Migration Guide: From LeMUR to LLM Gateway

Overview

This guide helps you migrate from LeMUR to AssemblyAI’s LLM Gateway. While LeMUR was designed specifically for processing transcripts, LLM Gateway provides a more flexible, unified interface for working with multiple language models.

Key differences

Migration steps

Step 1: Replace transcript processing

Before (LeMUR): LeMUR automatically retrieved transcript text using transcript IDs. It only used the basic text field — not speaker labels, utterances, or other structured data from the transcript.

After (LLM Gateway): You include the transcript text directly in your request. Since LeMUR only used the text field, migrating to LLM Gateway produces equivalent results.

1
# LeMUR automatically fetched transcript content
2
result = transcript.lemur.task(
3
    prompt="What was the emotional sentiment of the phone call?",
4
    final_model=aai.LemurModel.claude_sonnet_4_20250514
5
)

Step 2: Update model names

LLM Gateway uses different model identifiers:

Step 3: Modify response handling

Before: LeMUR returned a simple response object. After: LLM Gateway returns OpenAI-compatible response format.

1
result = transcript.lemur.task(prompt)
2
print(result.response)  # Direct access to response text

Complete migration example

Here’s a complete example showing how to migrate a LeMUR sentiment analysis task:

1
import requests
2
import time
3
4
base_url = "https://api.assemblyai.com"
5
headers = {"authorization": "<YOUR_API_KEY>"}
6
7
# Step 1: Transcribe audio
8
audio_data = {"audio_url": "https://assembly.ai/call.mp4"}
9
transcript_response = requests.post(f"{base_url}/v2/transcript", headers=headers, json=audio_data)
10
transcript_id = transcript_response.json()["id"]
11
12
# Poll for completion
13
while True:
14
    status_response = requests.get(f"{base_url}/v2/transcript/{transcript_id}", headers=headers)
15
    status = status_response.json()["status"]
16
    if status == "completed":
17
        break
18
    elif status == "error":
19
        raise RuntimeError("Transcription failed")
20
    time.sleep(3)
21
22
# Step 2: Apply LeMUR
23
lemur_data = {
24
    "prompt": "What was the emotional sentiment of the phone call?",
25
    "transcript_ids": [transcript_id],
26
    "final_model": "anthropic/claude-sonnet-4-20250514"
27
}
28
29
result = requests.post(f"{base_url}/lemur/v3/generate/task", headers=headers, json=lemur_data)
30
print(result.json()["response"])

Migration benefits

Moving to LLM Gateway provides several advantages:

More model choices

• 15+ models including Claude 4.5 Sonnet, GPT-5, and Gemini 2.5 Pro
• Better performance with newer, more capable models

Flexible input handling

• Multi-turn conversations for complex interactions
• Custom system prompts for better context control

Enhanced capabilities

• Tool calling for function execution
• Agentic workflows for multi-step reasoning
• OpenAI-compatible API for easier integration

Next steps

After migrating to LLM Gateway, explore additional capabilities:

• Multi-turn Conversations - Build conversational experiences
• Tool Calling - Enable function execution
• Agentic Workflows - Create multi-step reasoning

Need help?

If you encounter issues during migration:

1. Check model availability - Ensure your chosen model is supported
2. Verify API endpoints - LLM Gateway uses different URLs than LeMUR
3. Update response parsing - Response format follows OpenAI standards
4. Review token limits - Different models have different context windows