zvec-ai

⚡ Lightning-fast, lightweight embedded vector DB.
🚀 High performance made simple.
- Github: Discord:

[zvec-ai @ZvecAI]

🚀 zg (zvec-grep) is now open source!

We built a local search tool for people and AI agents—
indexing and search run on-device by default, combining natural-language understanding with the speed and precision of rg.

🔒 Local-first: File processing, indexing, and search stay on-device by default, keeping your code and documentation private and secure.

⚡️ Ready in three steps: Install zg, index your workspace, and connect your agent. zg automatically configures supported agents and runs across macOS, Linux, and Windows.

🔎 Comprehensive search: Semantic search, BM25, hybrid retrieval, and rg work together across code, documentation, and data.

🤖 More efficient agents: On SWE-QA-Bench, tool calls and input tokens were both reduced by roughly half, while the judge score improved by 1.50 points.

GitHub: https://t.co/42c050Kqql
Blog: https://t.co/sZUOa8g98y

zg is still evolving. Try it, star the repo, and share your feedback—every suggestion helps make it better. ⭐️

[zvec-ai @ZvecAI]

Zvec v0.7.0 is here! 🎉
This release expands the Zvec ecosystem and brings major upgrades to indexing, deployment, and usability. Upgrade today and give it a try! ✨

What’s new
🧭 zvec-grep (https://t.co/42c050KYfT)
A local-first workspace search CLI for humans and AI agents. It combines semantic search, BM25, and rg to find code and documents faster—with fewer search rounds, tokens, and less time spent.

🔗 ReMe integration
Zvec can now serve as a file-store backend for ReMe, providing in-process HNSW ANN search for agent memory management.

🚀 Production-ready DiskANN
Now supports Linux ARM64, macOS ARM64, and io_uring asynchronous I/O—with automatic fallback to the best available backend.

⚡ Upgraded index algorithms
Introduces the new IVF-RaBitQ index. RaBitQ now supports runtime AVX2/AVX512 dispatch, automatically selecting the optimal path for each CPU.

📦 Better deployment experience
macOS ARM64 C API library reduced from 37 MB to 22 MB—40% smaller
New musl libc and Alpine Linux support
Prebuilt SDK binaries for Linux, macOS, Windows, Android, and iOS with every release

🛠️ Improved usability
New DocIterator for streaming full-collection document traversal
New N-gram tokenizer for phrase, code, and short-text search

📚 Learn more
https://t.co/SkgIkIZJVA

We’d love to hear your feedback! 🚀

[zvec-ai @ZvecAI]

🚀 Zvec × DiskANN: Scale to Hundreds of Millions of Vectors Without Burning Through RAM

Traditional in-memory vector indexes become extremely expensive at scale, often forcing teams to upgrade to high-end servers and undermining the promise of lightweight deployment.

✅ Lower memory costs: Zvec DiskANN keeps only PQ-compressed codes in memory, while raw vectors and the graph structure live on SSD—reducing memory usage by an order of magnitude.

⚡ Higher performance: On the Cohere dataset, Zvec DiskANN delivers 1.5×–2.4× higher single-thread query throughput than Microsoft’s open-source implementation, while accelerating index construction by 36%–54%.

💡 With SSD storage, asynchronous I/O, and tiered caching, massive vector search can stay both cost-efficient and high-throughput.

Struggling with large-scale vector storage and low-latency retrieval? Let’s explore how Zvec DiskANN can help break the memory bottleneck.

📖 Read the full article:
https://t.co/KEeosqHcP1

[zvec-ai @ZvecAI]

🎉 Zvec 0.6.0 Release is Live!

This release brings major upgrades across retrieval, compute optimization, deployment experience, and the quantization framework. Upgrade and give it a try!

✨What's new:

• 🔍Retrieval: New Group-By search with per-group Top-K support, covering Flat, HNSW, and HNSW-RaBitQ indexes

• 📝Full-Text Search: Introduces a UAX #29 standard tokenizer, UTF-8 support, ASCII folding, and Snowball stemming for 34+ languages—a major upgrade for multilingual retrieval; Conjunction queries are now 22–38% faster with block-max pruning

• ⚡ Compute Optimization: INT8/INT4 quantization now supports random rotation, evening out the distribution to reduce variance and quantization error; INT4 recall improves by up to 50+ percentage points while QPS remains essentially unchanged

• 💾 Deployment Experience: DiskANN and libaio are now dynamically decoupled via dlopen, removing the libaio build dependency and plugin .so; libaio is automatically detected at runtime, with a graceful fallback when unavailable

• 🧩 Quantization Framework: The internal quantization module has been refactored into a pluggable Turbo framework, decoupling quantization logic from index code; future quantizers such as PQ and RaBitQ can be added independently without modifying index internals

📚 Learn more:

• 📄 Release Notes: https://t.co/1WZhVJVhQJ
• 🧭 Roadmap: https://t.co/VFkhEwBIIR

[zvec-ai @ZvecAI]

🚀 New blog post: Zvec now has native full-text search

With v0.5.0, Zvec can handle keyword recall, BM25 scoring, phrase queries, Boolean queries, and FTS × vector hybrid search inside the same embedded vector database.

No separate Elasticsearch sidecar.
No dual writes.
No extra index sync.

We also shared benchmarks against Elasticsearch and SQLite FTS5, plus a deep dive into how FTS fits into Zvec’s Segment architecture.

👉 Read the post:
https://t.co/DxRC2FMN4R

[zvec-ai @ZvecAI]

📱What if your phone could search its own memory?

To explore Zvec Flutter SDK on mobile, we built PocketSearch: an on-device search app that starts with intelligent local photo search on Android and iOS. It shows how Zvec can power fast, local retrieval today, and points toward richer mobile context for users and agents next.

If you're interested in on-device retrieval or mobile AI context, follow along.

🔗 Blog: https://t.co/rE7NjCw0KP

[zvec-ai @ZvecAI]

🎉 Zvec 0.5.0 Release is Live!

This release focuses on retrieval, indexing, ecosystem, and platform & hardware. Give it a try and let us know what you think! ✨

✨ What's new:

• 🔍 Retrieval: Native full-text search works out of the box; a single query fuses full-text search, semantic vectors, and conditional filters, with multi-way result merging handled automatically; the MultiQuery interface is implemented natively in C/C++, so every language SDK can integrate hybrid retrieval with ease

• 💾 On-Disk Indexing: New DiskANN on-disk index significantly reduces memory overhead in large-scale scenarios; it complements HNSW/IVF/Flat to cover the full range of retrieval needs, from pure in-memory to on-disk

• 🌐 Ecosystem: Brand-new official Go SDK and Rust SDK for smoother multi-language access; plus the new visual tool Zvec Studio for zero-code data browsing and query debugging

• 🖥️ Platform & Hardware: Added RISC-V support, further expanding hardware compatibility

📚 Learn more:

• 📄 Release Notes: https://t.co/en1SWObUV5
• 🧭 Roadmap: https://t.co/VFkhEwBIIR
• 🐛 Report bugs / requests: https://t.co/toHNgRd0Up
• 💬 Discuss & share: https://t.co/0abR6Fj9wi

[zvec-ai @ZvecAI]

A nice write-up on what local-first vector search can actually look like on desktop apps.

This post uses Zvec + Obsidian as an example to show:

1. local semantic search
2. image + text retrieval
3. related context surfacing while writing
4. fully on-device retrieval pipelines

Feels like we're moving from “vector databases in the cloud” toward “vector search as a native capability inside apps”.

Worth a read:
https://t.co/upAvMOaYuV

[zvec-ai @ZvecAI]

🎉 Zvec v0.4.0 is Live!

This release focuses on mobile coverage, search capabilities, and stability. Give it a try and let us know what you think! ✨

What's new:

• 📱 Mobile Coverage: Official Dart/Flutter SDK for Android (arm64-v8a) and iOS (arm64)

• 🔍 Retrieval Enhancement: Enlarged topK limit to support large-scale candidate recall for downstream re-ranking pipelines

• 🎯 Quantization Accuracy: Fixed SQ8 quantizer recall drop caused by missing int8 rounding in metadata computation

• 🖥️ Cross-platform Stability: Fixed Windows drive-root path handling and improved OS error reporting; relaxed collection path validation

For more details, please read the release blog:
https://t.co/1653eaVeSb

We also shared a deep dive into the engineering behind our lightweight core:
https://t.co/qd4PlLUGs4

[zvec-ai @ZvecAI]

Two ways to use Zvec with your AI Coding Agent:

1️⃣ MCP Server — Your agent operates the vector DB directly via natural language. No code needed.
https://t.co/wTuXuYqSRk

2️⃣ Agent Skill — Your agent writes Zvec code (Python/Node.js) for you.
Just describe your use case (RAG, e-commerce search, etc.) and get working code with schema design.
https://t.co/PyhzxwEa3z

More native AI support is on the way! 🚀

[zvec-ai @ZvecAI]

🎉 Zvec v0.3.0 is Live!

This release brings full multi-platform support for Windows, macOS, and Linux, along with official SDKs for Python, Node.js, and C.

We've integrated the RabitQ quantization algorithm and added batch optimization for IP/L2 distance calculations on x86. We have also added MCP and Skill framework support for AI Agents.

⚡Performance and simplicity are always our top priorities.

For more details, please read the blog post: https://t.co/uVf6RmiAM8