behavior_drift is one of Tessary's built-in classifiers. It learns how an agent normally behaves from its own history, the steps it takes and the order it takes them in, and flags traces that depart from that. It doesn't judge whether any answer was correct.
It catches the changes correctness checks miss: an agent that starts skipping a step it always ran, calling tools it never used, or taking paths it has never taken. The baseline is fitted per call site from its own traffic, so a new project is quiet by design until normal is established. New patterns graduate into normal only after recurring across many sessions. Firings are grouped by cause and triaged, and a confirmed finding opens a case.
7 questions
Answered, plainly.
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.
- 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