AG2

AG2 is an agent framework built on the autogen API. The AG2Instrumentor captures ConversableAgent chats, replies, and tool execution as OpenInference AGENT and TOOL spans in Arize AX. AG2 is a fork of the original 0.2-style AutoGen project. This integration is for AG2’s autogen module, not AG2 v1 or Microsoft’s separate autogen-agentchat package. See the Microsoft AutoGen AgentChat guide for the latter.

Prerequisites

Python 3.10+
An Arize AX account (sign up)

This guide uses AG2’s offline multi-agent and tool-call pattern, so it does not need an LLM provider key.

Launch Arize AX

Sign in to your Arize AX account.
From Space Settings, copy your Space ID and API Key. You will set them as ARIZE_SPACE_ID and ARIZE_API_KEY below.

Install

pip install arize-otel openinference-instrumentation-ag2 "ag2<1.0"

Configure credentials

export ARIZE_SPACE_ID="<your-space-id>"
export ARIZE_API_KEY="<your-api-key>"
export ARIZE_PROJECT_NAME="ag2-tracing-example"

Setup tracing

# instrumentation.py
import os

from arize.otel import register
from openinference.instrumentation.ag2 import AG2Instrumentor

tracer_provider = register(
    space_id=os.environ["ARIZE_SPACE_ID"],
    api_key=os.environ["ARIZE_API_KEY"],
    project_name=os.environ["ARIZE_PROJECT_NAME"],
)

AG2Instrumentor().instrument(tracer_provider=tracer_provider)
print("Arize AX tracing initialized for AG2.")

AG2 instrumentation supports the AG2 0.14.x autogen API. AG2 v1 uses a different middleware API and is not yet supported.

Spans and context captured

AG2 creates an AGENT span for each chat and agent reply. Reply spans nest under their chat span, and a function call made while producing a reply becomes a child TOOL span:

AG2 operation	Span name	Span kind
`initiate_chat` / `a_initiate_chat` (also used by `run` and `initiate_chats`)	`<agent>.initiate_chat`	`AGENT`
`generate_reply` / `a_generate_reply`	`<agent>.generate_reply`	`AGENT`
`execute_function` / `a_execute_function`	`<tool>`	`TOOL`

Tool spans use the registered function name (for example, get_weather) and record tool.name, tool_call.id, tool_call.function.arguments, resolved tool.parameters, and the result. Agent spans record the input and output messages and the agent name. The instrumentor preserves the active OpenTelemetry context, so an AG2 chat started inside one of your application’s spans remains in that trace. Context also flows through nested replies and tool execution. Use using_attributes to propagate a session ID, user ID, metadata, and tags to all AG2 spans within a context block. Pair it with the OpenAIInstrumentor when an AG2 agent uses OpenAI: this adds LLM child spans below the AGENT spans. The standard OpenInference privacy controls apply. Use suppress_tracing around calls you do not want to trace. Set OPENINFERENCE_HIDE_INPUTS=true, OPENINFERENCE_HIDE_OUTPUTS=true, OPENINFERENCE_HIDE_INPUT_MESSAGES=true, or OPENINFERENCE_HIDE_OUTPUT_MESSAGES=true before starting your process to mask the corresponding captured values. For programmatic masking or image truncation, configure TraceConfig, which takes precedence over environment variables.

Run AG2

# example.py
import json
from typing import Any

from instrumentation import tracer_provider
from autogen import ConversableAgent


def get_weather(city: str) -> str:
    return f"It is 72F and sunny in {city}."


def reply_with_weather(
    agent: ConversableAgent,
    messages: list[dict[str, Any]] | None = None,
    sender: Any = None,
    config: Any = None,
) -> tuple[bool, str]:
    _, result = agent.execute_function(
        {"name": "get_weather", "arguments": json.dumps({"city": "Portland"})},
        call_id="call-1",
    )
    return True, str(result["content"])


weather_agent = ConversableAgent(
    "weather_agent",
    llm_config=False,
    human_input_mode="NEVER",
)
weather_agent.register_function({"get_weather": get_weather})
weather_agent.register_reply(
    [ConversableAgent, None],
    reply_with_weather,
    position=0,
)

user_proxy = ConversableAgent(
    "user_proxy",
    llm_config=False,
    human_input_mode="NEVER",
)
chat = user_proxy.initiate_chat(
    weather_agent,
    message="What is the weather in Portland?",
    max_turns=1,
    silent=True,
)
print("weather_agent:", chat.chat_history[-1]["content"])

Expected output

Arize AX tracing initialized for AG2.
weather_agent: It is 72F and sunny in Portland.

More AG2 patterns

The quickstart above is deliberately offline. The following upstream-inspired patterns require OPENAI_API_KEY and the optional packages shown below. Add openai and openinference-instrumentation-openai to the install command, then instrument OpenAI before AG2:

from openinference.instrumentation.openai import OpenAIInstrumentor

OpenAIInstrumentor().instrument(tracer_provider=tracer_provider)

Use a real model available to your account in config_list.

Group chat

import os

from autogen import ConversableAgent, LLMConfig
from autogen.agentchat import initiate_group_chat
from autogen.agentchat.group.patterns import AutoPattern

llm_config = LLMConfig({"api_type": "openai", "model": "gpt-5.4-mini", "api_key": os.environ["OPENAI_API_KEY"]})
writer = ConversableAgent("writer", llm_config=llm_config)
reviewer = ConversableAgent("reviewer", llm_config=llm_config)
pattern = AutoPattern(initial_agent=writer, agents=[writer, reviewer], group_manager_args={"llm_config": llm_config})
result, _, _ = initiate_group_chat(pattern=pattern, messages="Write and review a one-sentence slogan.", max_rounds=3)
print(result.chat_history[-1]["content"])

Sequential chats

import os

from autogen import ConversableAgent, LLMConfig

llm_config = LLMConfig({"api_type": "openai", "model": "gpt-5.4-mini", "api_key": os.environ["OPENAI_API_KEY"]})
researcher = ConversableAgent("researcher", llm_config=llm_config)
editor = ConversableAgent("editor", llm_config=llm_config)
coordinator = ConversableAgent("coordinator", human_input_mode="NEVER")
coordinator.initiate_chats([
    {"recipient": researcher, "message": "List two benefits of tracing.", "max_turns": 1, "summary_method": "last_msg"},
    {"recipient": editor, "message": "Turn those benefits into one sentence.", "max_turns": 1, "summary_method": "last_msg"},
])

Structured outputs

from pydantic import BaseModel
import os

from autogen import ConversableAgent, LLMConfig

class Weather(BaseModel):
    city: str
    temperature_f: int

llm_config = LLMConfig(
    {"api_type": "openai", "model": "gpt-5.4-mini", "api_key": os.environ["OPENAI_API_KEY"]},
    response_format=Weather,
)
agent = ConversableAgent(
    "weather",
    llm_config=llm_config,
)
response = agent.run(message="Return Portland weather.", max_turns=1, user_input=False)
response.process()
print(Weather.model_validate_json(response.messages[-1]["content"]))

These patterns follow the AG2 instrumentor examples.

Verify in Arize AX

Open your Arize AX space and select project ag2-tracing-example.
You should see a new trace within ~30 seconds with AGENT spans for the chat and reply, plus a TOOL span named get_weather.
If no traces appear, see Troubleshooting.

Check from the skill, CLI, or SDK

Confirm spans are actually reaching your Arize AX project. Use whichever fits your workflow — the skill and CLI work for any framework; the SDK check is shown for each language.

Arize skill (agent)
AX CLI
SDK

Install the Arize Skills plugin and let your coding agent check for you:

npx skills add Arize-ai/arize-skills

Then prompt your agent:

Use the arize-trace skill to export and analyze recent traces from my project. Confirm spans are arriving, and summarize any errors or latency issues.

Export recent spans for your project — any rows mean traces are landing:

ax spans export "$ARIZE_PROJECT_NAME" --space "$ARIZE_SPACE_ID" \
  --limit 5 --stdout | jq 'length'

A non-zero count confirms spans reached Arize AX. Run ax auth login first if you have not authenticated. See the ax spans reference.

Query the project’s spans and check that at least one came back.

import os
from arize import ArizeClient

client = ArizeClient(api_key=os.environ["ARIZE_API_KEY"])
resp = client.spans.list(
    project=os.environ["ARIZE_PROJECT_NAME"],
    space=os.environ["ARIZE_SPACE_ID"],
    limit=5,
)
count = len(resp.spans)
print(
    f"{count} span(s) found" if count else "No spans yet — recheck setup"
)

// Reads ARIZE_API_KEY from the environment.
import { listSpans } from "@arizeai/ax-client";

const { data: spans } = await listSpans({
  project: process.env.ARIZE_PROJECT_NAME!,
  space: process.env.ARIZE_SPACE_ID!,
  limit: 5,
});
const count = spans.length;
console.log(
  count ? `${count} span(s) found` : "No spans yet — recheck setup",
);

client, err := arize.NewClient(
    arize.Config{APIKey: os.Getenv("ARIZE_API_KEY")},
)
if err != nil {
    log.Fatal(err)
}
resp, err := client.Spans.List(ctx, spans.ListRequest{
    Project: os.Getenv("ARIZE_PROJECT_NAME"),
    Space:   os.Getenv("ARIZE_SPACE_ID"),
    Limit:   5,
})
if err != nil {
    log.Fatal(err)
}
fmt.Printf("%d span(s) found\n", len(resp.Spans))

SDK span references: Python · TypeScript · Go.

Troubleshooting

No traces in Arize AX. Call AG2Instrumentor().instrument(...) before starting a chat.
Import or instrumentation error. Install an AG2 0.14.x release; AG2 v1 is not supported by this instrumentor.
The example asks for an LLM key. Keep both agents’ llm_config=False; this offline example runs a local tool instead of calling a model.
LLM spans are missing. Install and initialize OpenAIInstrumentor before AG2 when your agents use OpenAI.

OpenTelemetry

LLM Providers

Agent Frameworks

Coding Agents

Platforms

Orchestration

Evaluation

Prerequisites

Launch Arize AX

Install

Configure credentials

Setup tracing

Spans and context captured

Run AG2

Expected output

More AG2 patterns

Group chat

Sequential chats

Structured outputs

Verify in Arize AX

Check from the skill, CLI, or SDK

Troubleshooting

Resources

OpenInference AG2 Instrumentor

AG2 repository

​Prerequisites

​Launch Arize AX

​Install

​Configure credentials

​Setup tracing

​Spans and context captured

​Run AG2

​Expected output

​More AG2 patterns

​Group chat

​Sequential chats

​Structured outputs

​Verify in Arize AX

​Check from the skill, CLI, or SDK

​Troubleshooting

​Resources

OpenInference AG2 Instrumentor

AG2 repository

Prerequisites

Launch Arize AX

Install

Configure credentials

Setup tracing

Spans and context captured

Run AG2

Expected output

More AG2 patterns

Group chat

Sequential chats

Structured outputs

Verify in Arize AX

Check from the skill, CLI, or SDK

Troubleshooting

Resources