Every AI change proven before it ships. Every model release regression-checked.
Metrx is built to run the trial, state the result on your traffic, and record how it was produced. You set the authority for each change class; nothing promotes outside your contract.
AI configuration changes ship on judgment, not evidence.
Each change is a pre-registered randomized trial with a reproducible record.
A provider model update lands with no regression gate.
The Assure loop re-verifies incumbents on release and rolls back on contract violation.
Spend and behavior spread across teams with no shared account of them.
One evidence ledger for every workload — wins, losses, and rollbacks, with provenance.
Two loops, always running.
Find, prove, and apply better configurations.
Generate candidate configurations — model, prompt, parameters, routing — and prove model and route switches with pre-registered randomized trials on your traffic. Promotion only follows causal evidence.
Guard every configuration that's already live.
Watch every live configuration for drift, re-check incumbents when a provider ships a model release, and roll back when a breaker trips. The last proven config is restored automatically.
Authority you define, per change class
Runtime routing, prompt and cache config, workflow thresholds — each change class carries its own permission. A reversible model rollback and a prompt rewrite are never treated as the same kind of change. You set the matrix; only model and route switches can promote at all today — every other class stops at a proposal for your review.
An account of every workload
Agent-level spend attribution, the acceptance contract in force, and the evidence behind the current config — in one place, on your paper, ready for review.
Safe by construction, not by promise
A breaker reverts on technical failure on its monitoring cycle. A slow governor holds semantic regressions to a pre-registered exposure budget. Every affected task stays traceable.
Each promotion seals a reproducible evidence record — inputs, splits, readouts — that you or your auditors can re-run. Nothing rides on trust in a pitch.
Explore a live workspace →Put every workload on its best proven configuration.
Connect a workload, approve an acceptance contract, and let the loops run.
Figures shown in the demo workspace are synthetic. Every number carries its provenance.