How it works · 8 min
Control room, tool factory, agent operating layer.
KAIRO has three layers. The control room sits on top: priorities, missions, decisions. The tool factory sits in the middle: 460+ specialist tools across four sectors. The agent operating layer runs underneath: a 601-strong company roster plus a dedicated crew of agents on every tool that turns an input into a hardened output.
The control room
The control room is the surface for founders and operators. It asks: what should move today? Then it tells you. Live signals, priorities, missions, and the work that needs your judgment. The control room is the place you spend most of your time.
The tool factory
Underneath the control room is the tool factory. 460+ specialist tools across four sectors: the autonomous revenue layer, buyer-side AI, vertical revenue OS for MSPs, and the B2B identity graph. Each tool is a focused, single-purpose surface with its own input, its own output, and its own agent team.
Tools are exposed at /tools/[id]. The dynamic runner serves all 460+ tools from one factory engine, so we ship new tools by adding entries to the registry, not by deploying new pages.
The agent operating layer
Every tool has its own dedicated 5-agent crew — distinct agents that work only on that tool, never shared with another. Each crew has a lead who signs off, specialists who do the core work, an analyst who gathers the inputs, and a verifier who checks before it ships. The company roster of 601 leads, governs, and reviews above them.
The 5-agent challenge loop
Before any output reaches you, it goes through five rounds of peer critique. Each round, a different agent attacks the current draft, surfaces gaps, and returns a stronger version. The output that ships is the one that survived. The full critique trail is visible on the run page.
The accountability gates
Every agent has an accountability mode. Customer-facing outputs are human-gated. Financial and irreversible actions are human-gated. Internal research, drafts, and analysis are auto-execute. The 601 agents split into 231 human-gated, 82 human-notified, and 288 auto-execute. The gates are encoded in code, not in policy.
How it all fits
You open the control room. You see a live signal that an account is heating up. You click into the corresponding tool. The 5-agent team runs the work through the challenge loop. The output comes back, gated by your approval. You ship it.
That loop, repeated, is the platform.