agentmetrics-autogen 0.2.0

AgentMetrics observability integration for AutoGen agents

agentmetrics-autogen

AgentMetrics integration for AutoGen. Wrap team.run_stream() with one async context manager and every run reports back to your dashboard showing latency, tool calls, and errors. Token counts are not available from AutoGen's streaming events and are not tracked.

---

Install

pip install agentmetrics-autogen

---

Quickstart

from agentmetrics_autogen import AgentMetricsRunStream

tracker = AgentMetricsRunStream(
    agent_id="my-autogen-team",
    base_url="http://localhost:8099",
)

async with tracker.run(team, task="Analyze this dataset") as stream:
    async for event in stream:
        pass  # handle events as normal

tracker.flush()

---

API

AgentMetricsRunStream(agent_id, base_url)

Parameter	Default	Description
agent_id	"autogen-agent"	Label shown in the dashboard
base_url	"http://localhost:8099"	AgentMetrics server address

.run(team, **kwargs)

Returns an async context manager. Calls team.run_stream(**kwargs) internally. Yields a tracking iterator that intercepts ToolCallRequestEvent, ToolCallExecutionEvent, and TaskResult events to collect run metrics.

Emits a run summary to /v1/events on context exit (success or exception).

.flush(timeout=10.0)

Blocks until all in-flight HTTP requests complete.

---

What gets tracked

Each .run() call emits one event to /v1/events on completion or exception:

Field	Description
status	success or failed
duration_ms	Wall-clock run duration
tool_calls	Number of ToolCallRequestEvent events
tool_errors	Number of tool results with is_error=True
tool_names	Set of tool names from request events
error	Error message if TaskResult.stop_reason contains "error"

---

License

MIT