Open WebUI / Lite LLM User Tracking

[NoMoreFood]

I use Lite LLM to attach Open WebUI for a small workgroup to connect to Azure AI. It works great, but was hoping to track which user made the request using the Lite LLM database. Is there any way to use something like the X-OpenWebUI-User-Email HTTP header such that Lite LLM uses that to track the user? Right now everything appears to register as default_user_id.

[ishaan-jaff]

yes, Does OpenWebUI send X-OpenWebUI-User-Email in the request header ?

[sickOscar]

That would be a really cool add that would enable many use cases for me

[krrishdholakia]

We just need to read the header if available, and set it as the user field - https://docs.litellm.ai/docs/proxy/customers#tracking-customer-spend

Place to do it -

litellm/litellm/proxy/litellm_pre_call_utils.py

Line 288 in 44b3429

def add_litellm_data_for_backend_llm_call(

A PR here is welcome!

[Ithanil]

I'd kindly suggest to use X-OpenWebUI-User-Id instead of the E-Mail here. I think there's no need to store the plaintext end-user email here when an administrator having access to both OpenWebUI and LiteLLM could easily infer it from the user id (but noone else).

Also, I have achieved the same effect by passing "user" with every chat request from OpenWebUI - it's basically one line of code. But of course the proposed solution here would be more accessible.
By the way, with a little hack you can also do per-end-user rate limiting with LiteLLM, without creating any budgets:

From 1b265121d77ada6196fdabd14d2a244d6d090b58 Mon Sep 17 00:00:00 2001
From: Jan Kessler <jakessle@uni-mainz.de>
Date: Tue, 18 Mar 2025 14:04:55 +0100
Subject: [PATCH] support globally limiting end user budget/tpm/rpm (instead of
 creating budgets for each end user)

---
 litellm/__init__.py                     | 2 ++
 litellm/proxy/auth/user_api_key_auth.py | 4 ++++
 2 files changed, 6 insertions(+)

diff --git a/litellm/__init__.py b/litellm/__init__.py
index 762a058c7..f6ac2c60a 100644
--- a/litellm/__init__.py
+++ b/litellm/__init__.py
@@ -277,6 +277,8 @@ max_ui_session_budget: Optional[float] = 10  # $10 USD budgets for UI Chat sessi
 internal_user_budget_duration: Optional[str] = None
 tag_budget_config: Optional[Dict[str, BudgetConfig]] = None
 max_end_user_budget: Optional[float] = None
+max_end_user_tpm: Optional[float] = None
+max_end_user_rpm: Optional[float] = None
 disable_end_user_cost_tracking: Optional[bool] = None
 disable_end_user_cost_tracking_prometheus_only: Optional[bool] = None
 custom_prometheus_metadata_labels: List[str] = []
diff --git a/litellm/proxy/auth/user_api_key_auth.py b/litellm/proxy/auth/user_api_key_auth.py
index 7e293b758..7cddb3ec4 100644
--- a/litellm/proxy/auth/user_api_key_auth.py
+++ b/litellm/proxy/auth/user_api_key_auth.py
@@ -609,6 +609,10 @@ async def _user_api_key_auth_builder(  # noqa: PLR0915
                             end_user_params["end_user_max_budget"] = (
                                 budget_info.max_budget
                             )
+                    else:
+                        end_user_params["end_user_tpm_limit"] = litellm.max_end_user_tpm
+                        end_user_params["end_user_rpm_limit"] = litellm.max_end_user_rpm
+                        end_user_params["end_user_max_budget"] = litellm.max_end_user_budget
             except Exception as e:
                 if isinstance(e, litellm.BudgetExceededError):
                     raise e
-- 
2.49.0

I have restricted these limits to certain requests types though (e.g. not embedding).

[damgem]

We just need to read the header if available, and set it as the user field - https://docs.litellm.ai/docs/proxy/customers#tracking-customer-spend

Place to do it -

litellm/litellm/proxy/litellm_pre_call_utils.py

Line 288 in 44b3429

def add_litellm_data_for_backend_llm_call(

A PR here is welcome!

Thanks for the helpful pointer! I'm happy to report I've given it a go here: add user_header_name by damgem · Pull Request #9802 · BerriAI/litellm :)
This would support both parsing X-OpenWebUI-User-Id or X-OpenWebUI-User-Email :)

[krrishdholakia]

Thanks @damgem responded on PR!