litellm - 💡(How to fix) Fix [Bug]: Proxy-level callbacks (`async_pre_call_hook`) are bypassed on Anthropic endpoint (`/v1/messages`) [1 pull requests]

Check for existing issues

• I have searched the existing issues and checked that my issue is not a duplicate.

What happened?

When making requests to the LiteLLM Proxy via the Anthropic-compatible endpoint (/v1/messages), custom proxy-level hooks defined in CustomLogger (such as async_pre_call_hook) are completely bypassed and not executed.

However, when hitting the standard OpenAI-compatible endpoints (e.g. /chat/completions or /v1/chat/completions), these exact same hooks are successfully triggered.

Any request routed through the proxy, regardless of whether it originates from the OpenAI /v1/chat/completions endpoint or the Anthropic /v1/messages endpoint, should reliably trigger the configured async_pre_call_hook callbacks.

Steps to Reproduce

1. Create a custom callback script custom_hook.py that inherits from CustomLogger and implements async_pre_call_hook:

import sys
from litellm.integrations.custom_logger import CustomLogger

class MyCustomHook(CustomLogger):
    async def async_pre_call_hook(self, user_api_key_dict, cache, data, call_type, **kwargs):
        print(">>> async_pre_call_hook EXECUTED!", file=sys.stderr)
        return data

proxy_handler_instance = MyCustomHook()

1. Configure config.yaml to load this callback:

litellm_settings:
  callbacks:
    - custom_hook.proxy_handler_instance

1. Start the proxy and send a request to the standard OpenAI endpoint:

curl -X POST "http://localhost:4000/v1/chat/completions" \
-H "Content-Type: application/json" \
-H "Authorization: Bearer sk-0000" \
-d '{"model": "gpt-3.5-turbo", "messages": [{"role": "user", "content": "hi"}]}'

Observe the proxy logs: >>> async_pre_call_hook EXECUTED! is printed.

1. Now send an equivalent request to the Anthropic endpoint:

curl -X POST "http://localhost:4000/v1/messages" \
-H "Content-Type: application/json" \
-H "Authorization: Bearer sk-0000" \
-d '{
  "model": "gpt-3.5-turbo", 
  "max_tokens": 100, 
  "messages": [{"role": "user", "content": "hi"}]
}'

Observe the proxy logs: The request goes through successfully, but the hook execution log is completely missing.

Relevant log output

# Log for /chat/completions (OpenAI format)
>>> async_pre_call_hook EXECUTED!
INFO:     127.0.0.1:45312 - "POST /v1/chat/completions HTTP/1.1" 200 OK

# Log for /v1/messages (Anthropic format)
INFO:     127.0.0.1:45324 - "POST /v1/messages HTTP/1.1" 200 OK
# (Missing the hook execution log entirely)

What part of LiteLLM is this about?

Proxy

What LiteLLM version are you on ?

v.1.83.10

Twitter / LinkedIn details

No response