Expected ≠ observed.

Why we built it

Every team we spoke to had the same story: the assistant was great in the demo, then one day it said “done” and nothing happened — or the wrong thing happened. Nobody could tell which from the logs, because the logs were the agent’s own words.

COLVO grades the side effect. In testing it replays scenarios against an isolated sandbox and checks the resulting state. In production it stands between the agent and the provider, applies a mandate, holds risky actions for a human and verifies the outcome. The reply is treated as what it is: a claim.

Who we are

COLVO is built by VLN Technologies, a software company based in Italy. VLN builds and operates products for teams that run AI in customer-facing workflows. vln-network.it

How we work

• Deterministic first. A verdict rests on explicit checks against real state; a language model may comment, never decide.
• Say the limit. Our pricing, security and status pages state what is not there yet.
• Evidence over dashboards. Everything COLVO shows is traceable to an append-only record.