Rylvo

Control Observability

Open any reply.
See exactly why.

Every reply your agent sends becomes a trace: each guardrail, search, tool call and model call behind it, with its time, tokens and cost.

Start freeTour the views

Six ways in

One reply, one conversation, one person, one cause.

Start wherever the question is. Every view reads the same store, so the number on a chart and the replies behind it always agree.

  1. 01Traces

    Every reply as a row, with its trace open beside it. Your filters and your place stay put while you read.

    list
  2. 02Sessions

    A whole conversation, turn by turn, with its total cost and worst outcome.

    sessions
  3. 03Users

    The people your bots talked to, with what each one cost and how it went.

    users
  4. 04Failure modes

    What went wrong, grouped by cause and tracked over time, with where to fix each one.

    failures
  5. 05Evaluators

    A sample of live replies scored against your criteria, each score with its reason.

    evaluators
  6. 06Prices

    The rate every model is charged at. Nothing unpriced is ever shown as $0.

    prices

Find the pattern

Not one bad reply. The reason behind twelve.

Failure modes groups what went wrong by cause and shows whether each one is rising or falling. Every count opens exactly the replies behind it, and Edge Cases picks the same failures up as cases to fix.

Analytics

Rooms for the numbers you check.

Build a room of charts for a bot and a time range, start from a template, or reuse widgets from your library. Share a read-only snapshot by link. Set an alert on error rate, latency or spend.

Your tools too

Already have a stack? Send it there.

Finished traces go out over OTLP/HTTP about every five minutes, from Test and Production alike. A new destination gets your history first, oldest first.

  • Langfuse
  • Honeycomb
  • Grafana Tempo
  • Datadog
  • Any OpenTelemetry collector

Retention

How long traces are kept.

Your plan sets the window. Shorten it for your workspace, or keep traces longer with a Retention Pack.

  • Free7 daysTrace list only — full replay needs Lite or above
  • Starter14 daysTrace list only — full replay needs Lite or above
  • Lite14 daysFull trace replay
  • Pro90 daysFull trace replay
  • EnterpriseUnlimitedFull trace replay

Questions

Asked often.

Is every reply traced, or only a sample?
Every reply on the engine opens a trace: from the dashboard chat, a channel, Beam or the public API. What is limited is how long traces are kept, which your plan sets.
Can I keep test traffic out of my numbers?
Yes. Every trace carries its environment, and Traces and Analytics each show Test or Production. Test and Production even have their own analytics rooms.
Where do costs come from?
From the model price list, for every model your bots actually called. A model with no price is shown as Unpriced, never $0. Set your own rate and new replies use it at once; recost past replies after a preview.
Can I send traces to my own tools?
Yes. Add an OTLP/HTTP destination such as Langfuse, Honeycomb, Grafana Tempo, Datadog or any OpenTelemetry collector. Finished replies are sent in batches about every five minutes, and a new destination first receives your history. Only admins and owners can change where traces go.
Can Rylvo tell me when something breaks?
Yes. Analytics alerts watch error rate, warning rate, p95 latency, average cost per reply or daily spend, and notify every member when a number crosses your line. They are thresholds on your own traffic, not anomaly detection.
How long are traces kept, and who can delete one?
Included trace retention is 7 days on Free, 14 on Starter and Lite, 90 on Pro, and unlimited on Enterprise; an organization can set a shorter window, or extend one with a Retention Pack. Deleting a trace is a separate permission from viewing one.

Stop guessing
what your agent did.

Start freeRead the docs