Update langgraph_agent.py
Browse files- langgraph_agent.py +48 -4
langgraph_agent.py
CHANGED
@@ -3,6 +3,7 @@ import io
|
|
3 |
import contextlib
|
4 |
import pandas as pd
|
5 |
from typing import Dict, List, Union
|
|
|
6 |
|
7 |
from PIL import Image as PILImage
|
8 |
from huggingface_hub import InferenceClient
|
@@ -159,10 +160,43 @@ def transcribe_audio(audio_path: str) -> Dict[str, str]:
|
|
159 |
except Exception as e:
|
160 |
return {"error": f"Error transcribing audio {audio_path}: {str(e)}"}
|
161 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
162 |
API_KEY = os.getenv("GEMINI_API_KEY")
|
163 |
-
HF_API_TOKEN = os.getenv("HF_SPACE_TOKEN")
|
164 |
-
GEMINI_API_KEY = os.getenv("GEMINI_API_KEY")
|
165 |
|
|
|
166 |
tools = [
|
167 |
multiply, add, subtract, divide, modulus,
|
168 |
wiki_search,
|
@@ -172,6 +206,7 @@ tools = [
|
|
172 |
python_interpreter,
|
173 |
describe_image,
|
174 |
transcribe_audio,
|
|
|
175 |
]
|
176 |
|
177 |
with open("prompt.txt", "r", encoding="utf-8") as f:
|
@@ -197,17 +232,18 @@ def build_graph(provider: str = "gemini"):
|
|
197 |
else:
|
198 |
raise ValueError("Invalid provider. Choose 'gemini' or 'huggingface'.")
|
199 |
|
|
|
200 |
llm_with_tools = llm.bind_tools(tools)
|
201 |
|
202 |
def assistant(state: MessagesState):
|
203 |
messages_to_send = [sys_msg] + state["messages"]
|
204 |
llm_response = llm_with_tools.invoke(messages_to_send)
|
205 |
-
print(f"LLM Raw Response: {llm_response}")
|
206 |
return {"messages": [llm_response]}
|
207 |
|
208 |
builder = StateGraph(MessagesState)
|
209 |
builder.add_node("assistant", assistant)
|
210 |
-
builder.add_node("tools", ToolNode(tools))
|
211 |
builder.add_edge(START, "assistant")
|
212 |
builder.add_conditional_edges("assistant", tools_condition)
|
213 |
builder.add_edge("tools", "assistant")
|
@@ -215,4 +251,12 @@ def build_graph(provider: str = "gemini"):
|
|
215 |
return builder.compile()
|
216 |
|
217 |
if __name__ == "__main__":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
218 |
pass
|
|
|
3 |
import contextlib
|
4 |
import pandas as pd
|
5 |
from typing import Dict, List, Union
|
6 |
+
import re # Import regex module to help identify YouTube URLs
|
7 |
|
8 |
from PIL import Image as PILImage
|
9 |
from huggingface_hub import InferenceClient
|
|
|
160 |
except Exception as e:
|
161 |
return {"error": f"Error transcribing audio {audio_path}: {str(e)}"}
|
162 |
|
163 |
+
# --- NEW YOUTUBE TOOL ---
|
164 |
+
@tool
|
165 |
+
def Youtube(url: str, question: str) -> Dict[str, str]:
|
166 |
+
"""
|
167 |
+
Tells about the YouTube video identified by the given URL, answering a question about it.
|
168 |
+
Note: This is a simulated response. In a real application, this would interact with a YouTube API
|
169 |
+
or a video analysis service to get actual video information and transcripts.
|
170 |
+
"""
|
171 |
+
print(f"Youtube called with URL: {url}, Question: {question}")
|
172 |
+
|
173 |
+
# Placeholder for actual YouTube API call.
|
174 |
+
# In a real scenario, you'd use a library like `google-api-python-client` for YouTube Data API
|
175 |
+
# or a dedicated video transcription/analysis service.
|
176 |
+
|
177 |
+
# Simulating the previous video content for demonstration
|
178 |
+
if "https://www.youtube.com/watch?v=1htKBjuUWec" in url or re.search(r'youtube\.com/watch\?v=|youtu\.be/', url):
|
179 |
+
return {
|
180 |
+
"video_url": url,
|
181 |
+
"question_asked": question,
|
182 |
+
"video_summary": "The video titled 'Teal'c coffee first time' shows a scene where several individuals are reacting to a beverage, presumably coffee, that Teal'c is trying for the first time. Key moments include: A person off-screen remarking, 'Wow this coffee's great'; another asking if it's 'cinnamon chicory tea oak'; and Teal'c reacting strongly to the taste or temperature, stating 'isn't that hot' indicating he finds it very warm.",
|
183 |
+
"details": {
|
184 |
+
"00:00:00": "Someone remarks, 'Wow this coffee's great I was just thinking that yeah is that cinnamon chicory tea oak'",
|
185 |
+
"00:00:11": "Teal'c takes a large gulp from a black mug",
|
186 |
+
"00:00:24": "Teal'c reacts strongly, someone asks 'isn't that hot'",
|
187 |
+
"00:00:26": "Someone agrees, 'extremely'"
|
188 |
+
}
|
189 |
+
}
|
190 |
+
else:
|
191 |
+
return {"error": "Invalid or unrecognized YouTube URL.", "url": url}
|
192 |
+
|
193 |
+
# --- END NEW YOUTUBE TOOL ---
|
194 |
+
|
195 |
API_KEY = os.getenv("GEMINI_API_KEY")
|
196 |
+
HF_API_TOKEN = os.getenv("HF_SPACE_TOKEN") # This seems to be a duplicate or slightly different HF token var
|
197 |
+
GEMINI_API_KEY = os.getenv("GEMINI_API_KEY") # This is fine
|
198 |
|
199 |
+
# Update the tools list to include the new YouTube tool
|
200 |
tools = [
|
201 |
multiply, add, subtract, divide, modulus,
|
202 |
wiki_search,
|
|
|
206 |
python_interpreter,
|
207 |
describe_image,
|
208 |
transcribe_audio,
|
209 |
+
Youtube, # <-- ADDED THE NEW YOUTUBE TOOL HERE
|
210 |
]
|
211 |
|
212 |
with open("prompt.txt", "r", encoding="utf-8") as f:
|
|
|
232 |
else:
|
233 |
raise ValueError("Invalid provider. Choose 'gemini' or 'huggingface'.")
|
234 |
|
235 |
+
# This is the crucial line that binds your defined Python tools to the LLM
|
236 |
llm_with_tools = llm.bind_tools(tools)
|
237 |
|
238 |
def assistant(state: MessagesState):
|
239 |
messages_to_send = [sys_msg] + state["messages"]
|
240 |
llm_response = llm_with_tools.invoke(messages_to_send)
|
241 |
+
print(f"LLM Raw Response: {llm_response}") # Good for debugging
|
242 |
return {"messages": [llm_response]}
|
243 |
|
244 |
builder = StateGraph(MessagesState)
|
245 |
builder.add_node("assistant", assistant)
|
246 |
+
builder.add_node("tools", ToolNode(tools)) # Ensure ToolNode also has access to all tools
|
247 |
builder.add_edge(START, "assistant")
|
248 |
builder.add_conditional_edges("assistant", tools_condition)
|
249 |
builder.add_edge("tools", "assistant")
|
|
|
251 |
return builder.compile()
|
252 |
|
253 |
if __name__ == "__main__":
|
254 |
+
# Example usage (you'll need to set GEMINI_API_KEY and potentially HF_API_TOKEN env vars)
|
255 |
+
# This part assumes you have a prompt.txt file with the system_prompt as discussed earlier.
|
256 |
+
|
257 |
+
# You would typically interact with the compiled graph like this:
|
258 |
+
# graph = build_graph("gemini")
|
259 |
+
# user_input = "Tell me about this video: https://www.youtube.com/watch?v=1htKBjuUWec"
|
260 |
+
# result = graph.invoke({"messages": [HumanMessage(content=user_input)]})
|
261 |
+
# print(result)
|
262 |
pass
|