Vector Index Tuning
Guides an agent through tuning vector search indexes for speed, recall, and memory use
- Repository
- wshobson/agents
- Stars
- 42,095
- Licence
- MIT
- Files
- 2 files
What it does
Vector search lets software find items by meaning, using lists of numbers called embeddings, and its speed and accuracy depend on how the index is built. With this skill installed, the agent follows a guide on choosing the right index type for a data size, setting HNSW parameters like M, efConstruction and efSearch, and picking quantization methods that shrink memory use. It can also point you to worked templates in a reference file when you ask for concrete examples.
Great for
- Engineers running a production vector search or RAG system who need faster queries or lower memory costs
- Teams deciding between Flat, HNSW, IVF+PQ or DiskANN indexes as their dataset grows
- Developers who want a quick reference for what each HNSW or quantization setting trades off
Skip it if
- Beginners without a vector database or embeddings already in place, since the guidance assumes existing infrastructure
- Exact, project-specific tuning, because the advice is general rules of thumb and the agent cannot measure your real recall or latency from these files alone
Try asking
- “My product search has 3 million embeddings and queries are slow. Which index type and HNSW settings should I try?”
- “Help me pick a quantization strategy to cut the memory footprint of my 768-dimension vectors.”
- “Give me a template for tuning efSearch and measuring recall against a benchmark set.”
Install in 3 steps
- 1
Download the skill
The ZIP contains the 2 original files, the MIT licence and an attribution note.
vector-index-tuning.zip - 2
Add it to your agent
Claude Code (all projects)
mkdir -p ~/.claude/skills && unzip vector-index-tuning.zip -d ~/.claude/skills/Claude Code / Agent SDK (this project only)
mkdir -p .claude/skills && unzip vector-index-tuning.zip -d .claude/skills/Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.
- 3
Ask for the task
No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.
Get better results
- Tell the agent your vector count, dimensions, database, and latency or recall targets so it can recommend a specific index type.
- Ask for the templates in references/details.md when you need concrete code, since the main guide only summarizes them.
- Benchmark changes with your own real queries, as the skill itself advises, rather than trusting its defaults blindly.
What's inside
- references/details.md
- SKILL.md
Redistributed unmodified from wshobson/agents under the MIT licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.