Concurrency and scale — a native async client, lexical/hybrid retrieval, per-document TTL, a new framework connector, and a graph data-loss fix.

Added

• Native async client (#270) — AsyncDynavec with async/await upsert / search / get / delete for concurrent agent and web workloads.
• BM25 lexical + dense hybrid (#233, #264) — BM25Retriever, BM25HybridRetriever, and db.hybrid_search() fusing dense ANN and sparse BM25 via Reciprocal Rank Fusion.
• Per-document TTL (#22) — ttl_seconds= on upsert()/update() with automatic DynamoDB TTL provisioning.
• Semantic Kernel connector (#69) — DynavecStore / DynavecCollection implementing Microsoft Semantic Kernel's vector-store interface.
• search_many() (#258) — run several queries against a namespace concurrently.
• Optimistic concurrency on update() (#24) — expected_version=, ConflictError, UpsertResult.version.
• Weighted graph edges (#268) and client-side rate limiting (#262).

Changed

• Parallelized put_vectors across batches for higher ingest throughput (#259).
• Strict mypy now runs in CI (#207).

Fixed

• Graph node metadata preservation (#272, fixes #271) — add_edge() / link_docs() no longer wipe an existing node's ntype / props.
• Markdown front-matter date normalization on ingest (#253); cross-encoder reranker type fix (#254).

pip install --upgrade dynavec

Full Changelog: https://github.com/codeforstartups/dynavec/blob/development/CHANGELOG.md

dynavec 0.6.0 — a big feature release

Retrieval

• Query-expansion retrievers (#215) — MultiQuery + HyDE, RRF-fused
• Learned RRF fusion weights (#204, #223) — grid/random/Bayesian + save/load
• Cross-encoder reranking (#208)

Quantization

• OPQ (rotated PQ) (#198) and INT8 ScalarQuantizer (#201)

Graph

• Hybrid graph + ANN search (#206), shortest path (#237), node/edge deletion (#209)

Cache & observability

• Cache invalidation on write (#236), warm_cache() (#194), embedding cache (#213)
• search().explain() debug output (#228)

Ingestion & storage

• S3Source (#61), CsvSource + row-wise spreadsheets (#205), DynamoDB gzip (#231), item-size validation (#211)

Integrations & CLI

• OpenAI Assistants tool (#73), LlamaIndex metadata-filter translation (#219), Strands example
• CLI namespace export/import (#217), describe() (#218), --version (#238)

Embeddings & config

• OpenAI retry with Retry-After (#242), dimension>4096 warning (#241), eager max_workers validation (#250)

Upgrade: pip install --upgrade dynavec / uv pip install --upgrade dynavec

Full changelog: https://github.com/codeforstartups/dynavec/blob/development/CHANGELOG.md

dynavec 0.5.0

Added

• Office document ingestion (#188) — DocxSource, PptxSource, XlsxSource.
• Hugging Face Inference embedder (#191) — HFInferenceEmbedder.
• DSPy retrieval integration (#195) — DynavecRM(dspy.Retrieve), closes #68.
• Structured logging (#200) — opt-in JSON store events with secret redaction.
• ProductQuantizer persistence (#199) — save()/load() via safe np.savez.
• Dashboard dark mode (#187).

Performance

• Vectorized MMR reranking (#197) — O(k·N) instead of O(k·N·k).

Tests

• Property-based metric tests via Hypothesis (#192); rescore override coverage (#196).

Upgrade: pip install --upgrade dynavec / uv pip install --upgrade dynavec

Full changelog: https://github.com/codeforstartups/dynavec/blob/development/CHANGELOG.md

dynavec 0.4.0 — in-memory hot tier

Added

• In-memory hot tier (#186) — hot_tier=True + db.warm(namespace) serve a namespace entirely from RAM (no S3 Vectors query, no DynamoDB hydration) for in-memory-engine latency without a paid cluster. Write-through keeps it current; filters/rescore/MMR run on the hot path; non-authoritative namespaces fall back to S3 safely. New: warm(), hot_stats(), hot_tier* config.
• Retrieval quality runner (#185) — recall@k, MRR, nDCG@k.
• Async LangChain retrieval (#183) — asimilarity_search and friends (closes #70).
• Graph export (#182) to Mermaid / Graphviz DOT.
• Ollama embedder (#180) and URL (#181) + Markdown (#174) ingestion sources.
• DynamoDB cache TTL jitter (#145).

Upgrade: pip install --upgrade dynavec / uv pip install --upgrade dynavec

Full changelog: https://github.com/codeforstartups/dynavec/blob/development/CHANGELOG.md

dynavec 0.3.0

Added

• Async embeddings — aembed_documents / aembed_query on every backend (#172)
• New embedders — Mistral (#152), Voyage AI (#149), Bedrock Titan multimodal image embeddings (#173)
• SPFresh hot-tier incremental rebalancing (#170)
• PDF ingestion source (#169)
• FastMCP server for semantic + graph search (#165)
• Observability — native telemetry recorder + stdlib dashboard, plus a Next.js/TypeScript/Tailwind/Recharts dashboard under dashboard/
• max_pool_connections tuning across all boto3 clients (#175)
• list_vectors maintenance iterator (#166), optional score normalization (#147), cache hits/misses + stats() (#121), AWS doctor command (#144), chunk dedup (#120), py.typed marker (#154)

Changed

• Semantic cache bounded by bytes rather than entry count (#146)

Fixed

• Escape structured-storage key components (#140), drain QueryVectors pages + page_size (#151), pin crewai off yanked 1.14.0 (#161), Python 3.9 optional deps (#1)

Full changelog: https://github.com/codeforstartups/dynavec/blob/development/CHANGELOG.md

First public release of dynavec — serverless hybrid vector database on AWS (DynamoDB + Amazon S3 Vectors). Pluggable embedders, metrics/rerank, namespaces, knowledge graph, caching, MCP ingestion, framework adapters. Published to PyPI via Trusted Publishing.