Skip to main content
CloudThinker builds Frontier Investigation Agents: agents that investigate problems across your cloud and propose fixes. Writes run only under your approval policy. This page maps each practice in the field guide to a product page, so you can check every claim before you run a pilot.

​
What CloudThinker is

Every workspace has one built-in assistant, also called CloudThinker. It splits large tasks across temporary subagents that work in parallel, and your team can add custom agents for its own systems. The agents overview explains how they fit together. Agents work through connections, the saved credentials CloudThinker uses to reach a system: AWS, Azure, Google Cloud, local clouds such as GreenNode and FPT Cloud, Kubernetes, databases, observability tools, and more. On-premise systems are reachable through VMware, OpenStack, SSH, or a Worker, a folder on your own machine where CloudThinker reads files and runs commands over an outbound connection. Each request runs through one loop: detect the problem, analyze the root cause, resolve it under your approval policy, and validate that the fix held (full explanation).

​
The products

All of them share the same connections, approval policy, and audit trail.

​
Where the field guide’s controls live

​
Run the 90-day pilot

Chapter 9 recommends starting recommendation-only and earning autonomy on evidence. In CloudThinker, week one takes three moves.
Prerequisite: changing the Approval mode needs the workspace Admin role. Other roles see the picker as view only.
1

Switch the workspace to Manual

New workspaces start on Auto · Rules + AI · Ask me, so change it before you connect anything. Open Approvals in the sidebar, then the Approval mode tab. Under Workspace mode, select Manual.If you plan to run Automations, also go to Run types, click Override on the Automations row, and select Manual. That row defaults to Auto · Rules + AI · Refuse and doesn’t follow the workspace mode.Success state: the picker shows Manual, and every write waits for a person. Reads still run.
2

Connect one bounded, low-risk system with read-only access

Pick something you can afford to experiment on, such as a staging Kubernetes cluster or one non-production AWS account. Use a read-only IAM role or service account, as in the AWS, Azure, or Google Cloud setup. Follow the quickstart to add it under connections.Success state: the connection shows a Connected status.
3

Record a baseline, then compare monthly

Export the last 90 days of mean time to resolve (MTTR) and pages per week from your paging and ticket tools, and note your monthly cloud cost. CloudThinker can’t measure the time before you connected it. Resolve Insights starts tracking once Resolve receives incidents from your alert source. Chapter 8 covers which metrics to pick.Success state: you have a dated baseline to compare against Resolve → Insights at day 30, 60, and 90.
For the later phases (Approvers, Allow Rules, and switching to Auto), follow §9.6. Allow Rules decide only once CloudThinker turns on Rule enforcement for your workspace; until then the Approval mode alone decides, so Manual is the control that holds every write today. Bring the eight data-control questions from §6.3 to your evaluation. For private deployment or procurement questions, contact the team.

​
Next steps

Quickstart

Connect a cloud account and run your first investigation.

Approvals

Set the approval policy before agents make any change.