zespan-autogen 1.2.0

Zespan observability middleware for Microsoft AutoGen

zespan-autogen

Observability middleware for Microsoft AutoGen via Zespan.

Install

# For AutoGen v0.4 (autogen-agentchat)
pip install 'zespan-autogen[agentchat]'

# For AutoGen v0.2 (pyautogen / ag2)
pip install 'zespan-autogen[legacy]'

AutoGen v0.4 — run_with_tracing()

Replace await team.run(task=...) with await run_with_tracing(team, task, ...):

from zespan_autogen import run_with_tracing
from autogen_agentchat.teams import RoundRobinGroupChat
from autogen_agentchat.conditions import MaxMessageTermination
from autogen_ext.models.openai import OpenAIChatCompletionClient
from autogen_agentchat.agents import AssistantAgent

model = OpenAIChatCompletionClient(model="gpt-4o")
agent = AssistantAgent("assistant", model_client=model)
team = RoundRobinGroupChat([agent], termination_condition=MaxMessageTermination(5))

trace_id = await run_with_tracing(
    team=team,
    task="Analyze the sales data",
    api_key="zsp_your_key",
    project_id="your_project_id",
)

AutoGen v0.4 — ZespanModelClient (per-LLM-call tracing)

from zespan_autogen import ZespanModelClient
from autogen_ext.models.openai import OpenAIChatCompletionClient

base = OpenAIChatCompletionClient(model="gpt-4o")
model = ZespanModelClient(base, api_key="zsp_your_key", project_id="proj_id")
agent = AssistantAgent("assistant", model_client=model)

AutoGen v0.2 — register_hooks()

from autogen import AssistantAgent, UserProxyAgent
from zespan_autogen import register_hooks

assistant = AssistantAgent("assistant", llm_config={"config_list": [...]})
user = UserProxyAgent("user", human_input_mode="NEVER")

register_hooks(assistant, user, api_key="zsp_your_key", project_id="proj_id")
user.initiate_chat(assistant, message="Research quantum computing trends")

What's captured

• All agent messages (source, content, token counts)
• Tool call requests and execution results
• Agent handoffs and delegation
• Internal reasoning/thought events (thinking models)
• Memory query events
• Raw LLM request/response (via ZespanModelClient)