Conjuntos de herramientas habituales (web, calculadora y RAG)

La caja de herramientas inicial

Herramienta 1: Búsqueda web

import requests

def web_search(query: str, k: int = 5):
    r = requests.post('https://api.tavily.com/search',
        json={'api_key': TAVILY_KEY, 'query': query, 'max_results': k})
    results = r.json()['results']
    return [
        {'title': r['title'], 'url': r['url'], 'snippet': r['content'][:300]}
        for r in results
    ]

Web Search Tool Schema

schema_search = {'type': 'function', 'function': {
    'name': 'web_search',
    'description': 'Search the public web. Use for current events or facts after the knowledge cutoff.',
    'parameters': {'type': 'object', 'properties': {
        'query': {'type': 'string', 'description': 'What to search for'},
        'k': {'type': 'integer', 'default': 5}
    }, 'required': ['query']}
}}
import json
print(json.dumps(schema_search, indent=2))

Herramienta 2: Calculadora

import math

ALLOWED = {k: getattr(math, k) for k in ['sqrt','log','exp','sin','cos','tan','pi','e']}

def calculator(expression: str) -> str:
    try:
        return str(eval(expression, {'__builtins__': {}}, ALLOWED))
    except Exception as e:
        return f'Error: {e}'

print(calculator('sqrt(16) + 2'))

Calculator Tool Schema

schema_calc = {'type': 'function', 'function': {
    'name': 'calculator',
    'description': 'Evaluate a math expression like 2 + 3 * sqrt(16). Use for any arithmetic.',
    'parameters': {'type': 'object', 'properties': {
        'expression': {'type': 'string'}
    }, 'required': ['expression']}
}}
import json
print(json.dumps(schema_calc, indent=2))

Tool 3: RAG / Knowledge Retrieval

from langchain_openai import OpenAIEmbeddings
from langchain_community.vectorstores import Chroma

store = Chroma(persist_directory='./db', embedding_function=OpenAIEmbeddings())

def retrieve(query: str, k: int = 4):
    docs = store.similarity_search(query, k=k)
    return [{'text': d.page_content, 'source': d.metadata.get('source')} for d in docs]

Retrieve Tool Schema

schema_retrieve = {'type': 'function', 'function': {
    'name': 'retrieve',
    'description': 'Search the internal knowledge base for company-specific information (policies, docs, history).',
    'parameters': {'type': 'object', 'properties': {
        'query': {'type': 'string'},
        'k': {'type': 'integer', 'default': 4}
    }, 'required': ['query']}
}}
import json
print(json.dumps(schema_retrieve, indent=2))

Herramienta 4: Intérprete de código (extra)

def run_python(code: str) -> str:
    # Use a sandboxed environment (E2B, Firecracker, gVisor)
    return sandbox.run(code).stdout

Herramienta 5: Obtener contenido HTTP (extra)

def fetch_url(url: str) -> str:
    r = requests.get(url, timeout=10)
    return BeautifulSoup(r.text, 'html.parser').get_text()[:5000]

Consejos para describir herramientas

Evitar la superposición de herramientas

Evaluación del uso de herramientas

Herramientas ocultas

La mejor forma de mejorar la selección

Repaso