kiff.dev Each clip starts with a situation a team recognizes, then shows KIFF deciding the action before it runs — allowed, held for a human, or blocked. Real product, real decisions, signed receipts.
Before you trust an agent with real systems, you need to know every consequential action it can already reach.
In this clip A static scan of the agent's own code lists the actions a model-controlled input can trigger — before and after they're governed — and flags the ones that are legitimate in one state and catastrophic in another.
You have an agent making tool calls today, and nothing decides whether a given call should happen.
In this clip Pointing an existing agent at KIFF with one call — no rewrite — and watching it show up as a connected, governed runtime.
Refund limits, approval thresholds and state rules are spread across services and hard to audit.
In this clip Authoring a real refunds domain with a coding-agent skill, then validating and activating it — one versioned contract your agents decide against.
An agent acting on production can do the right thing at the wrong moment; you need the refusal to be real, not hoped-for.
In this clip The control room deciding each action against live state — allowed, held for a human, or blocked — with a signed receipt for each.
Weeks after an incident, compliance asks exactly what happened and what stopped it. "Trust me" isn't an answer.
In this clip The per-domain activity trail and a signed, tamper-evident receipt: the action, the state it was checked against, and the verdict — recorded before anything ran.
Connect an agent, govern one consequential action, and see the decision in minutes.