embedding-strategies
- Use it when:
- You're building semantic search or RAG and need to pick an embedding model, split documents into chunks, or measure retrieval quality.
- It does:
- Compares ten embedding models by dimensions, token limit and best use, and gives the agent chunking and preprocessing rules.
- You get:
- Python templates for Voyage AI, OpenAI and local embeddings, four chunkers, document and code pipelines, and a function that scores and compares models.
- Recommends:
- Voyage AI models for Claude apps, as Anthropic does, with domain models for code, finance and legal text.
- Terms:
- An embedding turns text into a list of numbers so similar meanings land close together; chunking splits documents into pieces to embed.
- Unlike rag-implementation:
- That one, from the same repo, covers the whole retrieval system; this one covers only the embedding step.
TL;DR
A reference with templates rather than a workflow. The SKILL.md has a 2026 table of embedding models, from voyage-3-large and OpenAI's text-embedding-3 models to open-source bge-large, all-MiniLM and multilingual-e5, and a pipeline: chunk, preprocess, embed. Its rules: match the model to code, prose or multilingual content, keep chunks on meaning boundaries, normalize vectors for cosine similarity, batch and cache requests, respect token limits, never mix models in one index, and keep metadata. The reference file holds six Python templates: Voyage AI embeddings through LangChain, including the code, finance and law models; OpenAI embeddings in batches of 100 with optional dimension reduction (Matryoshka embeddings, which still work when cut shorter); local BGE and E5 models with the query prefixes they expect; chunking by tokens, sentences, Markdown headers or a recursive splitter; a pipeline that cleans, chunks and embeds documents with metadata, and one that splits code into functions and classes with tree-sitter; and an evaluation that scores retrieval by precision, recall, MRR and NDCG and compares models side by side.
Summarizes the skill at 4bde316.
npx skills add https://github.com/wshobson/agents --skill embedding-strategies skills.sh pageSource on GitHubAll skills from wshobson/agents
- Total Downloads
- 11,800 on skills.sh, Nov 9, 2026
- Repo stars
- 42,334 wshobson/agents
- First seen
- Jan 20, 2026 on skills.sh
Weekly values
| Week to | Installs |
|---|---|
| Sep 21 | 187 |
| Sep 28 | 228 |
| Oct 5 | 285 |
| Oct 12 | 226 |
| Oct 19 | 183 |
| Oct 26 | 187 |
| Nov 2 | 175 |
| Nov 9 | 157 |
| Category |
Build agents, orchestrate workers and give agents memory. Everything for building software: planning, coding, debugging, design, frameworks, cloud, agents and skills. |
|---|---|
| Type |
Knowledge and rules the agent applies while it works: best practices, design principles, API references. |
| Effect |
Adds no side effects of its own. It shapes how the agent thinks, writes or reviews. Guidelines and routers are always effect:none, even when the task they support edits code. |
| Needs |
Instructions only. Works in any agent that supports skills. |
| Publisher |
Everyone else. Unverified vendor claims default to community. |
| Topics | |
| Stack |
SKILL.md
From wshobson/agents at 4bde316. Checked daily, last changed Nov 6, 2026.
Frontmatter
name: embedding-strategies
description: Select and optimize embedding models for semantic search and RAG applications. Use when choosing embedding models, implementing chunking strategies, or optimizing embedding quality for specific domains.Embedding Strategies
Guide to selecting and optimizing embedding models for vector search applications.
When to Use This Skill
- Choosing embedding models for RAG
- Optimizing chunking strategies
- Fine-tuning embeddings for domains
- Comparing embedding model performance
- Reducing embedding dimensions
- Handling multilingual content
Core Concepts
1. Embedding Model Comparison (2026)
| Model | Dimensions | Max Tokens | Best For |
|---|---|---|---|
| voyage-3-large | 1024 | 32000 | Claude apps (Anthropic recommended) |
| voyage-3 | 1024 | 32000 | Claude apps, cost-effective |
| voyage-code-3 | 1024 | 32000 | Code search |
| voyage-finance-2 | 1024 | 32000 | Financial documents |
| voyage-law-2 | 1024 | 32000 | Legal documents |
| text-embedding-3-large | 3072 | 8191 | OpenAI apps, high accuracy |
| text-embedding-3-small | 1536 | 8191 | OpenAI apps, cost-effective |
| bge-large-en-v1.5 | 1024 | 512 | Open source, local deployment |
| all-MiniLM-L6-v2 | 384 | 256 | Fast, lightweight |
| multilingual-e5-large | 1024 | 512 | Multi-language |
2. Embedding Pipeline
Document → Chunking → Preprocessing → Embedding Model → Vector
↓
[Overlap, Size] [Clean, Normalize] [API/Local]
Templates and detailed worked examples
Full template library and detailed worked examples live in references/details.md. Read that file when you need the concrete templates.
Best Practices
Do's
- Match model to use case: Code vs prose vs multilingual
- Chunk thoughtfully: Preserve semantic boundaries
- Normalize embeddings: For cosine similarity search
- Batch requests: More efficient than one-by-one
- Cache embeddings: Avoid recomputing for static content
- Use Voyage AI for Claude apps: Recommended by Anthropic
Don'ts
- Don't ignore token limits: Truncation loses information
- Don't mix embedding models: Incompatible vector spaces
- Don't skip preprocessing: Garbage in, garbage out
- Don't over-chunk: Lose important context
- Don't forget metadata: Essential for filtering and debugging