Helpers prepare the work. You decide what actually happens.
Helpers can research, draft, and propose — but sending, deleting, and calendar changes wait at a checkpoint. You see the draft, the reason, and the source before it goes anywhere.
With AI tools that act autonomously, something consequential can happen quietly. A reply goes out, a file gets deleted, a meeting gets changed — and you find out afterward.
Why this is different
SwarmMarshal holds consequential actions at a review checkpoint. You see the proposed action, open any source message behind it, and choose to approve, edit, or reject.
When to use it
You need this when...
Reviewing a drafted reply before it is sent.
Stopping broad or destructive actions unless explicitly approved.
Seeing why an agent thinks an action is ready.
Teaching the system what can happen automatically next time.
Real examples
Here's exactly how it works.
Each example starts with a situation you've probably been in — and ends with the answer and where it came from.
Sensitive reply
Situation
An agent drafts a message involving money, dates, or a public promise.
You do
Review the draft and the cited sources.
SwarmMarshal does
Holds the send action until approval.
Output
A send-ready draft with visible reasoning and source links.
Rule promotion
Situation
The system notices a repeated low-value sender pattern.
You do
Approve or reject the proposed rule.
SwarmMarshal does
Explains what would move and why it is low risk.
Output
A deterministic rule only if you accept it.
Calendar change
Situation
A follow-up requires moving a meeting or creating a deadline.
You do
Review the proposed calendar action.
SwarmMarshal does
Shows the source message and impact before writing the change.
Output
An approved calendar/task update.
How to use it
Typical workflow
Let agents prepare drafts or proposed actions.
Review the source, risk label, and exact action.
Approve, reject, edit, or ask for a different approach.
Promote narrow repeat behaviors only after they have earned trust.
How it works
Behind the scenes
Actions carry risk and side-effect metadata.
Policy checks decide whether approval is required.
The UI shows pending approvals and blocked tasks.
Approval decisions become correction signals for future behavior.
What you get out
The result is something you can actually use.
Pending approval
Risk label
Draft action
Rule proposal
Audit trail
See more
SwarmMarshal can do a lot more with your communication history.
Each feature solves a specific, recurring problem — and every answer links back to the real message behind it.
