Every handoff and tool call in your Agents SDK app.
The OpenAI Agents SDK runs agents that hand off to each other, call tools, and enforce guardrails; Obsivara shows you how each run actually behaved and what it cost. Every run traced through handoffs, tool calls, and model requests — each carrying tokens, latency, and dollars — so a run that loops or routes to the wrong agent points to the exact step. The SDK's tracing has pluggable processors; add Obsivara's or an OTel exporter, out of the request path.
OpenAI Agents SDK, fully observable.
Run and handoff tracing
Every agent run traced as a span tree — each handoff between agents, the tool calls made, and the model request behind each step, with inputs, outputs, and timing.
Guardrail and tool visibility
Input and output guardrail checks and tool invocations recorded with pass/fail and duration, so you see when a guardrail tripped or a tool failed and what happened next.
Per-agent cost attribution
Tokens and dollars rolled up per agent, run, and workflow, so multi-agent routing shows you which agent drives spend.
Failure attribution and health
Failed runs, tool errors, and guardrail trips classified and traced to the step that broke, with health scoring and predictive alerts.
Three steps. No code changes.
Add a trace processor or OTel exporter
The Agents SDK has built-in tracing with pluggable trace processors — register Obsivara's, or export the SDK's traces to Obsivara's OTLP endpoint. 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 agents
Traces flow as runs execute handoffs, tools, and guardrails — no changes to your agent or tool definitions.
See it in Obsivara
Agents appear in the inventory and knowledge map with per-run cost, latency, and health within minutes.
Signals tracked out of the box.
- →Agent run and handoff traces
- →Tool calls and guardrail outcomes
- →Tokens and cost per agent and run
- →Latency per step and model
- →Error rates and failing steps
- →Health score and predictive alerts
Common questions.
Does Obsivara use the Agents SDK's built-in tracing?
Yes. The SDK exposes pluggable trace processors; Obsivara registers as one, or you export its traces over OpenTelemetry — either way, no separate instrumentation of your agent logic is needed.
Are handoffs between agents traced?
Yes. Each handoff appears in the span tree, so you can follow a run as it routes from one agent to the next and see where it went wrong.
Does it capture guardrails?
Yes. Input and output guardrail checks are recorded with their results, so tripped guardrails are visible alongside the run they interrupted.