Milvus Marketplace
A data retrieval development assistant based on Claude Code Skills.
Core Philosophy
Focus: Data Retrieval Domain
We specialize in the data retrieval vertical:
- Vector search, semantic search
- RAG, knowledge base Q&A
- Image search, recommendation systems
- Milvus / Zilliz Cloud
Two-Layer Solution Architecture
┌─────────────────────────────────────────────────────────┐
│ Scenario Plugins (6 plugins) │
│ rag-toolkit, retrieval-system, multimodal-retrieval, │
│ rec-system, memory-system, data-analytics │
│ = Pre-built solutions = AI era caching mechanism │
└─────────────────────────────────────────────────────────┘
↑
Match / Combine
↑
┌─────────────────────────────────────────────────────────┐
│ core plugin │
│ Methodology (pilot) + Atomic operators │
│ (embedding, chunking, ...) │
└─────────────────────────────────────────────────────────┘
Core Ideas:
- Scenario plugins = AI cache: Pre-built scenario solutions act like cache - use directly if matched, no need to build from scratch
- core = methodology + operators: Universal development methodology + composable atomic capabilities
- Ship 60% first, iterate to 80%: Deliver working solutions quickly, then optimize based on feedback
Scenario Classification Principles
Scenarios are classified by architectural differences, not by industry or model:
- Different architecture = different scenario: Fundamentally different processes and code structures
- Only model/parameters differ = vertical applications within same scenario: Covered through configuration tables
plugins/retrieval-system/skills/
├── semantic-search/ # Category: architecture definition
│ ├── SKILL.md # Generic workflow + model selection table
│ └── verticals/ # Subcategory: vertical application guides
│ ├── legal.md # Legal search
│ ├── academic.md # Academic papers
│ └── ecommerce.md # E-commerce search
Workflow
User describes requirement
│
▼
pilot activates
│
├─→ Clarify data and query
│
├─→ Can match a scenario?
│ ├─ Yes → Use pre-built solution
│ └─ No → Combine core operators
│
├─→ Generate code → User tests
│
└─→ Collect feedback → Iterate
Installation
1. Add Marketplace
/plugin marketplace add zilliztech/milvus-marketplace
2. Install Plugins
# Core tools (required)
/plugin install core@milvus-marketplace
# Install scenario plugins as needed
/plugin install rag-toolkit@milvus-marketplace # RAG solutions
/plugin install retrieval-system@milvus-marketplace # Text search
/plugin install multimodal-retrieval@milvus-marketplace # Image/video/multimodal
/plugin install rec-system@milvus-marketplace # Recommendations
/plugin install memory-system@milvus-marketplace # Chat memory
/plugin install data-analytics@milvus-marketplace # Duplicate detection, clustering
3. Start Using
Simply describe what you want to build:
"Help me build a document Q&A system"
"I want to implement semantic search"
"Build an image search application"
The pilot will automatically activate, clarify requirements, and help you orchestrate the toolchain and generate code.
Plugin Overview
core - Core Capabilities
| Type | Skill | Purpose |
|---|
| Controller | pilot | AI application navigator - understands requirements, orchestrates tools, delivers code |
| Operator | embedding | Text/image vectorization |
| Operator | chunking | Document chunking |
| Operator | indexing | Milvus index management |
| Operator | data-ingestion | Batch data import |
| Operator | rerank | Search result reranking |
| Operator | pdf-extract | PDF text extraction |
| Operator | vlm-caption | Image captioning (VLM) |
| Environment | local-setup | Local Milvus deployment |
Scenario Plugins - Pre-built Solutions (6 Plugins, 17 Skills)
retrieval-system - Text Search
| Skill | Architecture | Vertical Applications |
|---|
| semantic-search | embedding → vector search | Legal, academic, news, e-commerce, code, patents |
| hybrid-search | vector + BM25 keyword + score fusion | E-commerce, legal, academic |
| filtered-search | vector search + scalar filtering | E-commerce, recruitment, real estate |
| multi-vector-search | multi-vector field joint search | Products, papers, resumes |
rag-toolkit - RAG / Q&A
