Skip to the content

Tour the fleet console in the demo

Walk Central Lite's console on a fictional company's synthetic data: its machines, an incident, a kill order, costs, outbound inspection and access.

Built, not yet claimed preview: this guide describes a package that has not passed its release gate.

Open the fleet console in your browser with nothing to install, on Halden Robotics, a fictional company whose machines, people and spend are synthetic, and follow one path through it: the inventory, an incident from its first finding to the kill order that stopped it, a kill order of your own from request to lift, the costs, outbound inspection, and who may do what. Nothing you do there leaves your browser.

1. Open the demo

Go to the console demo. It is the real fleet console with one thing changed: instead of talking to a Central Lite server, it answers itself in the page, from a synthetic world generated when the page loads, so its dates are always recent. You are signed in as an owner, so every section shows. Anything you change stays in this tab, and a reload starts afresh. Where the console would ask for your passkey, the demo shows a sheet that says what the passkey would confirm.

2. Read the inventory

Inventory lists Halden's 40 machines: Macs, Windows laptops and Linux workstations, build runners and CI jobs. Each row gives the machine's sentinels, its agentgarmd and policy versions, its key custody and its last report, and the card below the table counts the machines that reported AI tools AgentGarm does not govern. The sentinels and custody are examples, not what each platform has today: that is on coverage.

3. Follow an incident from finding to kill order

In Incidents, enter the task id 9ad68266-2f36-4d0d-94b3-e0b84a7fec55 and choose Show the timeline. It is a Claude Code task on a developer's laptop: secrets redacted out of a model request, a blocked write to the agent's own policy, the assurance level falling, then a kill order requested, approved by two approvers and applied, and the task suspended. The timeline also names the records the laptop could not send, as a gap, rather than hiding them. The assurance levels it shows are examples; what each agent and platform is rated today is on coverage.

See also

Investigate a task across the fleet

Read a task's timeline in Central Lite with its gaps, export a bundle the org root's evidence key signs, verify it offline and hold its records.

4. Request, approve and lift a kill order

In Kill orders, put host:ae01b0cb-4b76-4b9c-8766-d8e260c2d2ba (the machine build-runner-01) in Scope, give a reason, and choose Show what it reaches: one node. Request with a passkey opens the demo's sheet in place of your passkey; confirm it.

An order for one host needs two approvers. The new order opens below the list: once the console has checked its digest, choose Approve with an approver key and confirm the sheet, twice. The order is issued, and the runner answers that it applied it. To let its tasks run again, give a reason under the order and choose Request a lift with a passkey, then open the lift from the list and approve it twice the same way.

See also

Stop agents across a fleet

Suspend, terminate or lock down agents on one machine, one agent, one person or the whole fleet, with an order two approver keys sign and each node checks.

5. See what the agents cost

Costs shows the last 30 days by default: what was ingested from the machines, what the providers' own reports reconciled, the calls and tokens, and a daily trend. Change the range, or group By vendor account (reconciled), By principal or By credential. Every call through AgentGarm's proxy is priced from its catalog of models and prices, as on a real fleet; Cursor's and GitHub Copilot's spend, which runs on their own backends, is what their usage reports bill, read by their connectors.

See also

Attribute AI spend by machine and person

Read your enrolled machines' model spend by day, machine, provider and model, and reconcile it against your providers' own usage and cost reports.

6. See what agents upload to listed sites

Outbound inspection (Beta) shows the paste, webhook and file-sharing sites Halden's policies list: what the agents sent to each, what was stopped, and the machines not yet inspecting and why. The decisions at the top include a Java tool that refuses the certificate on four Macs. Open Edit sites to see the change a policy would need, and the replay that shows what Block would have stopped before you switch to it.

To see the first run an admin gets before any site is listed, open the section with nothing listed at /demo/?inspection=off#inspection and choose Choose sites.

See also

Roll out outbound inspection across a fleet

List the sites your fleet's agents should not send secrets to, run them in Audit, read what Block would stop, then block through a canary rollout.

7. Check who may do what

Roles & access lists the admins and the role each holds, over the whole organisation or a group of machines, including one suspended when the identity provider removed its person. Single sign-on, SCIM and SIEM show how sign-in, provisioning and the forwarding of security events are set up, with one SIEM sink lagging behind.

See also

Give people the right access in the console

Give each Central Lite admin one of six roles, for the whole org or a host group, and know who may request a kill order and who holds each approver key.