llm-integration
Integrate LLMs into applications - APIs, prompting, fine-tuning, and context management
When & Why to Use This Skill
The LLM Integration skill empowers developers to seamlessly incorporate production-grade Large Language Models (LLMs) into their applications. It simplifies the complexities of connecting to major providers like Anthropic, OpenAI, and Google, while providing robust frameworks for prompt engineering, context management, and cost optimization. By streamlining API interactions and implementing best practices like streaming and exponential backoff, it ensures high-performance and reliable AI-driven functionality.
Use Cases
- Multi-Provider Integration: Quickly connect and switch between different LLM APIs (Claude, OpenAI, Google) to find the best model for specific application needs.
- Responsive AI Interfaces: Implement streaming responses to create low-latency, real-time chat experiences that improve user engagement.
- Cost-Effective Scaling: Optimize operational expenses by managing token usage and selecting the most efficient models (e.g., Claude Haiku vs. Opus) based on task complexity.
- Production-Grade Reliability: Build resilient systems with built-in troubleshooting for common API challenges such as rate limits, context window overflows, and output quality control.
| name | llm-integration |
|---|---|
| description | Integrate LLMs into applications - APIs, prompting, fine-tuning, and context management |
| sasmp_version | "1.3.0" |
| bonded_agent | 02-llm-integration |
| bond_type | PRIMARY_BOND |
| version | "2.0.0" |
LLM Integration
Integrate Large Language Models with production-grade reliability.
When to Use This Skill
Invoke this skill when:
- Connecting to Claude, OpenAI, or other LLM APIs
- Designing effective prompts and system messages
- Optimizing token usage and costs
- Implementing streaming responses
Parameter Schema
| Parameter | Type | Required | Description | Default |
|---|---|---|---|---|
provider |
enum | Yes | anthropic, openai, google, local |
- |
task |
string | Yes | Integration goal | - |
streaming |
bool | No | Enable streaming | true |
max_tokens |
int | No | Response token limit | 4096 |
Quick Start
# Anthropic Claude
from anthropic import Anthropic
client = Anthropic()
response = client.messages.create(
model="claude-sonnet-4-20250514",
max_tokens=1024,
messages=[{"role": "user", "content": "Hello!"}]
)
# OpenAI
from openai import OpenAI
client = OpenAI()
response = client.chat.completions.create(
model="gpt-4-turbo",
messages=[{"role": "user", "content": "Hello!"}]
)
Prompt Templates
System Prompt
SYSTEM = """You are {role}, an expert in {domain}.
Your task: {task}
Constraints: {constraints}
Output format: {format}"""
Chain-of-Thought
COT = """Think step by step:
1. Understand the problem
2. Break it down
3. Solve each part
4. Combine results"""
Cost Optimization
| Model | Input $/1M | Output $/1M | Best For |
|---|---|---|---|
| Claude Haiku | $0.25 | $1.25 | High volume |
| Claude Sonnet | $3 | $15 | Complex tasks |
| Claude Opus | $15 | $75 | Most demanding |
Troubleshooting
| Issue | Solution |
|---|---|
| 429 Rate Limited | Exponential backoff |
| Context overflow | Truncate/summarize |
| Poor output quality | Add examples, lower temp |
| High costs | Use cheaper model, cache |
Best Practices
- Always implement retry with backoff
- Use streaming for better UX
- Cache repeated queries
- Monitor token usage
Related Skills
ai-agent-basics- Agent architecturerag-systems- Retrieval augmentationtool-calling- Function calling