API Integration Patterns

Choosing Your Approach

API Best Practices

Streaming

Always use streaming for user-facing applications. It dramatically improves perceived latency:

import anthropic

client = anthropic.Anthropic()

with client.messages.stream(
    model="claude-opus-4-6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Explain RAG"}]
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)

Retry Logic

• Implement exponential backoff for rate limits (429) and server errors (500+)
• Set reasonable timeouts (30-120s depending on task complexity)
• Use circuit breakers for sustained failures

Batching

For non-real-time workloads, use batch APIs (available from Anthropic and OpenAI) for 50% cost savings.

Architecture Pattern: LLM Gateway

Client → API Gateway → LLM Router → Claude API
                                    → OpenAI API (fallback)
                                    → Local Model (sensitive data)
         └→ Cache Layer (Redis)
         └→ Rate Limiter
         └→ Usage Tracker

🌼 Daisy+ in Action: Layered API Architecture

Daisy+ uses a layered API architecture: Odoo's XML-RPC for internal operations, a FastAPI REST layer for external integrations (with Redis caching and model-level invalidation), and MCP for AI-native interactions. API keys protect all external endpoints, and the cache-aside pattern keeps responses fast without stale data. This three-layer approach means every type of client — human, traditional software, or AI agent — has an optimized integration path.