Trace every turn of your AutoGen conversation.
AutoGen builds systems where agents talk to each other to solve a task; Obsivara shows you what that conversation actually did and cost. Every turn traced across agents — each message, tool call, and model request carrying tokens, latency, and dollars — so a crew that loops or stalls points to the exact agent and step. AutoGen v0.4's runtime emits OpenTelemetry; point it at Obsivara, or instrument with the SDK, out of the request path.
AutoGen, fully observable.
Conversation-level tracing
Every multi-agent conversation traced turn by turn — which agent spoke, the tool calls it made, and the model request behind each reply, as one span tree.
Multi-agent cost attribution
Tokens and dollars rolled up per agent, conversation, and team, so you see which agent in the group chat drives spend and whether the back-and-forth earns it.
Loop and termination visibility
Conversations that fail to terminate or ping-pong between agents flagged, with the turn count and the step where the run stopped converging.
Failure attribution and health
Tool errors, model failures, and stalled turns classified and traced to the responsible agent, with health scoring and predictive alerts.
Three steps. No code changes.
Instrument with OpenTelemetry or the SDK
AutoGen v0.4's actor runtime emits OpenTelemetry — point its exporter at Obsivara's OTLP endpoint, or wrap your app with the Obsivara SDK. Your agents keep calling models directly; Obsivara records each run out-of-band, with no proxy in the path and no added latency.
Run your conversations
Traces flow as agents converse and call tools — no changes to your agent definitions or group-chat logic.
See it in Obsivara
Agent teams appear in the inventory and knowledge map with per-conversation cost, latency, and health within minutes.
Signals tracked out of the box.
- →Conversation and per-turn traces
- →Tokens and cost per agent and conversation
- →Tool calls and error rates
- →Turn counts and non-terminating runs
- →Latency per agent and model
- →Health score and predictive alerts
Common questions.
Does Obsivara support AutoGen v0.4 and the earlier AgentChat API?
Yes. v0.4's runtime emits OpenTelemetry that Obsivara ingests directly; earlier versions and AgentChat are covered via the Obsivara SDK wrapper.
Do I need to change my AutoGen agents?
No. You instrument via OpenTelemetry or the SDK; your agents, group chats, and termination logic stay exactly as they are.
Can Obsivara trace tool calls inside a conversation?
Yes. Each agent's tool invocations are traced with arguments, duration, and outcome, nested under the conversation turn that made them.