Skip to main content
actguard patches openai.SyncAPIClient.request and openai.AsyncAPIClient.request — private transport-layer methods that every high-level OpenAI API call passes through.

​
Requirements

RequirementVersion
openai SDK≥ 1.76.0
Python≥ 3.9
Why 1.76.0? The request() signature changed across SDK versions. Prior to 1.76.0 it included a remaining_retries parameter that our replacement does not accept, creating a potential TypeError. Version 1.76.0 removed that parameter, making the signature an exact match for the patched version. Installing an older SDK will trigger a UserWarning at runtime.
Install or upgrade: 
pip install "openai>=1.76.0"

​
What gets patched

openai._base_client.SyncAPIClient.request   → actguard wrapper (sync)
openai._base_client.AsyncAPIClient.request  → actguard wrapper (async)
Every OpenAI API surface (Chat Completions, Responses API, Embeddings, etc.) routes through these methods, so all calls are automatically tracked.

​
Chat completions (non-streaming)

import openai
from actguard import BudgetGuard

client = openai.OpenAI()

with BudgetGuard(user_id="alice", usd_limit=0.10) as guard:
    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[{"role": "user", "content": "Explain async/await in Python."}],
    )
    print(response.choices[0].message.content)

print(f"${guard.usd_used:.6f}  ({guard.tokens_used} tokens)")

​
Chat completions (streaming)

For streaming, actguard automatically injects stream_options={"include_usage": true} into chat completion requests so the final chunk includes usage data. This injection only applies to /chat/completions endpoints and is harmless to existing code. 
with BudgetGuard(user_id="alice", usd_limit=0.10) as guard:
    stream = client.chat.completions.create(
        model="gpt-4o",
        messages=[{"role": "user", "content": "Write a haiku."}],
        stream=True,
    )
    for chunk in stream:
        delta = chunk.choices[0].delta.content or ""
        print(delta, end="", flush=True)

print(f"\n${guard.usd_used:.6f}")

​
Async client

import asyncio
import openai
from actguard import BudgetGuard

client = openai.AsyncOpenAI()

async def main():
    async with BudgetGuard(user_id="alice", usd_limit=0.10) as guard:
        response = await client.chat.completions.create(
            model="gpt-4o",
            messages=[{"role": "user", "content": "Hello!"}],
        )
    print(f"${guard.usd_used:.6f}")

asyncio.run(main())

​
Responses API

The Responses API is also tracked. Usage is read from the response.completed event chunk when streaming, or from response.usage for non-streaming calls.

​
Version warning

If you have an older OpenAI SDK installed, actguard emits a warning when patch_openai() runs (i.e., the first time you enter a BudgetGuard): 
UserWarning: actguard requires openai>=1.76.0; detected 1.50.0.
Budget tracking may fail with this SDK version.
Upgrade to silence the warning and ensure correct behaviour: 
pip install "openai>=1.76.0"