refactor(proxy_server.py): move admin ui to be locally hosted within proxy

litellm/proxy/admin_ui.py

@@ -0,0 +1,143 @@
+def add_new_model():
+    import streamlit as st
+    import json, requests, uuid
+
+    model_name = st.text_input(
+        "Model Name - user-facing model name", placeholder="gpt-3.5-turbo"
+    )
+    st.subheader("LiteLLM Params")
+    litellm_model_name = st.text_input(
+        "Model", placeholder="azure/gpt-35-turbo-us-east"
+    )
+    litellm_api_key = st.text_input("API Key")
+    litellm_api_base = st.text_input(
+        "API Base",
+        placeholder="https://my-endpoint.openai.azure.com",
+    )
+    litellm_api_version = st.text_input("API Version", placeholder="2023-07-01-preview")
+    litellm_params = json.loads(
+        st.text_area(
+            "Additional Litellm Params (JSON dictionary). [See all possible inputs](https://github.com/BerriAI/litellm/blob/3f15d7230fe8e7492c95a752963e7fbdcaf7bf98/litellm/main.py#L293)",
+            value={},
+        )
+    )
+    st.subheader("Model Info")
+    mode_options = ("completion", "embedding", "image generation")
+    mode_selected = st.selectbox("Mode", mode_options)
+    model_info = json.loads(
+        st.text_area(
+            "Additional Model Info (JSON dictionary)",
+            value={},
+        )
+    )
+
+    if st.button("Submit"):
+        try:
+            model_info = {
+                "model_name": model_name,
+                "litellm_params": {
+                    "model": litellm_model_name,
+                    "api_key": litellm_api_key,
+                    "api_base": litellm_api_base,
+                    "api_version": litellm_api_version,
+                },
+                "model_info": {
+                    "id": str(uuid.uuid4()),
+                    "mode": mode_selected,
+                },
+            }
+            # Make the POST request to the specified URL
+            complete_url = ""
+            if st.session_state["api_url"].endswith("/"):
+                complete_url = f"{st.session_state['api_url']}model/new"
+            else:
+                complete_url = f"{st.session_state['api_url']}/model/new"
+
+            headers = {"Authorization": f"Bearer {st.session_state['proxy_key']}"}
+            response = requests.post(complete_url, json=model_info, headers=headers)
+
+            if response.status_code == 200:
+                st.success("Model added successfully!")
+            else:
+                st.error(f"Failed to add model. Status code: {response.status_code}")
+
+            st.success("Form submitted successfully!")
+        except Exception as e:
+            raise e
+
+
+def streamlit_ui():
+    import streamlit as st
+    import requests
+
+    st.header("Admin Configuration")
+
+    # Add a navigation sidebar
+    st.sidebar.title("Navigation")
+    page = st.sidebar.radio("Go to", ("Proxy Setup", "Add Models", "List Models"))
+
+    # Initialize session state variables if not already present
+    if "api_url" not in st.session_state:
+        st.session_state["api_url"] = None
+    if "proxy_key" not in st.session_state:
+        st.session_state["proxy_key"] = None
+
+    # Display different pages based on navigation selection
+    if page == "Proxy Setup":
+        # Use text inputs with intermediary variables
+        input_api_url = st.text_input(
+            "Proxy Endpoint",
+            value=st.session_state.get("api_url", ""),
+            placeholder="http://0.0.0.0:8000",
+        )
+        input_proxy_key = st.text_input(
+            "Proxy Key",
+            value=st.session_state.get("proxy_key", ""),
+            placeholder="sk-...",
+        )
+        # When the "Save" button is clicked, update the session state
+        if st.button("Save"):
+            st.session_state["api_url"] = input_api_url
+            st.session_state["proxy_key"] = input_proxy_key
+            st.success("Configuration saved!")
+    elif page == "Add Models":
+        add_new_model()
+    elif page == "List Models":
+        # Check if the necessary configuration is available
+        if (
+            st.session_state.get("api_url", None) is not None
+            and st.session_state.get("proxy_key", None) is not None
+        ):
+            # Make the GET request
+            try:
+                complete_url = ""
+                if isinstance(st.session_state["api_url"], str) and st.session_state[
+                    "api_url"
+                ].endswith("/"):
+                    complete_url = f"{st.session_state['api_url']}models"
+                else:
+                    complete_url = f"{st.session_state['api_url']}/models"
+                response = requests.get(
+                    complete_url,
+                    headers={
+                        "Authorization": f"Bearer {st.session_state['proxy_key']}"
+                    },
+                )
+                # Check if the request was successful
+                if response.status_code == 200:
+                    models = response.json()
+                    st.write(models)  # or st.json(models) to pretty print the JSON
+                else:
+                    st.error(
+                        f"Failed to get models. Status code: {response.status_code}"
+                    )
+            except Exception as e:
+                st.error(f"An error occurred while requesting models: {e}")
+        else:
+            st.warning(
+                "Please configure the Proxy Endpoint and Proxy Key on the Proxy Setup page."
+            )
+
+
+if __name__ == "__main__":
+    streamlit_ui()

litellm/proxy/proxy_server.py

@@ -100,7 +100,7 @@ from typing import Union
 app = FastAPI(
     docs_url="/",
     title="LiteLLM API",
-    description="Proxy Server to call 100+ LLMs in the OpenAI format",
+    description="Proxy Server to call 100+ LLMs in the OpenAI format\n\nAdmin Panel on `/admin` endpoint",
 )
 router = APIRouter()
 origins = ["*"]
@@ -346,6 +346,29 @@ def load_from_azure_key_vault(use_azure_key_vault: bool = False):
         )
 
 
+async def run_streamlit_ui():
+    try:
+        # Start Streamlit without opening the browser automatically
+        process = subprocess.Popen(
+            [
+                "streamlit",
+                "run",
+                "admin_ui.py",
+                "--server.headless=true",
+                "--browser.serverAddress=0.0.0.0",
+                "--server.enableCORS=false",
+            ]
+        )
+        # Wait for the server to start before exiting the context manager
+        await asyncio.sleep(1)
+        print("Streamlit UI server has started successfully.")
+        # Keep the background task running
+        while True:
+            await asyncio.sleep(3600)
+    except Exception as e:
+        print(f"An error occurred: {e}")
+
+
 def cost_tracking():
     global prisma_client
     if prisma_client is not None:
@@ -982,6 +1005,9 @@ async def startup_event():
             duration=None, models=[], aliases={}, config={}, spend=0, token=master_key
         )
 
+    # Run streamlit_ui as a background task
+    asyncio.create_task(run_streamlit_ui())
+
 
 #### API ENDPOINTS ####
 @router.get(
@@ -1728,6 +1754,8 @@ async def add_new_model(model_params: ModelParams):
             }
         )
 
+        print(f"updated model list: {config['model_list']}")
+
         # Save the updated config
         with open(f"{user_config_file_path}", "w") as config_file:
             yaml.dump(config, config_file, default_flow_style=False)
@@ -1737,6 +1765,7 @@ async def add_new_model(model_params: ModelParams):
             router=llm_router, config_file_path=user_config_file_path
         )
 
+        print(f"llm_model_list: {llm_model_list}")
         return {"message": "Model added successfully"}
 
     except Exception as e:
@@ -2036,6 +2065,36 @@ async def retrieve_server_log(request: Request):
     return FileResponse(filepath)
 
 
+#### ADMIN UI ENDPOINTS ####
+
+
+@router.get("/admin")
+async def admin_page(request: Request):
+    from fastapi.responses import HTMLResponse
+
+    # Assuming your Streamlit app is running on localhost port 8501
+    html_content = """
+    <html>
+    <head>
+        <title>Admin Page</title>
+        <style>
+            html, body, iframe {
+                margin: 0;
+                padding: 0;
+                width: 100%;
+                height: 100%;
+                border: none;
+            }
+        </style>
+    </head>
+    <body>
+        <iframe src="http://localhost:8501"></iframe>
+    </body>
+</html>
+    """
+    return HTMLResponse(content=html_content)
+
+
 #### BASIC ENDPOINTS ####
 
 

pyproject.toml

@@ -33,6 +33,7 @@ proxy = [
     "pyyaml",
     "rq",
     "orjson",
+    "streamlit"
 ]
 
 extra_proxy = [

ui/app.py

@@ -5,7 +5,7 @@ from dotenv import load_dotenv
 
 load_dotenv()
 import streamlit as st
-import base64, binascii, os
+import base64, binascii, os, json
 from admin import admin_page
 from auth import auth_page, verify_with_otp
 import urllib.parse