AutoGen¶

Test AutoGen multi-agent conversations with Tenro.

Experimental

This framework integration is experimental. Examples may break when the framework updates. See Compatibility for support definitions.

Tested with: autogen-agentchat==0.5.0, autogen-ext==0.5.0

What you'll use¶

Decorator	Purpose
`@link_tool`	Your custom tools (search, database, APIs)
`@link_agent`	Entry point for tracing

No @link_llm needed. Tenro intercepts LLM calls at the HTTP level automatically.

Customer support example¶

A customer support conversation with multiple agents.

"""Customer Support: Testing knowledge base retrieval with AutoGen."""

from __future__ import annotations

from examples.experimental.autogen.myapp.agents import CustomerSupportAgent, search_kb

from tenro import Provider, ToolCall
from tenro.simulate import agent, llm, tool
from tenro.testing import tenro


@tenro
def test_customer_support_answers_question() -> None:
    """Test customer support agent uses knowledge base and LLM.

    Flow: LLM decides to call search_kb → tool returns result → LLM answers.
    """
    # Simulate the tool that LLM will call
    tool.simulate(search_kb, result="Full refunds within 30 days.")

    # Simulate LLM: first requests tool, then gives final answer
    llm.simulate(
        Provider.OPENAI,
        responses=[
            [ToolCall("search_kb", query="refund")],
            "You can get a full refund within 30 days of purchase. TERMINATE",
        ],
    )

    result = CustomerSupportAgent().run("How do I get a refund?")

    assert result == "You can get a full refund within 30 days of purchase. TERMINATE"
    agent.verify(CustomerSupportAgent)
    llm.verify_many(Provider.OPENAI, count=2)
    tool.verify_many(search_kb, count=1)

RAG pipeline example¶

A retrieval-augmented generation conversation with document search.

"""RAG Pipeline: Testing document retrieval with AutoGen."""

from __future__ import annotations

from examples.experimental.autogen.myapp.agents import RAGPipeline, fetch_docs

from tenro import Provider, ToolCall
from tenro.simulate import agent, llm, tool
from tenro.testing import tenro


@tenro
def test_rag_pipeline_synthesizes_answer() -> None:
    """Test RAG pipeline fetches documents and generates answer.

    Flow: LLM decides to call fetch_docs → tool returns docs → LLM synthesizes.
    """
    # Simulate the tool that LLM will call
    tool.simulate(
        fetch_docs,
        result="Machine learning uses algorithms to learn.\nDeep learning is a subset of ML.",
    )

    # Simulate LLM: first requests tool, then synthesizes answer
    llm.simulate(
        Provider.OPENAI,
        responses=[
            [ToolCall("fetch_docs", topic="AI")],
            "ML uses algorithms to learn from data. TERMINATE",
        ],
    )

    result = RAGPipeline().run("What is machine learning?", "AI")

    assert result == "ML uses algorithms to learn from data. TERMINATE"
    agent.verify(RAGPipeline)
    llm.verify_many(Provider.OPENAI, count=2)
    tool.verify_many(fetch_docs, count=1)

Multi-turn conversation example¶

A multi-agent conversation handling complex queries.

"""Multi-Turn Conversation: Testing sequential LLM calls with AutoGen."""

from __future__ import annotations

from examples.experimental.autogen.myapp.agents import ConversationAgent

from tenro import Provider
from tenro.simulate import agent, llm
from tenro.testing import tenro


@tenro
def test_multi_turn_conversation() -> None:
    """Test agent handles multi-turn conversation with context."""
    llm.simulate(
        Provider.OPENAI,
        responses=[
            "A list in Python is created with square brackets: my_list = [1, 2, 3] TERMINATE",
            "To add items, use append(): my_list.append(4) TERMINATE",
        ],
    )

    responses = ConversationAgent().run(
        ["How do I create a list in Python?", "How do I add items to it?"]
    )

    assert responses == [
        "A list in Python is created with square brackets: my_list = [1, 2, 3] TERMINATE",
        "To add items, use append(): my_list.append(4) TERMINATE",
    ]
    agent.verify(ConversationAgent)
    llm.verify_many(Provider.OPENAI, count=2)

Key patterns¶

Agentic loop (LLM calls tool)¶

When the LLM decides to call a tool, then responds with the result:

from tenro import Provider, ToolCall
from tenro.simulate import llm, tool
# Assuming search_knowledge_base is defined with @link_tool("search_kb")

# 1. Set up tool result (use function reference)
tool.simulate(search_knowledge_base, result={"content": "Full refunds within 30 days."})

# 2. Set up LLM responses: first triggers tool, second is final response
llm.simulate(Provider.OPENAI, responses=[
    ToolCall(search_knowledge_base, query="refund"),
    "You can get a full refund within 30 days.",
])

Multi-agent conversations¶

from tenro import Provider, ToolCall
from tenro.simulate import llm
# Sequential responses for different agents with tool calls
llm.simulate(Provider.OPENAI, responses=[
    ToolCall(lookup, query="info"),
    "Agent 1: I found the information.",
    "Agent 2: Based on the findings..."
])

Verifying¶

from tenro import Provider
from tenro.simulate import llm, tool
tool.verify_many(search_knowledge_base, count=1)
llm.verify_many(Provider.OPENAI, count=2)  # Tool request + final answer