From a1ea01c6f17294ec07a6cedf238e651c2e09a709 Mon Sep 17 00:00:00 2001
From: Vrajesh Sulakhe <vbsulakhe21@gmail.com>
Date: Fri, 2 Oct 2026 20:52:35 +0530
Subject: [PATCH] fix(agentic): preserve provider prefixes for organization
 models

---
 .../chat_completion_agentic_loop.py           |  2 +-
 litellm/llms/custom_httpx/llm_http_handler.py |  2 +-
 .../test_agentic_loop_prefix_44069.py         | 82 +++++++++++++++++++
 3 files changed, 84 insertions(+), 2 deletions(-)
 create mode 100644 tests/test_litellm/test_agentic_loop_prefix_44069.py

diff --git a/litellm/litellm_core_utils/chat_completion_agentic_loop.py b/litellm/litellm_core_utils/chat_completion_agentic_loop.py
index 9d7c9864e62..2e842df4718 100644
--- a/litellm/litellm_core_utils/chat_completion_agentic_loop.py
+++ b/litellm/litellm_core_utils/chat_completion_agentic_loop.py
@@ -171,7 +171,7 @@ async def _execute_chat_completion_agentic_plan(
         raise ValueError("Agentic loop plan missing patched messages")
 
     full_model_name = patch.model or model
-    if "/" not in full_model_name:
+    if custom_llm_provider and not full_model_name.startswith(f"{custom_llm_provider}/"):
         full_model_name = f"{custom_llm_provider}/{full_model_name}"
 
     optional_params_for_followup: Final = {**optional_params, **patch.optional_params}
diff --git a/litellm/llms/custom_httpx/llm_http_handler.py b/litellm/llms/custom_httpx/llm_http_handler.py
index dd97db45a88..bfd60bb0f19 100644
--- a/litellm/llms/custom_httpx/llm_http_handler.py
+++ b/litellm/llms/custom_httpx/llm_http_handler.py
@@ -5494,7 +5494,7 @@ class BaseLLMHTTPHandler:
             raise ValueError("Agentic loop plan missing patched messages")
 
         full_model_name = patch.model or model
-        if "/" not in full_model_name:
+        if custom_llm_provider and not full_model_name.startswith(f"{custom_llm_provider}/"):
             full_model_name = f"{custom_llm_provider}/{full_model_name}"
 
         optional_params_for_followup: Final = dict(optional_params)
diff --git a/tests/test_litellm/test_agentic_loop_prefix_44069.py b/tests/test_litellm/test_agentic_loop_prefix_44069.py
new file mode 100644
index 00000000000..c3c5c7ddc7b
--- /dev/null
+++ b/tests/test_litellm/test_agentic_loop_prefix_44069.py
@@ -0,0 +1,82 @@
+from typing import Final, Literal
+from unittest.mock import AsyncMock, patch
+
+import pytest
+
+import litellm
+from litellm.integrations.custom_logger import CustomLogger
+from litellm.litellm_core_utils.chat_completion_agentic_loop import (
+    _execute_chat_completion_agentic_plan,
+)
+from litellm.llms.custom_httpx.llm_http_handler import BaseLLMHTTPHandler
+from litellm.types.integrations.custom_logger import (
+    AgenticLoopPlan,
+    AgenticLoopRequestPatch,
+)
+from litellm.types.utils import ModelResponse
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("execution_path", ("http", "sdk"))
+@pytest.mark.parametrize("use_patch_model", (False, True), ids=("original-model", "patch-model"))
+@pytest.mark.parametrize(
+    ("model_name", "custom_llm_provider", "expected_model"),
+    (
+        ("zai-org/GLM-5.3-Flash", "hosted_vllm", "hosted_vllm/zai-org/GLM-5.3-Flash"),
+        ("hosted_vllm/zai-org/GLM-5.3-Flash", "hosted_vllm", "hosted_vllm/zai-org/GLM-5.3-Flash"),
+        ("hosted_vllm/zai-org/GLM-5.3-Flash", "", "hosted_vllm/zai-org/GLM-5.3-Flash"),
+    ),
+    ids=("organization-model", "already-prefixed", "no-provider"),
+)
+async def test_agentic_followup_preserves_provider_prefix(
+    execution_path: Literal["http", "sdk"],
+    use_patch_model: bool,
+    model_name: str,
+    custom_llm_provider: str,
+    expected_model: str,
+) -> None:
+    model: Final = "original-model" if use_patch_model else model_name
+    plan: Final = AgenticLoopPlan(
+        run_agentic_loop=True,
+        request_patch=AgenticLoopRequestPatch(
+            model=model_name if use_patch_model else None,
+            messages=[{"role": "user", "content": "Continue"}],
+        ),
+    )
+    followup: Final = AsyncMock(wraps=litellm.acompletion)
+
+    with patch("litellm.acompletion", new=followup):
+        response: Final = (
+            await BaseLLMHTTPHandler()._execute_chat_completion_agentic_plan(
+                plan=plan,
+                model=model,
+                messages=[],
+                optional_params={"mock_response": "Follow-up complete"},
+                kwargs={"custom_llm_provider": custom_llm_provider},
+                custom_llm_provider=custom_llm_provider,
+                depth=0,
+                max_loops=3,
+                fingerprints=[],
+                fingerprint="followup",
+            )
+            if execution_path == "http"
+            else await _execute_chat_completion_agentic_plan(
+                plan=plan,
+                callback=CustomLogger(),
+                model=model,
+                optional_params={"mock_response": "Follow-up complete"},
+                kwargs={"custom_llm_provider": custom_llm_provider},
+                logging_obj=None,
+                custom_llm_provider=custom_llm_provider,
+                depth=0,
+                max_loops=3,
+                fingerprints=[],
+                fingerprint="followup",
+            )
+        )
+
+    assert isinstance(response, ModelResponse)
+    assert response.model == expected_model.removeprefix("hosted_vllm/")
+    followup.assert_awaited_once()
+    assert followup.await_args is not None
+    assert followup.await_args.kwargs["model"] == expected_model