Customize, a Frontier Investigation Agent
Works the way your team works.
Give the agents your rules, your runbooks and your tools. Build your own agents, teach them how your team works, and decide what they may do on their own.
Build your own agent.
Describe the agent you need in one sentence and CloudThinker builds it. Or fill in a short form. Then mention it by name in any chat.
Your instructions
Give it a role, a goal and your rules in plain language: the currency to report in, the language to reply in, what it should never do.
Only the access it needs
Pick the connections it may use, built-in or your own MCP servers. Nothing else is in reach.
The right model for the job
Choose a Light, Pro or Ultra model tier per agent, and set whether it may save what it learns as new skills.
Teach it how you work.
Turn your checklists, conventions and procedures into skills. Write one once and every agent applies it, every time it matters.
Skills, in plain language
Let CloudThinker draft a skill, write it yourself, or upload one. Attach it to chat, code review, incidents or assessments.
Start from the Skills Hub
Browse the open Skills Hub and install a proven skill in one click. Then adapt it to your stack.
Your team's commands
Save the prompts your team repeats as commands and run them with a slash. Code review also follows the rule files already in your repo.
Plug in your tools.
The agents already connect to your clouds, clusters, databases and observability. Add the internal tools nothing else integrates with.
Any tool with an MCP server
Point CloudThinker at your own MCP server. Allow each tool, require approval for it, or turn it off.
Your runbooks and docs
Upload runbooks, policies and docs, or point to a website. The agents look things up before they act.
Memory of your environment
The agents remember your environment and your team's preferences across conversations. Correct them in plain chat.
Run on your schedule.
Turn any instruction into an automation. The agents investigate when it is time, or the moment something happens.
On a schedule
Daily, weekly, monthly or any cron. A weekly cost report, a nightly health check.
On a webhook
Any system can start the work by calling a webhook. The agent receives the payload.
On a repository event
A push, a merge request, a comment or a pipeline result starts the run, filtered by branch.
You decide what runs alone.
The agents find the why and propose the fix. You set exactly what they may do without asking.
Manual or Auto
In Manual, set each tool to always allow, needs approval or never allow. In Auto, every write is checked before it runs.
Rules for every runbook command
When Resolve follows your runbook, each command that changes something is allowed, waits for approval, or is denied.
Your data, masked
Choose which sensitive data is masked before the agents see it, and add your own identifier formats.
And the details that make it yours.
Engineers get their time back. For creative work.
Agents investigate every issue across application, cloud, security, database and Kubernetes, find the why and propose the fix. Your engineers approve, under your policies and audit trail.