GAIA_agent_space / custom_tools.py
Jofthomas's picture
initial commit
06e83d9
import requests
from pydantic import BaseModel, Field
from huggingface_hub import InferenceClient
from openai import OpenAI
from bs4 import BeautifulSoup
from markdownify import markdownify as md
from langchain_core.tools import tool, Tool
from langchain_experimental.utilities import PythonREPL
from pypdf import PdfReader
from io import BytesIO
from youtube_transcript_api import YouTubeTranscriptApi
from pytube import extract
import pandas as pd, requests, json, tempfile, io, os, re
class SpreadsheetInput(BaseModel):
file_url: str = Field(..., description="URL of the .xlsx, .xls, or .csv file")
sheet: str | int | None = Field(
None,
description="Sheet name or 0-based index (only for Excel; ignored for CSV)",
)
# Quick one-shot aggregate (optional)
agg: str | None = Field(
None,
description="Optional quick aggregate of the form COL:FUNC (e.g. 'Sales:sum'). "
"Supported FUNC values: sum, mean, count. "
"If omitted, the whole sheet is returned as JSON.",
)
@tool(args_schema=SpreadsheetInput)
def spreadsheet_tool(file_url: str, sheet: str | int | None = None,
agg: str | None = None) -> str:
"""
Download a spreadsheet/CSV, load into pandas, optionally run a quick aggregate,
and return JSON or a scalar string.
"""
try:
# 1 ─ fetch
resp = requests.get(file_url, timeout=30)
resp.raise_for_status()
suffix = os.path.splitext(file_url.split("?")[0])[-1].lower()
# 2 ─ parse
if suffix in {".xlsx", ".xls"}:
df = pd.read_excel(io.BytesIO(resp.content), sheet_name=sheet or 0)
elif suffix == ".csv":
df = pd.read_csv(io.BytesIO(resp.content))
else:
return f"Unsupported file type: {suffix}"
# 3 ─ aggregate or full dump
if agg:
col, func = agg.split(":", 1)
col = col.strip()
func = func.strip().lower()
if func == "sum":
result = df[col].sum()
elif func == "mean":
result = df[col].mean()
elif func == "count":
result = df[col].count()
else:
return f"Unsupported aggregate function: {func}"
# Return a JSON scalar so the agent can parse easily
return json.dumps({"result": float(result)})
# Return full table (records-oriented JSON)
return df.to_json(orient="records")
except Exception as e:
return f"spreadsheet_tool failed: {e}"
class HtmlTableInput(BaseModel):
url: str = Field(..., description="Web page with an HTML table")
table_index: int = Field(0, description="Which table on the page (0-based)")
as_markdown: bool = Field(False, description="Return markdown instead of JSON")
@tool(args_schema=HtmlTableInput)
def html_table_query(url: str, table_index: int = 0, as_markdown: bool = False) -> str:
"""Fetch an HTML table and return it as JSON or markdown."""
try:
html = requests.get(url, timeout=20).text
df = pd.read_html(html)[table_index]
return df.to_markdown(index=False) if as_markdown else df.to_json(orient="records")
except Exception as e:
return f"html_table_query failed: {e}"
# --- Basic operations --- #
class WikipediaSearchInput(BaseModel):
"""Schema for WikipediaSearchTool"""
query: str = Field(..., description="The search phrase, e.g. 'Mercedes Sosa discography'")
lang: str = Field("en", description="Language code, default English Wikipedia")
top_k: int = Field(1, ge=1, le=10, description="How many top pages to fetch and return (1–10)")
@tool(args_schema=WikipediaSearchInput)
def wikipedia_search(query: str, lang: str = "en", top_k: int = 2) -> str:
"""
Search Wikipedia and return the top-k page extracts (intro only) as markdown.
Each hit contains:
β€’ Title (as H2)
β€’ Source URL
β€’ Intro/lead section text
"""
base = f"https://{lang}.wikipedia.org/w/api.php"
# STEP 1 – run search
try:
resp = requests.get(
base,
params={
"action": "query",
"list": "search",
"srsearch": query,
"srlimit": top_k,
"format": "json",
"utf8": 1,
},
timeout=15,
)
resp.raise_for_status()
hits = resp.json()["query"]["search"]
if not hits:
return f"No Wikipedia results for **{query}** in `{lang}`."
md_blocks = []
pageids = [str(h["pageid"]) for h in hits]
# STEP 2 – get plain-text extracts for those pageids
extracts = requests.get(
base,
params={
"action": "query",
"prop": "extracts",
"explaintext": True,
"exintro": True, # only the lead/intro paragraph(s)
"pageids": "|".join(pageids),
"format": "json",
"utf8": 1,
},
timeout=15,
)
extracts.raise_for_status()
pages = extracts.json()["query"]["pages"]
for pid in pageids:
page = pages[pid]
title = page["title"]
url = f"https://{lang}.wikipedia.org/?curid={pid}"
intro = page.get("extract", "").strip() or "_No extract available_"
md_blocks.append(f"## {title}\n[{url}]({url})\n\n{intro}")
return "\n\n---\n\n".join(md_blocks)
except Exception as e:
return f"wikipedia_search failed: {e}"
@tool
def multiply(a: float, b: float) -> float:
"""Multiplies two numbers.
Args:
a (float): the first number
b (float): the second number
"""
return a * b
@tool
def add(a: float, b: float) -> float:
"""Adds two numbers.
Args:
a (float): the first number
b (float): the second number
"""
return a + b
@tool
def subtract(a: float, b: float) -> int:
"""Subtracts two numbers.
Args:
a (float): the first number
b (float): the second number
"""
return a - b
@tool
def divide(a: float, b: float) -> float:
"""Divides two numbers.
Args:
a (float): the first float number
b (float): the second float number
"""
if b == 0:
raise ValueError("Cannot divided by zero.")
return a / b
@tool
def modulus(a: int, b: int) -> int:
"""Get the modulus of two numbers.
Args:
a (int): the first number
b (int): the second number
"""
return a % b
@tool
def power(a: float, b: float) -> float:
"""Get the power of two numbers.
Args:
a (float): the first number
b (float): the second number
"""
return a**b
# --- Functions --- #
@tool
def query_image(query: str, image_url: str, need_reasoning: bool = False) -> str:
"""Ask anything about an image using a Vision Language Model
Args:
query (str): The query about the image, e.g. how many persons are on the image?
image_url (str): The URL to the image
need_reasoning (bool): Set to True for complex query that require a reasoning model to answer properly. Set to False otherwise.
"""
try:
#use full model for image recognition
model_name = "gpt-4.1"
client = OpenAI()
response = client.responses.create(
model=model_name,
input=[{
"role": "user",
"content": [
{"type": "input_text", "text": query},
{
"type": "input_image",
"image_url": image_url,
},
],
}],
)
return response.output_text
except Exception as e:
return f"query_image failed: {e}"
@tool
def automatic_speech_recognition(file_url: str, file_extension: str) -> str:
"""Transcribe an audio file to text
Args:
file_url (str): the URL to the audio file
file_extension (str): the file extension, e.g. mp3
"""
# PROVIDER = 'huggingface'
try:
response = requests.get(file_url)
response.raise_for_status()
# write to disk
file_extension = file_extension.replace('.','')
with open(f'tmp.{file_extension}', 'wb') as file:
file.write(response.content)
audio_file = open(f'tmp.{file_extension}', "rb")
client = OpenAI()
transcription = client.audio.transcriptions.create(
model="whisper-1",
file=audio_file
)
return transcription.text
except Exception as e:
return f"automatic_speech_recognition failed: {e}"
@tool
def get_webpage_content(page_url: str) -> str:
"""Load a web page and return it to markdown if possible
Args:
page_url (str): the URL of web page to get
"""
try:
r = requests.get(page_url)
r.raise_for_status()
text = ""
# special case if page is a PDF file
if r.headers.get('Content-Type', '') == 'application/pdf':
pdf_file = BytesIO(r.content)
reader = PdfReader(pdf_file)
for page in reader.pages:
text += page.extract_text()
else:
soup = BeautifulSoup((r.text), 'html.parser')
if soup.body:
# convert to markdown
text = md(str(soup.body))
else:
# return the raw content
text = r.text
return text
except Exception as e:
return f"get_webpage_content failed: {e}"
# ======= Python code interpreter =======
# WARNING: Python REPL can execute arbitrary code on the host machine (e.g., delete files, make network requests). Use with caution.
class PythonREPLInput(BaseModel):
code: str = Field(description="The Python code string to execute.")
python_repl = PythonREPL()
python_repl_tool = Tool(
name="python_repl",
description="""A Python REPL shell (Read-Eval-Print Loop).
Use this to execute single or multi-line python commands.
Input should be syntactically valid Python code.
Always end your code with `print(...)` to see the output.
Do NOT execute code that could be harmful to the host system.
You are allowed to download files from URLs.
Do NOT send commands that block indefinitely (e.g., `input()`).""",
func=python_repl.run,
args_schema=PythonREPLInput
)
@tool
def get_youtube_transcript(page_url: str) -> str:
"""Get the transcript of a YouTube video
Args:
page_url (str): YouTube URL of the video
"""
try:
# get video ID from URL
video_id = extract.video_id(page_url)
# get transcript
ytt_api = YouTubeTranscriptApi()
transcript = ytt_api.fetch(video_id)
# keep only text
txt = '\n'.join([s.text for s in transcript.snippets])
return txt
except Exception as e:
return f"get_youtube_transcript failed: {e}"