Mirascope and Langfuse Integration

Langfuse Integration

We’re thrilled to announce our integration with the LLM Engineering Platform Langfuse. LLM Development without any observability tool is a dangerous endeavor, so we at Mirascope have made it easy for you to slot in Langfuse into your workflow.

Mirascope and Langfuse

Mirascope integrates with Langfuse with a decorator @with_langfuse. This gives you full control on what you want to observe and what you don’t.

Call

This is a basic call example that will work across all Mirascope call functions such as stream , call_async , and stream_async .

1import os
2from mirascope.langfuse import with_langfuse
3from mirascope.anthropic import AnthropicCall
4
5os.environ["LANGFUSE_PUBLIC_KEY"] = "pk-lf-..."
6os.environ["LANGFUSE_SECRET_KEY"] = "sk-lf-..."
7os.environ["LANGFUSE_HOST"] = "https://cloud.langfuse.com"
8
9
10@with_langfuse
11class BookRecommender(AnthropicCall):
12    prompt_template = "Please recommend some {genre} books"
13
14    genre: str
15
16
17recommender = BookRecommender(genre="fantasy")
18response = recommender.call()  # this will automatically get logged with langfuse
19print(response.content)

‍

And that’s it! Now your Mirascope class methods will be sent to Langfuse traces and calls will be sent to Langfuse generation. Mirascope automatically sends tags, latency, cost, and other metadata alongside inputs and outputs.

Extract

You can also take advantage of Mirascope extractors and observe those in Langfuse. Extractors allow you to efficiently pull out relevant information from your data. By monitoring these extractors in Langfuse, you can assess the quality of your data and ensure your results are accurate.

1import os
2from typing import Literal, Type
3
4from pydantic import BaseModel
5
6from mirascope.anthropic import AnthropicExtractor
7from mirascope.langfuse import with_langfuse
8
9os.environ["LANGFUSE_PUBLIC_KEY"] = "pk-lf-..."
10os.environ["LANGFUSE_SECRET_KEY"] = "sk-lf-..."
11os.environ["LANGFUSE_HOST"] = "https://cloud.langfuse.com"
12
13
14class TaskDetails(BaseModel):
15    description: str
16    due_date: str
17    priority: Literal["low", "normal", "high"]
18
19
20@with_langfuse
21class TaskExtractor(AnthropicExtractor[TaskDetails]):
22    extract_schema: Type[TaskDetails] = TaskDetails
23    prompt_template = """
24    Extract the task details from the following task:
25    {task}
26    """
27
28    task: str
29
30
31task = "Submit quarterly report by next Friday. Task is high priority."
32task_details = TaskExtractor(
33    task=task
34).extract()  # this will be logged automatically with langfuse
35assert isinstance(task_details, TaskDetails)
36print(task_details)

‍

Head over to Langfuse and you can view your results!