What can replay not tell me?

Replay can’t tell you why the model made the choice it made, only whether a given input and context reliably produce a given output. You can vary one condition and watch the output move, which narrows down what mattered, but the model’s actual reasoning, the internal weighing that got it there, stays outside what you observed. Replay also can’t tell you how common the failure is: one replayed case is one case, not a rate, so it doesn’t say whether this happens on one percent of sessions or one in a million. And replay is only as honest as the capture. If the original trace didn’t record a piece of context, a tool argument, a system prompt version, replay can’t recover it, and you’re debugging an incomplete picture without knowing it’s incomplete.

keep reading

More on this.

Two ways to run Tessary.

Tessary is an open-source agent reliability platform. Cloud and self-hosted run the same workflow on the OpenTelemetry traces your agent already emits.

Tessary Cloud

We host it for you. Send your first trace with nothing to deploy and no model key.

what's includedper organization
traces
10,000 per calendar month
stored trace data
1 GB
retention
30 days
model credit
$10, one-time, for triage and root-cause analysis
credit card
not required

Self-hosted Tessary

Run the open-source code on your own infrastructure with one command. Add your own model key for triage and root-cause analysis.

Self-host Tessary for me by following https://github.com/tessaryai/tessary/blob/main/setup.md

docker compose -f oci://docker.io/tessaryai/tessary:compose up -d -y