菜单
Zvec in-process vector database. Covers collections, indexing, embeddings, reranking, and persistence. Use when embedding Zvec into applications or tuning retrieval/storage behavior. Keywords: Zvec, HNSW-RaBitQ, vector database, ANN.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/itechmeat/llm-code --skill zvec命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
| name | zvec |
| description | Zvec in-process vector database. Covers collections, indexing, embeddings, reranking, and persistence. Use when embedding Zvec into applications or tuning retrieval/storage behavior. Keywords: Zvec, HNSW-RaBitQ, vector database, ANN. |
| metadata | {"version":"0.6.0","release_date":"2026-07-20"} |
Zvec is a lightweight, in-process vector database meant to be embedded into applications ("SQLite for vectors").
references/overview.mdreferences/concepts.mdreferences/quickstart.mdreferences/installation.mdreferences/indexing.mdreferences/embedding.mdreferences/reranker.mdreferences/collections.mdreferences/data-operations.mdreferences/configuration.mdMinimal “embed Zvec” checklist
zvec.init(...) (logging, query_threads).create_and_open(path=..., schema=..., option=...).Doc(id=..., fields=..., vectors=...) via insert() or upsert().collection.query(vectors=VectorQuery(...), topk=...).collection.optimize() periodically after heavy ingestion.Bulk ingest + keep query latency stable
insert() / upsert().collection.stats and run optimize() when flat buffers grow.Hybrid retrieval patterns
collection.query(filter=..., topk=...).vectors=... and filter=....VectorQuery items and rerank using WeightedReRanker or RRF.Memory-sensitive ANN on x86_64
HNSW-RaBitQ when HNSW-quality recall matters but memory is the limiting factor.total_bits=7, num_clusters=16) and tune query-time ef before changing quantization bits.Safe evolution of live collections
add_column(), drop_column(), alter_column().create_index() / drop_index() (scalar). Vector indexes cannot be dropped.HNSW-RaBitQ on unsupported hardware; current docs limit it to x86_64 with AVX2 or better.Fp32Quantizer implementation and scalar FP32 distance kernels.enable_rotate) spreads variance evenly across dimensions before quantizing, cutting quantization error. On the cohere-1m benchmark this took HNSW INT8 recall from 0.9285 to 0.9397, Flat INT8 from 0.9695 to 0.9881, and HNSW INT4 recall from 0.2114 to 0.7117 — a large jump that makes INT4 viable in more scenarios.fetch_vector, is_linear, and bf_pks query modes), exposed through the Python API.VectorViewClause directly at the source numpy buffer instead of memcpy-ing through serialize_vector, removing redundant copies for dense vector queries.stemmer_lang) reduces words to their root form.VectorViewClause establishes a zero-copy vector query path with unified query validation, lowering per-query memory overhead.PO, PL) directly through search parameters for finer performance control.is_dirty is exposed on the core Index interface, and copy-on-write MMAP options (with a corrected MMAP_POPULATE placement) are configurable.create_index() / drop_index() and query it with natural-language or structured expressions, alongside vector indexes.MultiQuery API combines dense vectors, sparse vectors, scalar filters, and text in one query with consistent reranking across Python, Go, Rust, and C++.fetch() accepts an output_fields parameter to control which fields are returned.pip install zvec-studio) for visual data browsing and query testing.topK ceiling, stricter query_params validation, and fixes an SQ8 quantizer recall regression.