Quicksilver Uses Structured Company Models and Auth Kernels to Govern AI Decisions

Quicksilver is an autonomous company operating system that models organizational knowledge — including entities, policies, workflows, and objectives — as structured content in Sanity, rather than relying on document search. An LLM proposes actions based on this structured model, but a deterministic TypeScript authorization kernel — containing no LLM — decides whether each action can proceed autonomously, needs human approval, or is hard-blocked. Risk scoring, policy conflict detection, and capability checks are all computed by the kernel, with an independent reviewer model offering advisory input only. Workflow definitions stored in Sanity include executable process states and guards, allowing the kernel to enforce decision lifecycles without redeployment. A single editable field in Sanity controls the system's autonomy ceiling, meaning administrators can make the system more cautious at any time, but cannot override the kernel's own safety thresholds through content alone.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)
Log in to join the discussion and vote.
Log in