How it works
One message, from arrival to decision.
This is the whole path. It is worth reading once, because everything Sedge shows you afterwards is a view onto some part of it.
A message arrives
In a group you marked as protected. Nothing else is looked at.
It is queued for analysis
The body travels through the queue and is never written to the database.
The policy is resolved
Platform, then organisation, then WhatsApp account, then community, then group. The most specific layer wins.
Rules are scored
Rolling windows, duplicates, links, mentions, media, posting pattern and the member’s own history each contribute their weight.
The total meets a threshold
The highest band the score reaches decides the action: nothing, log, warn, review, or remove.
Recorded, nothing done
- The event is written with the action it would have taken.
- The member is not contacted and nothing changes in WhatsApp.
- You get a real preview on your own traffic.
Carried out, then reviewed
- The action runs against WhatsApp and the outcome is recorded — including a failure.
- A removal opens a review case with all the evidence attached.
- A moderator can reverse it.
Setting up
Eleven steps, and you can stop after any of them.
The connection wizard saves where you got to. Close the tab, come back tomorrow, and it resumes on the step you left.
Connect an account
Choose a provider and authenticate. The official Cloud API uses credentials; the linked-device adapter shows a code to scan, and asks you to acknowledge the account risk first.
Test the connection
Sedge checks it can read what it needs to before you go any further, and tells you plainly if it cannot.
Discover communities and groups
Sedge lists what the connected account can see. Nothing is protected yet, and nothing is changed in WhatsApp.
Select the groups to protect
Only what you tick. Groups you leave alone are not evaluated, not counted against your plan, and not stored beyond their name.
Choose review groups
Optionally nominate a group where removed members are invited while a decision is pending. Separate ones for spam and inactivity, if you want them.
Configure the spam rules
Start from the platform defaults and adjust the weights, the thresholds and the windows.
Configure inactivity
Pick the removal day count and which warnings lead up to it — or leave inactivity switched off entirely.
Dry run
Replay the policy against recent traffic and see the score every message would have produced, before anything is live.
Watch in monitor mode
Events accumulate with the action they would have taken. This is the step worth being patient about.
Review the results
Look at what would have been removed. If a rule is too eager, change its weight and watch the effect on a real message.
Enable enforcement
Group by group, with a confirmation that names what will change. Every removal from then on opens a review case.
Modes
Three states, and only one of them touches WhatsApp.
Monitor runs the identical code path as enforce. The only difference is whether the resulting action is executed — which is what makes a dry run a real preview rather than an approximation.
- Monitor
Rules run and events are recorded, but nobody is removed and no message is deleted. This is the safe default for a new group.
- Enforce
Rules run and the configured actions are carried out in WhatsApp.
- Paused
The connection stays up. No rules are evaluated and nothing is recorded.
A group becomes protected in monitor mode. It never starts enforcing on its own.
After a removal
What actually happens to the member.
WhatsApp has no atomic “move”, so Sedge does not pretend to have one.
The member is removed from the protected group, and a review case is opened with all the evidence attached.
If you nominated a review group, Sedge tries to add them to it.
If their privacy settings do not allow being added, an invitation is sent instead and the case shows “invitation sent, waiting for them to accept”.
If neither works, the case says so, stays open, and nothing about it is reported as a success.
A moderator decides: restore, keep removed, whitelist, ban, or ask for more information.
Start in monitor mode. Decide later.
Nothing is removed until you say so, and every decision after that is reversible.