binary)", "distance_metrics": "cosine, dot, euclidean (L2), manhattan", "persistence": "both (in-memory + on-disk / memmap)", "license": "Apache-2.0", "license_osi_open": true, "license_note": "Core DB (incl. sparse+dense hybrid via the Query API) is Apache-2.0. Some cloud/enterprise features (e.g., RBAC) are proprietary but do not gate the OSS vector+hybrid engine.", "managed_cloud": true, "best_for": "purpose-built vector search with rich payload pre-filtering, native sparse(BM25/BM42)+dense fusion (RRF/DBSF), and an easy single-binary self-host", "source_url": "https://github.com/qdrant/qdrant", "confidence": "high" }, { "db_id": "redis", "name": "Redis (Query Engine / Vector Sets)", "developer": "Redis Ltd. (originally Salvatore Sanfilippo / antirez)", "db_type": "single-binary-server", "primary_language": "C", "deployment": "single self-hosted in-memory server; scales horizontally via Redis Cluster (hash-slot sharding + replicas)", "index_types": "HNSW and FLAT (brute-force) — used by both the Query Engine vector index and the Vector Sets type", "hybrid_search": "partial", "metadata_filtering": "yes (pre-filter)", "quantization": "scalar int8 (Q8) and binary in the Vector Sets type; classic Query Engine FLOAT32/FLOAT16 vectors are unquantized", "distance_metrics": "cosine, L2 (euclidean), inner product (dot)", "persistence": "both (in-memory primary; RDB snapshot + AOF for durability)", "license": "Tri-license (Redis 8+): AGPLv3 OR RSALv2 OR SSPLv1", "license_osi_open": true, "license_note": "VERIFIED. Timeline: BSD-3 → (Mar 2024) dual RSALv2/SSPLv1 source-available, non-OSI → (May 2025, Redis 8.0) AGPLv3 re-added as a third option. license_osi_open=true because — unlike Elasticsearch — AGPLv3 covers the ACTUAL Redis 8 product you build/run, and vector search (Query Engine, formerly the RediSearch module, + the new Vector Sets type) is folded into Redis 8 core under the tri-license. RSALv2 and SSPLv1 remain non-OSI options; pre-8, vector search was a source-available module only. AGPL network-copyleft still applies.", "managed_cloud": true, "best_for": "ultra-low-latency in-memory vector search co-located with caching for real-time RAG", "source_url": "https://github.com/redis/redis", "confidence": "high" }, { "db_id": "weaviate", "name": "Weaviate", "developer": "Weaviate B.V.", "db_type": "single-binary-server", "primary_language": "Go", "deployment": "single self-hosted binary / Docker; horizontally scalable cluster (Raft metadata, sharding + replication) on k8s", "index_types": "HNSW, flat/brute-force, dynamic (flat auto-upgrades to HNSW)", "hybrid_search": "yes", "metadata_filtering": "yes (pre-filter)", "quantization": "multiple (product PQ, binary BQ, scalar SQ, rotational RQ)", "distance_metrics": "cosine, dot, L2-squared, hamming, manhattan", "persistence": "both (LSM object store on-disk + in-memory HNSW graph)", "license": "BSD-3-Clause", "license_osi_open": true, "license_note": "Core is BSD-3-Clause (permissive, cleanly OSI-open) — genuinely open, no source-available trap on the DB core. Native BM25F+dense fusion (rankedFusion/relativeScoreFusion) built in.", "managed_cloud": true, "best_for": "developer-friendly native BM25F + dense hybrid search with built-in vectorizer modules", "source_url": "https://github.com/weaviate/weaviate", "confidence": "high" }, { "db_id": "opensearch", "name": "OpenSearch (k-NN plugin)", "developer": "OpenSearch Software Foundation (Linux Foundation); forked from Elasticsearch by AWS", "db_type": "search-engine", "primary_language": "Java", "deployment": "self-hosted distributed JVM cluster (shards + replicas) on k8s/VMs", "index_types": "HNSW (Lucene / faiss / nmslib engines), IVF (faiss), flat/exact k-NN", "hybrid_search": "yes", "metadata_filtering": "yes (pre-filter)", "quantization": "multiple (byte/scalar, faiss product PQ, binary/BQ, FP16 disk-based)", "distance_metrics": "cosine (cosinesimil), L2, inner product, L1, L-infinity, hamming (binary)", "persistence": "on-disk (Lucene segments)", "license": "Apache-2.0", "license_osi_open": true, "license_note": "Apache-2.0 and now under vendor-neutral Linux Foundation governance (OpenSearch Software Foundation, 2024). THE truly-open, permissive alternative to source-available Elasticsearch — the key sovereignty pick in the ES/OpenSearch split.", "managed_cloud": true, "best_for": "Apache-2.0 full search engine + native BM25 + kNN hybrid (normalization processor); the permissive Elasticsearch alternative", "source_url": "https://github.com/opensearch-project/OpenSearch", "confidence": "high" }, { "db_id": "elasticsearch", "name": "Elasticsearch", "developer": "Elastic N.V.", "db_type": "search-engine", "primary_language": "Java", "deployment": "self-hosted distributed JVM cluster (shards + replicas) on k8s/VMs", "index_types": "HNSW (Lucene), flat/brute-force exact kNN, int8/int4/BBQ-quantized HNSW", "hybrid_search": "yes", "metadata_filtering": "yes (pre-filter)", "quantization": "multiple (int8 scalar, int4, BBQ better-binary-quantization)", "distance_metrics": "cosine, dot_product, L2 (euclidean), max_inner_product", "persistence": "on-disk (Lucene segments)", "license": "Elastic License 2.0 / SSPL 1.0 (all downloadable binaries) + AGPLv3 source-only option (added 8.16, 2024)", "license_osi_open": false, "license_note": "CORRECTED to FALSE (candidate had true). Every official downloadable distribution is Elastic License 2.0 / SSPL 1.0 — both source-available, NON-OSI. Elastic added an OSI-approved AGPLv3 option in 8.16 (announced 2024) but per Elastic's own licensing FAQ, 'Elastic does not currently offer any downloadable distribution that is covered by the AGPL' — AGPL applies to source code only, so you must self-compile to get an OSI-open build. Practical/mainstream posture remains source-available; contrast Redis 8 (AGPL covers the shipped product) and OpenSearch (clean Apache-2.0).", "managed_cloud": true, "best_for": "mature search engine with native RRF hybrid (BM25 + dense kNN + ELSER sparse) at enterprise scale — but source-available licensing; choose OpenSearch if you need a permissive OSI license", "source_url": "https://github.com/elastic/elasticsearch", "confidence": "medium" }, { "db_id": "milvus", "name": "Milvus (incl. Milvus Lite embedded mode)", "developer": "Zilliz / LF AI & Data Foundation", "db_type": "distributed-cluster", "primary_language": "Go, C++ (Milvus Lite: Python wrapper over the C++ core)", "deployment": "k8s distributed cluster (disaggregated compute/storage); single-node Standalone via Docker; and Milvus Lite embedded local .db file via pymilvus — same API across all three", "index_types": "Full Milvus: HNSW, IVF (FLAT/SQ8/PQ), DiskANN, SCANN, FLAT, GPU (CAGRA/GPU_IVF), SPARSE_INVERTED_INDEX. Milvus Lite embedded: FLAT + IVF_FLAT (dense) and SPARSE_INVERTED_INDEX (sparse) ONLY — no HNSW/DiskANN locally", "hybrid_search": "yes", "metadata_filtering": "yes (pre-filter)", "quantization": "multiple (scalar SQ8, product PQ, binary) in full Milvus; limited in Lite", "distance_metrics": "cosine, L2 (euclidean), inner product (dot); Hamming/Jaccard for binary vectors", "persistence": "on-disk (object-storage-backed; indexes served in-memory/mmap). Milvus Lite: local file, memory-bound", "license": "Apache-2.0", "license_osi_open": true, "license_note": "Apache-2.0 across full Milvus and Milvus Lite (same project/repo — the two research-pass rows were deduped into this one entry). Native dense+sparse+BM25 hybrid with RRF is confirmed even in Lite.", "managed_cloud": true, "best_for": "billion-scale distributed vector search with the widest index selection and GPU indexing; Milvus Lite gives the same API for embedded local prototyping that lifts to Standalone/Distributed/Zilliz Cloud unchanged", "source_url": "https://github.com/milvus-io/milvus", "confidence": "high" }, { "db_id": "vespa", "name": "Vespa", "developer": "Vespa.ai (spun out of Yahoo!)", "db_type": "distributed-cluster", "primary_language": "Java, C++", "deployment": "self-hosted distributed cluster (stateless container + stateful content nodes); also Vespa Cloud", "index_types": "HNSW (real-time modifiable), multi-vector HNSW, flat/brute-force", "hybrid_search": "yes", "metadata_filtering": "yes (pre-filter)", "quantization": "multiple (int8, binary/hamming via tensor cell types)", "distance_metrics": "angular (cosine), euclidean, dotproduct, prenormalized-angular, hamming, geodegrees", "persistence": "both (on-disk with in-memory attributes/tensors)", "license": "Apache-2.0", "license_osi_open": true, "license_note": null, "managed_cloud": true, "best_for": "very large-scale hybrid retrieval (text + tensor in one ranking expression) with complex multi-phase ML ranking in a single engine", "source_url": "https://github.com/vespa-engine/vespa", "confidence": "high" }, { "db_id": "faiss", "name": "FAISS (Facebook AI Similarity Search)", "developer": "Meta (Fundamental AI Research)", "db_type": "library-only", "primary_language": "C++ (with Python/NumPy bindings; some GPU-accelerated)", "deployment": "in-process library — NOT a database: no server, no metadata store, no query language; you manage persistence and IDs yourself", "index_types": "flat, IVF, IVF-PQ, HNSW, LSH, IVF-SQ, PQ/OPQ, IVF-HNSW composites", "hybrid_search": "no", "metadata_filtering": "limited", "quantization": "product (PQ/OPQ), scalar (SQ), binary", "distance_metrics": "L2, inner product (cosine via normalization)", "persistence": "both", "license": "MIT", "license_osi_open": true, "license_note": "Not a database — a raw ANN library. db_type intentionally 'library-only'; no persistence/metadata/serving layer of its own.", "managed_cloud": false, "best_for": "the raw ANN engine underneath many of the above — maximum index control when you'll build the storage/metadata/serving layer yourself", "source_url": "https://github.com/facebookresearch/faiss", "confidence": "high" } ] }