RequestyAI Python API library

Deliver AI products in days, not months, using requesty.ai

Step up your AI game with a simple integration:

• Real-time insights into your AI flows: See how your clients use your LLMs
• Hassle free auditing and logging: Don't stress about storage and infra
• Powerful analytics: Discover trends using your Requesty's insight-explorer

Installation

pip install requestyai

AInsights

A client library to effortlessly integrate powerful insights into your AI flows.

Create a client with minimal configuration and capture and request/response pair for future analysis.

Notes

Thread-safety

The insights client is thread-safe. You can safely use a single client from multiple threads.

Asynchronous

The insights client is asynchronous. capture()-ing requests/responses is a non-blocking operation, and it will not interrupt the flow of your application.

Usage pattern #1: Use a global instance

Just create a simple file (ainsights.py is a reasonable name) in your project, and import it everywhere. The client is thread-safe.

Check out this working sample app

Notes:

• Make sure to set the environment variable OPENAI_API_KEY to your OpenAI API key.
• Make sure to set the environment variable REQUESTY_API_KEY to your Requesty API key.

import os
from requestyai import AInsights


ainsights = AInsights.new_client(api_key=os.environ["REQUESTY_API_KEY"])

Then, calling OpenAI and capturing is as easy as:

It doesn't matter if you're using simple free-text outputs, JSON outputs or tools, the insights client will capture everything.

import os
from openai import OpenAI
from .ainsights import ainsights

openai.api_key = os.environ["OPENAI_API_KEY"]


if __name__ == '__main__':
    user_input = input("Ask anything: ")

    messages = [{"role": "user", "content": user_input}]
    args = {"model": "gpt-4o-mini", "temperature": 0.7, "max_tokens": 150}
    response = openai.chat.completions.create(messages=messages, **args)

    ainsights.capture(messages=messages, response=response, args=args)

    print(response.choices[0].message.content)

Usage pattern #2: Create a client instance

If you prefer creating and using client instance, just add an AInsights instance next to your OpenAI one, and capture every interaction by adding a single call.

It doesn't matter if you're using simple free-text outputs, JSON outputs or tools, the insights client will capture everything.

Check out this working sample app

Notes:

• Make sure to set the environment variable OPENAI_API_KEY to your OpenAI API key.
• Make sure to set the environment variable REQUESTY_API_KEY to your Requesty API key.

from openai import OpenAI
from pydantic import BaseModel
from requestyai import AInsights


class Model:
    def __init__(self, openai_api_key, openai_args, requesty_api_key):
        self.__model = OpenAI(api_key=openai_api_key)
        self.__args = openai_args
        self.__insights = AInsights.new_client(api_key=requesty_api_key)

    def chat(self, user_id: str, user_input: str):
        messages = [
            {"role": "system", "content": "You are a helpful search assistant."},
            {"role": "user", "content": user_input}
        ]

        meta = {"class": "Search assistant"}

        class Response(BaseModel):
            answer: str

        response = self.__model.beta.chat.completions.parse(
            messages=messages, response_format=Response, **self.args
        )

        self.__insights.capture(messages=messages,
                                response=response,
                                args=self.args,
                                meta=meta,
                                user_id=user_id)

        content = response.choices[0].message.content
        if not content:
            return None

        return Response.model_validate_json(content)

Meta tagging

If you want to add additional, custom, tags to your model interactions, the capture() method allows you to specify an extra argument, called meta.

You can then use those tags to group and inspect relevant traffic on the UI.

See "Usage pattern #2" above for specific details.

User tracking

If you want your insights to be tied to a specific user, you can specify the user_id argument when calling capture(...).

The user_id is not part of the meta dictionary, because the platform can leverage this specific piece of information to track usage patterns and provide additional insights.

See "Usage pattern #2" above for specific details.

Sample applications

Check out the samples directory for working examples you can try out in no time.

Set up the required virtual Python environment easily using poetry:

poetry install
poetry shell

From inside the virtual environment shell, run the samples using:

REQUESTY_KEY="<YOUR_REQUESTY_KEY>" OPENAI_KEY="<YOUR_OPENAI_KEY>" \
python -m samples.openai.basic.app