Update app.py
Browse files
app.py
CHANGED
|
@@ -1,50 +1,79 @@
|
|
| 1 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
import os
|
|
|
|
| 3 |
import gradio as gr
|
| 4 |
from mcp import StdioServerParameters
|
| 5 |
from smolagents import MCPClient, CodeAgent, InferenceClientModel
|
| 6 |
|
| 7 |
-
#
|
| 8 |
SERVER_PATH = pathlib.Path(__file__).with_name("mcp_server.py")
|
| 9 |
|
| 10 |
-
#
|
| 11 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
|
| 13 |
|
|
|
|
| 14 |
def respond(message: str, history: list):
|
| 15 |
-
"""
|
| 16 |
params = StdioServerParameters(command="python", args=[str(SERVER_PATH)])
|
| 17 |
|
| 18 |
with MCPClient(params) as tools:
|
| 19 |
-
|
| 20 |
-
agent = CodeAgent(tools=tools, model=model)
|
| 21 |
answer = agent.run(message)
|
| 22 |
|
| 23 |
-
#
|
| 24 |
history.append({"role": "user", "content": message})
|
| 25 |
history.append({"role": "assistant", "content": answer})
|
| 26 |
return history, history
|
| 27 |
|
| 28 |
|
|
|
|
| 29 |
with gr.Blocks(title="Enterprise SQL Agent") as demo:
|
| 30 |
chat_state = gr.State([])
|
| 31 |
-
|
| 32 |
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
show_label=False,
|
| 36 |
-
)
|
| 37 |
textbox.submit(respond, [textbox, chat_state], [chatbot, chat_state])
|
| 38 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 39 |
gr.Markdown(
|
| 40 |
-
""
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
- List customers sorted by last order date.
|
| 44 |
-
- Find clients from the West with recent orders.
|
| 45 |
-
|
| 46 |
-
_Powered by smolagents + MCP + Hugging Face Inference API_
|
| 47 |
-
"""
|
| 48 |
)
|
| 49 |
|
| 50 |
if __name__ == "__main__":
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
app.py — Gradio front‑end + smolagents CodeAgent
|
| 3 |
+
================================================
|
| 4 |
+
|
| 5 |
+
This version avoids private/gated models and works on any free Hugging Face
|
| 6 |
+
Space **without extra secrets**. It relies on:
|
| 7 |
+
|
| 8 |
+
* `mcp_server.py` sitting next to this file
|
| 9 |
+
* A public chat‑completion capable model exposed via the HF Inference API
|
| 10 |
+
(defaults to **microsoft/Phi‑3‑mini‑4k‑instruct**, ~3 B params, free‑tier‑OK)
|
| 11 |
+
* `smolagents[mcp]` for the agent loop
|
| 12 |
+
* **Optional**: set `HF_MODEL_ID` or `HF_API_TOKEN` in **Settings → Secrets**
|
| 13 |
+
if you want a different (or gated) model.
|
| 14 |
+
|
| 15 |
+
If you hit the free‑tier rate‑limit you can still point to OpenAI by setting the
|
| 16 |
+
env var `OPENAI_API_KEY` — the code will auto‑switch to OpenAI chat.
|
| 17 |
+
"""
|
| 18 |
import os
|
| 19 |
+
import pathlib
|
| 20 |
import gradio as gr
|
| 21 |
from mcp import StdioServerParameters
|
| 22 |
from smolagents import MCPClient, CodeAgent, InferenceClientModel
|
| 23 |
|
| 24 |
+
# ---------- Tool server ------------------------------------------------------
|
| 25 |
SERVER_PATH = pathlib.Path(__file__).with_name("mcp_server.py")
|
| 26 |
|
| 27 |
+
# ---------- Model selection --------------------------------------------------
|
| 28 |
+
# 1) Use OpenAI automatically if OPENAI_API_KEY is set.
|
| 29 |
+
# 2) Otherwise fall back to a public HF Inference model that supports chat‑completion.
|
| 30 |
+
OPENAI_KEY = os.getenv("OPENAI_API_KEY")
|
| 31 |
+
HF_MODEL_ID = os.getenv("HF_MODEL_ID", "microsoft/Phi-3-mini-4k-instruct")
|
| 32 |
+
|
| 33 |
+
if OPENAI_KEY:
|
| 34 |
+
from smolagents.models import OpenAIChatModel # lazy import only if needed
|
| 35 |
+
BASE_MODEL = OpenAIChatModel() # defaults gpt‑4o‑preview
|
| 36 |
+
else:
|
| 37 |
+
BASE_MODEL = InferenceClientModel(model_id=HF_MODEL_ID)
|
| 38 |
|
| 39 |
|
| 40 |
+
# ---------- Gradio callback ---------------------------------------------------
|
| 41 |
def respond(message: str, history: list):
|
| 42 |
+
"""Run the user prompt through a CodeAgent that can call MCP SQL tools."""
|
| 43 |
params = StdioServerParameters(command="python", args=[str(SERVER_PATH)])
|
| 44 |
|
| 45 |
with MCPClient(params) as tools:
|
| 46 |
+
agent = CodeAgent(tools=tools, model=BASE_MODEL)
|
|
|
|
| 47 |
answer = agent.run(message)
|
| 48 |
|
| 49 |
+
# Append to chat history (OpenAI messages format)
|
| 50 |
history.append({"role": "user", "content": message})
|
| 51 |
history.append({"role": "assistant", "content": answer})
|
| 52 |
return history, history
|
| 53 |
|
| 54 |
|
| 55 |
+
# ---------- UI ---------------------------------------------------------------
|
| 56 |
with gr.Blocks(title="Enterprise SQL Agent") as demo:
|
| 57 |
chat_state = gr.State([])
|
| 58 |
+
gr.Markdown("## Enterprise SQL Agent — ask natural‑language questions about your data 🏢➡️📊")
|
| 59 |
|
| 60 |
+
chatbot = gr.Chatbot(type="messages", label="Chat")
|
| 61 |
+
textbox = gr.Textbox(placeholder="e.g. Who are my inactive Northeast customers?", show_label=False)
|
|
|
|
|
|
|
| 62 |
textbox.submit(respond, [textbox, chat_state], [chatbot, chat_state])
|
| 63 |
|
| 64 |
+
with gr.Accordion("Example prompts"):
|
| 65 |
+
gr.Markdown(
|
| 66 |
+
"""
|
| 67 |
+
* Who are my **Northeast** customers with no orders in 6 months?
|
| 68 |
+
* List customers sorted by **LastOrderDate**.
|
| 69 |
+
* Draft re‑engagement emails for inactive accounts.
|
| 70 |
+
"""
|
| 71 |
+
)
|
| 72 |
+
|
| 73 |
gr.Markdown(
|
| 74 |
+
"_Powered by MCP + smolagents + Gradio • Model: {}_".format(
|
| 75 |
+
"OpenAI (gpt‑4o)" if OPENAI_KEY else HF_MODEL_ID
|
| 76 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 77 |
)
|
| 78 |
|
| 79 |
if __name__ == "__main__":
|