Spaces:
Sleeping
Sleeping
feat: smart_read_file, extract clean text, extra tools
Browse files
tools.py
CHANGED
@@ -3,12 +3,14 @@ import base64
|
|
3 |
import json
|
4 |
import inspect
|
5 |
import time
|
6 |
-
|
|
|
|
|
7 |
|
8 |
from datetime import datetime, timezone
|
|
|
9 |
|
10 |
from langchain.tools import tool
|
11 |
-
|
12 |
from langchain_community.tools.tavily_search import TavilySearchResults
|
13 |
from langchain_core.messages import HumanMessage
|
14 |
from langchain_google_genai.chat_models import ChatGoogleGenerativeAIError
|
@@ -21,6 +23,10 @@ from langchain_google_community import SpeechToTextLoader
|
|
21 |
from langchain_community.tools import YouTubeSearchTool
|
22 |
from youtube_transcript_api import YouTubeTranscriptApi
|
23 |
from langchain_community.tools.file_management.read import ReadFileTool
|
|
|
|
|
|
|
|
|
24 |
|
25 |
from basic_agent import print_conversation
|
26 |
|
@@ -115,28 +121,94 @@ def search_and_extract(query: str) -> list[dict]:
|
|
115 |
return structured_results
|
116 |
|
117 |
|
118 |
-
youtube_search_api = YouTubeSearchTool()
|
119 |
-
|
120 |
@tool
|
121 |
-
def
|
122 |
-
"""
|
|
|
|
|
|
|
|
|
|
|
123 |
if CUSTOM_DEBUG:
|
124 |
print_tool_call(
|
125 |
-
|
126 |
-
tool_name='
|
127 |
-
args={'
|
128 |
)
|
129 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
130 |
if CUSTOM_DEBUG:
|
131 |
-
print_tool_response(
|
132 |
-
return response
|
133 |
|
|
|
|
|
134 |
|
135 |
def extract_video_id(url: str) -> str:
|
136 |
parsed = urlparse(url)
|
137 |
return parse_qs(parsed.query).get("v", [""])[0]
|
138 |
|
139 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
140 |
@tool
|
141 |
def load_youtube_transcript(url: str) -> str:
|
142 |
"""Load a YouTube transcript using youtube_transcript_api."""
|
@@ -165,43 +237,21 @@ def load_youtube_transcript(url: str) -> str:
|
|
165 |
return error_str
|
166 |
|
167 |
|
168 |
-
|
169 |
-
gemini = ChatGoogleGenerativeAI(model="gemini-1.5-flash")
|
170 |
|
171 |
@tool
|
172 |
-
def
|
173 |
-
"""
|
174 |
-
Uses Gemini Vision to answer a question about an image.
|
175 |
-
- image_path: file path to the image to analyze (.png)
|
176 |
-
- question: the query to ask about the image
|
177 |
-
"""
|
178 |
-
try:
|
179 |
-
base64_img = encode_image_to_base64(image_path)
|
180 |
-
except OSError:
|
181 |
-
response = f"OSError: Invalid argument (invalid image path or file format): {image_path}. Please provide a valid PNG image."
|
182 |
-
print_tool_response(response)
|
183 |
-
return response
|
184 |
-
|
185 |
-
base64_img_str = f"data:image/png;base64,{base64_img}"
|
186 |
if CUSTOM_DEBUG:
|
187 |
print_tool_call(
|
188 |
-
|
189 |
-
tool_name='
|
190 |
-
args={'
|
191 |
)
|
192 |
-
|
193 |
-
{"type": "text", "text": question},
|
194 |
-
{"type": "image_url", "image_url": base64_img_str},
|
195 |
-
])
|
196 |
-
try:
|
197 |
-
response = gemini.invoke([msg])
|
198 |
-
except ChatGoogleGenerativeAIError:
|
199 |
-
response = "ChatGoogleGenerativeAIError: Invalid argument provided to Gemini: 400 Provided image is not valid"
|
200 |
-
print_tool_response(response)
|
201 |
-
return response
|
202 |
if CUSTOM_DEBUG:
|
203 |
-
print_tool_response(response
|
204 |
-
return response
|
205 |
|
206 |
|
207 |
@tool
|
@@ -223,43 +273,109 @@ def search_and_extract_from_wikipedia(query: str) -> list:
|
|
223 |
|
224 |
@tool
|
225 |
def transcribe_audio(file_path: str) -> list:
|
226 |
-
"""Transcribe audio from
|
|
|
227 |
if CUSTOM_DEBUG:
|
228 |
print_tool_call(
|
229 |
transcribe_audio,
|
230 |
tool_name='transcribe_audio',
|
231 |
args={'file_path': file_path},
|
232 |
)
|
233 |
-
|
234 |
-
|
235 |
-
|
236 |
-
|
237 |
-
|
238 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
239 |
|
240 |
-
|
241 |
-
|
|
|
|
|
|
|
242 |
|
243 |
if CUSTOM_DEBUG:
|
244 |
print_tool_response(docs_content)
|
245 |
return docs_content
|
246 |
|
247 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
248 |
read_tool = ReadFileTool()
|
249 |
|
250 |
|
251 |
@tool
|
252 |
-
def
|
253 |
-
"""
|
|
|
|
|
|
|
|
|
|
|
254 |
if CUSTOM_DEBUG:
|
255 |
print_tool_call(
|
256 |
-
|
257 |
-
tool_name='
|
258 |
args={'file_path': file_path},
|
259 |
)
|
260 |
-
|
261 |
-
|
262 |
-
|
263 |
-
|
264 |
-
|
265 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
import json
|
4 |
import inspect
|
5 |
import time
|
6 |
+
import trafilatura
|
7 |
+
|
8 |
+
from typing import Callable, Union
|
9 |
|
10 |
from datetime import datetime, timezone
|
11 |
+
from markitdown import MarkItDown
|
12 |
|
13 |
from langchain.tools import tool
|
|
|
14 |
from langchain_community.tools.tavily_search import TavilySearchResults
|
15 |
from langchain_core.messages import HumanMessage
|
16 |
from langchain_google_genai.chat_models import ChatGoogleGenerativeAIError
|
|
|
23 |
from langchain_community.tools import YouTubeSearchTool
|
24 |
from youtube_transcript_api import YouTubeTranscriptApi
|
25 |
from langchain_community.tools.file_management.read import ReadFileTool
|
26 |
+
from langchain.chains.summarize import load_summarize_chain
|
27 |
+
from langchain.prompts import PromptTemplate
|
28 |
+
from langchain_core.documents import Document
|
29 |
+
from langchain_openai import ChatOpenAI
|
30 |
|
31 |
from basic_agent import print_conversation
|
32 |
|
|
|
121 |
return structured_results
|
122 |
|
123 |
|
|
|
|
|
124 |
@tool
|
125 |
+
def aggregate_information(results: list[str], query: str) -> str:
|
126 |
+
"""
|
127 |
+
Processes a list of unstructured text chunks (e.g., search results) and produces a concise, query-specific summary.
|
128 |
+
|
129 |
+
Each input text is filtered and summarized individually in the context of the provided query. Irrelevant results are discarded.
|
130 |
+
Relevant content is aggregated and synthesized into a final, coherent answer that directly addresses the query.
|
131 |
+
"""
|
132 |
if CUSTOM_DEBUG:
|
133 |
print_tool_call(
|
134 |
+
aggregate_information,
|
135 |
+
tool_name='aggregate_information',
|
136 |
+
args={'results': results, 'query': query},
|
137 |
)
|
138 |
+
if not results:
|
139 |
+
response = "No search results provided."
|
140 |
+
if CUSTOM_DEBUG:
|
141 |
+
print_tool_response(response)
|
142 |
+
return response
|
143 |
+
|
144 |
+
# Convert to LangChain Document objects
|
145 |
+
docs = [Document(page_content=chunk) for chunk in results]
|
146 |
+
|
147 |
+
llm = ChatOpenAI(model="gpt-4o-mini", temperature=0.2)
|
148 |
+
|
149 |
+
# Map Prompt — Summarize each document in light of the query
|
150 |
+
map_prompt = PromptTemplate.from_template(
|
151 |
+
"You are analyzing a search result in the context of the question: '{query}'.\n\n"
|
152 |
+
"Search result:\n{text}\n\n"
|
153 |
+
"Instructions:\n"
|
154 |
+
"- If the result contains information relevant to answering the query, summarize the relevant parts clearly.\n"
|
155 |
+
"- If the result is not helpful or unrelated, return 'IGNORE'.\n"
|
156 |
+
"- Do not include generic information or filler.\n"
|
157 |
+
"- Focus on extracting facts, key statements, or numbers that directly support the query.\n\n"
|
158 |
+
"Relevant Summary:"
|
159 |
+
)
|
160 |
+
|
161 |
+
# Combine Prompt — Aggregate the summaries to one final answer
|
162 |
+
combine_prompt = PromptTemplate.from_template(
|
163 |
+
"You are aggregating information to answer the following question: '{query}'.\n\n"
|
164 |
+
"Here are the summaries from filtered search results:\n{text}\n\n"
|
165 |
+
"Using the most relevant points, write a clear, concise, and complete answer to the original query.\n"
|
166 |
+
"If there's conflicting information, mention it briefly. Otherwise, focus on consensus.\n\n"
|
167 |
+
"Final Answer:"
|
168 |
+
)
|
169 |
+
|
170 |
+
chain = load_summarize_chain(
|
171 |
+
llm,
|
172 |
+
chain_type="map_reduce",
|
173 |
+
map_prompt=map_prompt.partial(query=query),
|
174 |
+
combine_prompt=combine_prompt.partial(query=query),
|
175 |
+
)
|
176 |
+
|
177 |
+
summary = chain.invoke({'input_documents': docs})
|
178 |
+
output_text = summary.get('output_text', str(summary))
|
179 |
+
output_text = json.dumps({'summary': output_text})
|
180 |
+
|
181 |
if CUSTOM_DEBUG:
|
182 |
+
print_tool_response(output_text)
|
|
|
183 |
|
184 |
+
return output_text
|
185 |
+
|
186 |
|
187 |
def extract_video_id(url: str) -> str:
|
188 |
parsed = urlparse(url)
|
189 |
return parse_qs(parsed.query).get("v", [""])[0]
|
190 |
|
191 |
|
192 |
+
@tool
|
193 |
+
def get_audio_from_youtube(urls: list[str], save_dir:str="./tmp/") -> list[str | PurePath | None]:
|
194 |
+
"""Extracts audio from a YouTube video URL."""
|
195 |
+
|
196 |
+
if CUSTOM_DEBUG:
|
197 |
+
print_tool_call(
|
198 |
+
get_audio_from_youtube,
|
199 |
+
tool_name='get_audio_from_youtube',
|
200 |
+
args={'urls': urls, 'save_dir': save_dir},
|
201 |
+
)
|
202 |
+
loader = YoutubeAudioLoader(urls, save_dir)
|
203 |
+
audio_blobs = list(loader.yield_blobs())
|
204 |
+
paths = [str(blob.path) for blob in audio_blobs]
|
205 |
+
|
206 |
+
if CUSTOM_DEBUG:
|
207 |
+
print_tool_response(json.dumps({'paths': paths}))
|
208 |
+
|
209 |
+
return paths
|
210 |
+
|
211 |
+
|
212 |
@tool
|
213 |
def load_youtube_transcript(url: str) -> str:
|
214 |
"""Load a YouTube transcript using youtube_transcript_api."""
|
|
|
237 |
return error_str
|
238 |
|
239 |
|
240 |
+
youtube_search_api = YouTubeSearchTool()
|
|
|
241 |
|
242 |
@tool
|
243 |
+
def youtube_search_tool(query: str, number_of_results:int=3) -> list:
|
244 |
+
"""Search YouTube for a query and return the top number_of_results."""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
245 |
if CUSTOM_DEBUG:
|
246 |
print_tool_call(
|
247 |
+
youtube_search_tool,
|
248 |
+
tool_name='youtube_search_tool',
|
249 |
+
args={'query': query, number_of_results: number_of_results},
|
250 |
)
|
251 |
+
response = youtube_search_api.run(f"{query},{number_of_results}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
252 |
if CUSTOM_DEBUG:
|
253 |
+
print_tool_response(response)
|
254 |
+
return response
|
255 |
|
256 |
|
257 |
@tool
|
|
|
273 |
|
274 |
@tool
|
275 |
def transcribe_audio(file_path: str) -> list:
|
276 |
+
"""Transcribe audio from an audio file in file_path using Google Speech-to-Text."""
|
277 |
+
docs, docs_content = [], []
|
278 |
if CUSTOM_DEBUG:
|
279 |
print_tool_call(
|
280 |
transcribe_audio,
|
281 |
tool_name='transcribe_audio',
|
282 |
args={'file_path': file_path},
|
283 |
)
|
284 |
+
try:
|
285 |
+
loader = SpeechToTextLoader(
|
286 |
+
project_id=os.getenv("GOOGLE_CLOUD_PROJECT_ID"),
|
287 |
+
file_path=file_path,
|
288 |
+
is_long = False, # Set to True for long audio files
|
289 |
+
)
|
290 |
+
|
291 |
+
docs = loader.load()
|
292 |
+
except Exception as e:
|
293 |
+
print(f"Error loading audio file: {e}")
|
294 |
+
try:
|
295 |
+
loader = SpeechToTextLoader(
|
296 |
+
project_id=os.getenv("GOOGLE_CLOUD_PROJECT_ID"),
|
297 |
+
file_path=file_path,
|
298 |
+
is_long=True, # Set to True for long audio files
|
299 |
+
)
|
300 |
|
301 |
+
docs = loader.load()
|
302 |
+
except Exception as e:
|
303 |
+
docs_content = [f"Error loading audio file: {e}"]
|
304 |
+
|
305 |
+
docs_content = [doc.page_content for doc in docs] if docs else docs_content
|
306 |
|
307 |
if CUSTOM_DEBUG:
|
308 |
print_tool_response(docs_content)
|
309 |
return docs_content
|
310 |
|
311 |
|
312 |
+
@tool
|
313 |
+
def extract_clean_text_from_url(url: str) -> str:
|
314 |
+
"""Extract the main readable content from a webpage using trafilatura."""
|
315 |
+
if CUSTOM_DEBUG:
|
316 |
+
print_tool_call(
|
317 |
+
extract_clean_text_from_url,
|
318 |
+
tool_name='extract_clean_text_from_url',
|
319 |
+
args={'url': url},
|
320 |
+
)
|
321 |
+
downloaded = trafilatura.fetch_url(url)
|
322 |
+
response = ""
|
323 |
+
if not downloaded:
|
324 |
+
response = "Failed to download the page. Please check the URL."
|
325 |
+
|
326 |
+
if not "Failed" in response:
|
327 |
+
response = trafilatura.extract(downloaded)
|
328 |
+
|
329 |
+
response = response or "No meaningful content found."
|
330 |
+
if CUSTOM_DEBUG:
|
331 |
+
print_tool_response(response)
|
332 |
+
return response
|
333 |
+
|
334 |
+
|
335 |
read_tool = ReadFileTool()
|
336 |
|
337 |
|
338 |
@tool
|
339 |
+
def smart_read_file(file_path: str) -> str:
|
340 |
+
"""
|
341 |
+
Smart tool to read a file based on its type.
|
342 |
+
|
343 |
+
- Use `read_file_tool` for simple text, CSV, code files.
|
344 |
+
- Use MarkItDown for PDFs, Word, Excel, HTML, and other complex formats.
|
345 |
+
"""
|
346 |
if CUSTOM_DEBUG:
|
347 |
print_tool_call(
|
348 |
+
smart_read_file,
|
349 |
+
tool_name='smart_read_file',
|
350 |
args={'file_path': file_path},
|
351 |
)
|
352 |
+
_, ext = os.path.splitext(file_path.lower())
|
353 |
+
|
354 |
+
if ext in [".mp3", ".wav", ".m4a", ".flac"]:
|
355 |
+
# If the file is an audio file, transcribe it
|
356 |
+
return transcribe_audio.invoke({"file_path": file_path})
|
357 |
+
|
358 |
+
if ext in [".png", ".jpg", ".jpeg", ".gif", ".bmp"]:
|
359 |
+
# If the file is an image, use image_query_tool to analyze it
|
360 |
+
q = "What can you tell me about this image?"
|
361 |
+
return image_query_tool.invoke({"image_path": file_path, "question": q})
|
362 |
+
|
363 |
+
if any(ext in url_pattern for url_pattern in ["http://", "https://", "www."]):
|
364 |
+
if "youtube.com/watch?v=" in file_path:
|
365 |
+
transcript = load_youtube_transcript.invoke({"url": file_path})
|
366 |
+
if "Error loading" in transcript:
|
367 |
+
return get_audio_from_youtube.invoke({'urls': [file_path], 'save_dir': './tmp/'})
|
368 |
+
else:
|
369 |
+
return extract_clean_text_from_url.invoke(file_path)
|
370 |
+
|
371 |
+
md = MarkItDown()
|
372 |
+
try:
|
373 |
+
result = md.convert(file_path)
|
374 |
+
result = result.text_content
|
375 |
+
except Exception as e:
|
376 |
+
# print("Error reading file with MarkItDown:", e)
|
377 |
+
result = read_tool.invoke({"file_path": file_path})
|
378 |
+
|
379 |
+
if CUSTOM_DEBUG:
|
380 |
+
print_tool_response(result)
|
381 |
+
return result
|