Durable state
Tasks, evidence, decisions, and approvals survive beyond one prompt or model response.
Evidence
No borrowed logos. No internal experiment disguised as client work.
Corteus is early. This page shows public artifacts and the operating discipline we have actually built. Named commercial outcomes will appear only when they exist and can be disclosed.
Public artifacts
Small tools that make production claims easier to challenge.
01 / GitHub↗
Provider dependence
wrapper-test
An open-source test surface for examining whether an AI product has durable product behavior beyond one model-provider wrapper.
02 / GitHub↗
Model change exposure
model-rot
A public artifact for making silent behavioral change across model updates visible and reviewable.
03 / GitHub↗
Evaluation coverage
eval-audit
A public artifact for checking whether an eval suite represents the workload and decisions a production system actually carries.
Internal operating system
We built the operating layer before claiming autonomous work.
This is an internal Corteus system, not a customer result. It coordinates research, review, approvals, monitoring, and evidence while keeping external authority bounded.
Bounded agents
AI capabilities receive the minimum tools and authority needed for a defined contribution.
Deterministic control
Queues, retries, quotas, idempotency, and state transitions remain code-owned.
Human authority
External and reputational actions require an exact, content-bound approval.
What comes next