~/wiki
Récent+ Ajouter

API Wrapper vs RAG Architectures

Mis à jour le 2026-06-11Confiance : high
api-wrapperrag-architecturesemantic-searchkeyword-searchai-system-designinformation-retrieval

Fundamental architectural distinction in AI information systems between simple API passthrough and intelligent retrieval-augmented generation approaches. Critical decision point determining system capabilities, user experience, and competitive positioning.

API Wrapper Architecture

Definition: Direct passthrough layer that translates LLM requests into structured API calls and returns raw results.

Characteristics:

  • Minimal processing of queries or results
  • Dependency on upstream API search quality
  • Structured parameter requirements
  • Raw data delivery to LLM clients

Example Flow:

LLM → "contrat de travail CDD"
    ↓
Wrapper → API Call: search="CDD", code="Code du travail", field="ARTICLE"
    ↓
Raw API Results → Full articles dumped to LLM context

Advantages:

  • Simple implementation and maintenance
  • Direct access to authoritative sources
  • Lower computational requirements
  • Faster development cycle

Limitations:

  • Poor search quality for natural language queries
  • No semantic understanding or relevance ranking
  • Raw content dumps with noise and irrelevant information
  • Requires LLM to know data structure and query syntax

RAG (Retrieval-Augmented Generation) Architecture

Definition: Intelligent information retrieval system using semantic search, relevance ranking, and context optimization.

Characteristics:

  • Natural language query processing
  • Semantic vector search with embeddings
  • Multi-source data fusion and ranking
  • Filtered, relevant passage delivery

Example Flow:

LLM → "What are employee rights when a CDD ends?"
    ↓
RAG System → Query understanding + expansion
           → Vector search across multiple sources
           → Passage-level relevance scoring
           → Cross-source ranking and deduplication
    ↓
Top 5 Most Relevant Passages → Delivered to LLM

Advantages:

  • Superior search quality for natural queries
  • Intelligent content filtering and ranking
  • Cross-source knowledge synthesis
  • Optimized context delivery

Disadvantages:

  • Complex implementation requiring ML pipeline
  • Higher computational and infrastructure costs
  • Longer development and tuning cycles
  • Dependency on embedding model quality

API Wrapper Dominance: All existing French legal MCP servers follow wrapper pattern:

  • openlegi: Legifrance PISTE API wrapper
  • mcp-vosdroits: Service-Public.gouv.fr wrapper
  • data.gouv.fr MCP: Dataset API wrapper

Quality Comparison Example: "fin CDD droits salarié"

System Results Quality
OpenLegi 1 article (CSE database - irrelevant) Poor
mcp-vosdroits 5 URLs (empty descriptions) + noisy content Fair
RAG Approach Top 5 relevant passages from CDD termination law Excellent

Strategic Implications

When API Wrappers Sufficient:

  • Simple, well-structured queries
  • Users familiar with data source organization
  • Speed prioritized over quality
  • Limited development resources

When RAG Architecture Required:

  • Natural language query support needed
  • Quality and relevance critical
  • Cross-source knowledge synthesis required
  • Competitive differentiation through superior UX

Market Opportunity: In domains dominated by API wrappers, RAG architectures can provide substantial competitive advantage through superior user experience and result quality.

Implementation Considerations

Technical Complexity:

  • API Wrapper: Days to weeks
  • RAG System: Months with specialized ML expertise

Infrastructure Requirements:

  • API Wrapper: Simple web service
  • RAG System: Vector databases, embedding models, ML pipeline

Maintenance Overhead:

  • API Wrapper: Minimal (mostly upstream API changes)
  • RAG System: Continuous model tuning and data pipeline maintenance

See also