WhippleScriptBuild your own Claw.
Give it a goal, a memory, a schedule, a message channel, and a queue of work. It keeps a record of what it did. It brings you in when it needs a decision. The core of that loop is 24 lines.
For people who build automations
Build AI workflows that do real work.
You describe the work that you want done. An AI writes the workflow. WhippleScript makes sure that the workflow is safe to run. You do not write code.
morning-ops
08:00
New requestSupport email received
✓08:01
Research accountCustomer record and recent activity
✓08:02
Draft responseApproved model prepares a reply
✓08:03
Your approvalSend this response?
?What is new
An AI could always write you an automation. Nobody could tell you if it was safe.
That is why your IT department says no. WhippleScript answers the question with a compiler. It checks each workflow against the rules that your organization already set, before the workflow runs. Your IT department can now say yes.
Turn an inbox into a process.
It reads a request, collects the documents, drafts a response, and routes the work. Nothing goes out until you approve it.
Keep working after you close the window.
It researches, runs the comparison, and continues until it reaches an answer or a stopping point that you set.
When the work is consequential
Connect the things that are usually difficult to connect safely.
The useful workflows combine private records, the open internet, financial systems, internal documents, external messages, and language models. Each combination is a place where an automation can do damage.
Ask for a workflow that reads a customer account, checks current information on the web, prepares a payment request, and asks you before money moves or a message goes out.
Customer record+Web research+Approved AI+Your decision
Before the workflow runs, WhippleScript checks the connections. It tells you when private information would reach the wrong model. It tells you when information from the web or an inbox could steer an important action. It tells you when the workflow needs a person to decide.
How it works
Describe the routine. Keep control of the parts that matter.
- Open GaugeDesk.GaugeDesk is the workbench. You install nothing. WhippleScript runs below it as the governed engine.
- Describe the work.Use your own words. Start with the routine that you want to automate.
- The AI writes the workflow.It runs the workflow against a test provider first. You see the result before the workflow touches a real system.
- Choose the boundaries.Decide what the workflow can connect to, and when it must ask you.
- Come back to a process.See the progress, review the decisions, and change the workflow as the work changes.
Bring it to your organization
Two objections stop most automation projects. This page answers the first one.
The second objection belongs to the person who approves your tools. Send that person one page. It states what the compiler refuses to build, and it names the evidence for each refusal.
“Can I let my team build their own AI automations without reviewing every one of them?”
Policy set one time→Anyone can author→Compiler checks each one