Everything you need to ship AI agents you can debug.
Five capabilities, built on OpenTelemetry, designed to be adopted to fit your framework, vendor and backend. Point your OpenTelemetry exporter at Trefur, set your key, ship the agent. Trefur shows you what it does.
Your stack stays as it is: the trace, the cost, the alert, and the fix.
What changes when you turn it on.
These are the outcomes engineering teams shipping production agents see in the first week. Each one is backed by a capability you can click into below.
Find the failing decision in seconds, not days
Search by trace_id, customer interaction, error class, or build. Open the trace, see the exact step that went wrong, see what the model saw before it answered.
See your model bill by step
Every token, every dollar, attributed to the step that emitted it. Roll up by agent, by team, by deployment. See drift the day it starts, not the day finance asks.
Hear about regressions before customers do
Alert envelopes on retry rate, error rate, cost-per-run, p95 latency. Routed to Slack, PagerDuty, Teams, or any webhook — with the offending trace already attached.
Swap model vendors without breaking your dashboards
One canonical trace shape across OpenAI, Anthropic, Bedrock, Azure OpenAI, Gemini. Move workloads to the cheapest or fastest provider without losing observability continuity.
Hand engineering a repro, not a screenshot
Reconstruct any failing run from production — exact prompt, exact tool response, exact retry. Engineering opens it, sees the problem, ships the fix.
The capabilities behind it.
Pick the one closest to your problem.
The agent trace, end to end
One root span per agent run. Every LLM call, tool call, retry, and outcome as a child span. Schema-validated, OpenTelemetry-canonical, replayable.
See exactly what every run costs
Tokens and dollars attributed to the step that emitted them. Roll up by agent, build, deployment, model vendor. Catch cost drift before the invoice.
Alerts that reach the right human
Retry-rate spikes, cost-per-run drift, error-rate envelopes, latency walks. Routed to Slack, PagerDuty, Teams, webhook — with the trace attached.
One trace across every model vendor
OpenAI, Anthropic, Bedrock, Azure OpenAI, Gemini, Groq, MiniMax. Move workloads between vendors without losing trace continuity or breaking dashboards.
Replay a failing run from production
Reconstruct the exact prompt the model saw, the exact tool response, the exact retry. Hand engineering a repro, not a screenshot.
Together, this is what changes.
Each feature is useful on its own. Run them together and the shape of the work changes. You get ahead of invoices, screenshots, and customer escalations. You ship faster because the agent is fully visible.
- Ship agents to production with confidence — every decision is captured the moment it happens
- Make model-vendor choices on price and quality, not on which one your tracing supports
- Hand finance, ops, and engineering the same source of truth — by run, by agent, by team
- Adopt on your current stack — OpenTelemetry-native, drops into existing pipelines
Start free. First trace in under five minutes.
Point your OpenTelemetry exporter at Trefur, set your key, run your agent. The trace appears.