Your AI keeps working on its own. When it hits something that needs human judgment, Ratifia routes the decision to the right person, captures the verdict, and hands the answer back — so your agent can keep going.
One call inside your workflow · No infrastructure to run · Every decision audited
One decision — wherever your humans already work.
verbto sign off on what your AI does — approve the action before it ships.
A play on ratify — to give something your formal stamp of approval.
AI is starting to act — not just answer.
An AI that drafts a claim summary is one thing. An AI that issues a $12,400 payout, changes a patient's care plan, deploys to production, emails a customer about a breach, or grants access permissions — that's something else.
Some of those should happen on their own. Some shouldn't. Ratifia is where you draw the line — and prove where it was drawn.
Ratifia was built for the actions with consequences — the ones a regulator, a customer, or an on-call engineer will ask about later.
No new agent architecture, and Ratifia never runs your code. Your workflow waits on the engine you already use — Ratifia steps in only when a policy says a person needs to decide.
Ratifia routes the decision to a person, waits, then resumes your workflow with the verdict.
You don't rebuild your agent. Add one line where a human should decide — Inngest, Temporal, Trigger.dev, or your own loop.
const decision = await ratifia.awaitApproval(ctx, { policy: "payouts-over-5k", route: { surface: "slack", channel: "#finance-approvals" }, context: { blocks: [ toolCall, evidence, aiOutput ] }, }) if (!decision.approved) return escalate() // ...your code resumes here, with the verdict
awaitApproval() — your engine holds the pause, for as long as a human takes.decision — approved or not, plus who, when, and why.The simple loop is backed by the controls a security, compliance, or risk team asks about on the first call.
Behind every decision is an actual governance product — a reviewer inbox, policies, roles, and an immutable audit log.
Get early access, or follow along as we build the control layer for AI actions in the open.
Product updates and the occasional deep dive on human-in-the-loop for AI. No spam — unsubscribe anytime.