Behavioural drift detection
Continuous monitoring that detects when your system's behaviour shifts — after a provider update, a silent model revision, or a gradual change in traffic — and tells you before your users do.
Changelog
4 releases · Latest 14 Jul 2026
The record of changes
What shipped, when, and why it matters. Dated, versioned, and written to be read after something changes.
01 — Releases
Declarative guardrail policies with human escalation
Define guardrails as versioned policy files — input and output checks, tool constraints, and uncertainty-triggered escalation to a human reviewer — enforced at the harness boundary.
Eval suites as merge gates
Run evaluation suites in CI and block merges on behavioural regressions — with golden sets, configurable thresholds, and side-by-side diffs for provider migrations.
Trace Explorer with step-level replay
Inspect agent runs step by step — every model call, tool invocation, and retry — and replay any step in isolation to reproduce failures without re-running the whole pipeline.