Les agents LangGraph peuvent appeler des outils en boucle, ce qui signifie qu'un outil de recherche non contraint peut épuiser les crédits sur une seule requête. Ce tutoriel construit un agent LangGraph avec la recherche Scavio comme nœud d'outil qui suit l'utilisation des crédits par conversation et cesse de chercher lorsqu'il atteint une limite budgétaire configurable. Chaque recherche coûte 0,005 $.
Prérequis
- Python 3.8+
- langgraph et langchain installés
- Une clé API Scavio depuis scavio.dev
- Une clé API LLM (OpenAI, Anthropic, ou locale)
Parcours
Étape 1: Définir l'outil de recherche avec suivi budgétaire
Créer un outil compatible LangGraph qui suit et limite les dépenses de recherche.
Python
import os, requests, json
from langchain_core.tools import tool
from typing import Optional
API_KEY = os.environ['SCAVIO_API_KEY']
SH = {'x-api-key': API_KEY, 'Content-Type': 'application/json'}
search_state = {'calls': 0, 'cost': 0.0, 'max_calls': 10}
@tool
def web_search(query: str, platform: Optional[str] = None) -> str:
"""Search the web for current information. Supports google, reddit, youtube, amazon, walmart platforms."""
if search_state['calls'] >= search_state['max_calls']:
return f"Budget limit reached ({search_state['max_calls']} searches, ${search_state['cost']:.3f}). Synthesize from existing results."
body = {'query': query, 'country_code': 'us'}
if platform: body['platform'] = platform
data = requests.post('https://api.scavio.dev/api/v1/search',
headers=SH, json=body).json()
search_state['calls'] += 1
search_state['cost'] += 0.005
results = data.get('organic_results', [])[:5]
formatted = [f"{r['position']}. {r['title'][:60]} - {r.get('snippet', '')[:100]}" for r in results]
return f"Search results for '{query}' ({search_state['calls']}/{search_state['max_calls']} budget):\n" + '\n'.join(formatted)
print(web_search.invoke({'query': 'best serp api 2026'}))Étape 2: Construire le graphe de l'agent LangGraph
Créer le graphe de l'agent avec un nœud d'outil de recherche et un routage conditionnel.
Python
from langgraph.graph import StateGraph, END
from langgraph.prebuilt import ToolNode
from langchain_core.messages import HumanMessage, AIMessage
from typing import TypedDict, Annotated, Sequence
import operator
class AgentState(TypedDict):
messages: Annotated[Sequence, operator.add]
def should_continue(state):
last = state['messages'][-1]
if hasattr(last, 'tool_calls') and last.tool_calls:
return 'tools'
return END
tool_node = ToolNode([web_search])
# Use your preferred LLM:
# from langchain_openai import ChatOpenAI
# llm = ChatOpenAI(model='gpt-4o').bind_tools([web_search])
# Or from langchain_anthropic import ChatAnthropic
# llm = ChatAnthropic(model='claude-sonnet-4-20250514').bind_tools([web_search])
def agent_node(state):
response = llm.invoke(state['messages'])
return {'messages': [response]}
graph = StateGraph(AgentState)
graph.add_node('agent', agent_node)
graph.add_node('tools', tool_node)
graph.set_entry_point('agent')
graph.add_conditional_edges('agent', should_continue, {'tools': 'tools', END: END})
graph.add_edge('tools', 'agent')
app = graph.compile()
print('LangGraph agent compiled with budget-aware search.')Étape 3: Exécuter l'agent avec suivi budgétaire
Exécuter l'agent sur une question de recherche et surveiller l'utilisation des crédits.
Python
def run_with_budget(question, max_searches=5):
search_state['calls'] = 0
search_state['cost'] = 0.0
search_state['max_calls'] = max_searches
result = app.invoke({'messages': [HumanMessage(content=question)]})
final = result['messages'][-1].content
print(f'\n--- Agent Response ---')
print(final[:500])
print(f'\n--- Budget ---')
print(f'Searches used: {search_state["calls"]}/{max_searches}')
print(f'Cost: ${search_state["cost"]:.3f}')
return final
# Research question with budget limit
run_with_budget('Compare the top 3 SERP APIs for Python developers in 2026', max_searches=5)Étape 4: Ajouter une stratégie de recherche multi-plateforme
Configurer l'agent pour rechercher sur plusieurs plateformes afin d'obtenir des réponses plus riches.
Python
def research_with_sources(topic, max_searches=8):
prompt = f"""Research this topic using multiple search platforms for a complete picture:
Topic: {topic}
Strategy:
1. Search Google for overview and top results
2. Search Reddit for real user opinions
3. Search YouTube for tutorial coverage
4. Synthesize findings with sources
Use the platform parameter: google (default), reddit, youtube, amazon
Budget: {max_searches} searches maximum."""
search_state['calls'] = 0
search_state['cost'] = 0.0
search_state['max_calls'] = max_searches
result = app.invoke({'messages': [HumanMessage(content=prompt)]})
print(f'\nResearch complete. {search_state["calls"]} searches, ${search_state["cost"]:.3f}')
return result['messages'][-1].content
research_with_sources('best search API for AI agents', max_searches=8)Exemple Python
Python
import os, requests
from langchain_core.tools import tool
SH = {'x-api-key': os.environ['SCAVIO_API_KEY'], 'Content-Type': 'application/json'}
budget = {'n': 0, 'max': 5}
@tool
def web_search(query: str) -> str:
"""Search the web for current information."""
if budget['n'] >= budget['max']:
return 'Budget reached. Use existing results.'
data = requests.post('https://api.scavio.dev/api/v1/search',
headers=SH, json={'query': query, 'country_code': 'us'}).json()
budget['n'] += 1
results = data.get('organic_results', [])[:3]
return '\n'.join(f"{r['title'][:50]}" for r in results)
print(web_search.invoke({'query': 'langgraph tutorial'}))
print(f'Budget: {budget["n"]}/{budget["max"]} (${budget["n"] * 0.005:.3f})')Exemple JavaScript
JavaScript
// LangGraph.js equivalent:
const SH = { 'x-api-key': process.env.SCAVIO_API_KEY, 'Content-Type': 'application/json' };
let budget = { n: 0, max: 5 };
async function webSearch(query) {
if (budget.n >= budget.max) return 'Budget reached.';
const data = await fetch('https://api.scavio.dev/api/v1/search', {
method: 'POST', headers: SH,
body: JSON.stringify({ query, country_code: 'us' })
}).then(r => r.json());
budget.n++;
return (data.organic_results || []).slice(0, 3)
.map(r => r.title.slice(0, 50)).join('\n');
}
console.log(await webSearch('langgraph tutorial'));
console.log(`Budget: ${budget.n}/${budget.max} ($${(budget.n * 0.005).toFixed(3)})`);Sortie attendue
JSON
Search results for 'best serp api 2026' (1/5 budget):
1. Scavio - Unified Search API for Developers - Best SERP API with multi-platform...
2. SerpAPI - Google Search API - Reliable Google search results...
3. DataForSEO - SEO Data Provider - Comprehensive SEO data...
LangGraph agent compiled with budget-aware search.
--- Agent Response ---
Based on my research across 3 searches, the top SERP APIs for Python developers...
--- Budget ---
Searches used: 3/5
Cost: $0.015