Examples

This section contains practical, copy-paste-ready examples for common LM Deluge workflows. Each example is tested against the current codebase.

Quick Reference

Example	What you’ll learn
Chat Loops	Build interactive multi-turn conversations
Streaming	Stream responses token-by-token (OpenAI)
Batch Processing	Cost-effective batch API processing
Computer Use	Claude Computer Use with Anthropic and Bedrock

Basic Patterns

Simple Request

import asyncio
from lm_deluge import LLMClient, Conversation

async def main():
    client = LLMClient("claude-4.5-haiku", max_new_tokens=1024)
    response = await client.start(Conversation().user("Hello!"))
    print(response.completion)

asyncio.run(main())

With Tools

import asyncio
from lm_deluge import LLMClient, Conversation, Tool

def add(a: int, b: int) -> int:
    """Add two numbers."""
    return a + b

async def main():
    client = LLMClient("gpt-4o-mini")
    tools = [Tool.from_function(add)]

    conv = Conversation().user("What is 2 + 3?")
    conv, response = await client.run_agent_loop(conv, tools=tools)
    print(response.completion)

asyncio.run(main())

Structured Output

from pydantic import BaseModel
from lm_deluge import LLMClient

class Person(BaseModel):
    name: str
    age: int

client = LLMClient("gpt-4o-mini")
response = client.process_prompts_sync(
    ["Extract: John is 25 years old"],
    output_schema=Person,
)[0]
print(response.completion)  # {"name": "John", "age": 25}

What’s Where

Looking for something specific? Here’s where to find it:

• Tool creation & agent loops: Tool Use and Building Agents
• MCP servers: MCP Integration
• Prompt caching: Caching
• File uploads: Working with Files
• Structured outputs: Structured Outputs
• Rate limiting: Rate Limiting