Advanced Examples

Patterns that combine multiple agents, retrieval, planning, and reasoning. All snippets use verified APIs from the buddy package.

Multi-agent team

A coordinator delegates to specialist agents. Members are passed via members=.

from buddy import Agent, Team
from buddy.models.openai import OpenAIChat
from buddy.tools.tavily import TavilyTools

researcher = Agent(
    name="researcher",
    role="Find accurate, up-to-date information",
    model=OpenAIChat(id="gpt-4o-mini"),
    tools=[TavilyTools()],
    show_tool_calls=True,
)

writer = Agent(
    name="writer",
    role="Write clear, well-structured content",
    model=OpenAIChat(id="gpt-4o"),
    instructions=["Use markdown.", "Include headers and bullet points."],
)

team = Team(
    name="content_team",
    members=[researcher, writer],
    mode="coordinate",
    model=OpenAIChat(id="gpt-4o"),
    instructions=[
        "First, have the researcher gather facts.",
        "Then, have the writer compose the final piece.",
    ],
    markdown=True,
)

team.print_response("Write a 300-word article about recent advances in quantum computing.")

members=, not agents=

Team takes its participants through the required members parameter.

RAG with a vector database

Index sources into a vector store, then let the agent search them.

from buddy import Agent
from buddy.models.openai import OpenAIChat
from buddy.knowledge.url import UrlKnowledge
from buddy.vectordb.chroma import ChromaDb

knowledge = UrlKnowledge(
    urls=["https://example.com/handbook"],
    vector_db=ChromaDb(collection="handbook", persistent_client=True),
)
knowledge.load()  # read, chunk, embed, index

agent = Agent(
    model=OpenAIChat(id="gpt-4o-mini"),
    knowledge=knowledge,
    search_knowledge=True,
    instructions=["Search your knowledge base before answering.", "Cite sources."],
    markdown=True,
)

agent.print_response("What does the handbook say about onboarding?")

Embeddings

ChromaDb uses OpenAIEmbedder by default, so an OPENAI_API_KEY is needed for indexing. Pass a different embedder= to use another provider. Buddy also ships an irag knowledge base (buddy.knowledge.irag) that uses spaCy + TF-IDF and needs no external vector DB.

Planning agent

PlanningAgent decomposes a goal into an executable plan.

from buddy.planning.planner import PlanningAgent, PlanStrategy

agent = PlanningAgent(
    name="planner",
    planning_strategy=PlanStrategy.HIERARCHICAL,
    plan_validation=True,
    adaptive_replanning=True,
    max_planning_depth=4,
)

plan = agent.create_execution_plan(
    goal="Research and summarize recent AI safety developments",
    context={"depth": "comprehensive"},
    constraints=["Stay within 500 words"],
    resources=["web_search", "document_reader"],
)

print(f"Goal: {plan.goal}")
for step in plan.steps:
    print(f"  [{step.step_type.value}] {step.name}: {step.description}")

validation = agent.validate_plan(plan)
print("Plan valid:", validation["valid"])

Optional feature

Planning is an optional capability — check check_feature("planning") or get_available_features() at runtime.

Reasoning

Enable step-by-step reasoning with reasoning=True. You can bound the number of reasoning steps.

from buddy import Agent
from buddy.models.openai import OpenAIChat

agent = Agent(
    model=OpenAIChat(id="gpt-4o"),
    reasoning=True,
    reasoning_min_steps=1,
    reasoning_max_steps=10,
    markdown=True,
)

response = agent.run("If a train travels 60 km in 45 minutes, what is its speed in km/h?")
print(response.content)
print(response.reasoning_content)  # the reasoning trace, when available

Multi-model comparison

Run the same prompt across providers to compare latency and quality.

import time
from buddy import Agent
from buddy.models.openai import OpenAIChat
from buddy.models.anthropic import Claude
from buddy.models.google import Gemini

PROMPT = "Explain the concept of entropy in 2 sentences."

models = [
    ("OpenAI", OpenAIChat(id="gpt-4o-mini")),
    ("Anthropic", Claude(id="claude-3-5-sonnet-20241022")),
    ("Google", Gemini(id="gemini-1.5-flash")),
]

for name, model in models:
    agent = Agent(name=name, model=model)
    start = time.perf_counter()
    response = agent.run(PROMPT)
    print(f"{name} ({time.perf_counter() - start:.2f}s): {response.content}")

Provider extras and model ids

Each provider needs its extra (pip install buddy-ai[anthropic], [google], ...) and its own API key. The id strings above are illustrative — use whichever model ids your accounts have access to.