Observation First
Every new group logs rule matches before it can delete. The minimum observation period is 24 hours.
GroupGuard gives WhatsApp admins 14 local moderation rules, a guided setup, and conservative defaults. New groups start by observing. You decide when one rule can delete.
You want less spam. You do not want a moderation tool to put the WhatsApp account behind your group at greater risk.
Moderation should reduce the mess without creating a new emergency.
The default path earns trust before it deletes anything
Run the guided setup on your computer and link WhatsApp with a QR code. Use a number you can afford to lose, not your main personal or business account.
Select the groups GroupGuard may inspect. Each one starts with a conservative spam guard, admin exemption, no DMs, and no deletion.
Review at least 24 hours of matches. When the rule looks right, unlock enforcement from the server. Start with one group and one narrow rule.
GroupGuard favors a missed action over a repeated or uncertain account mutation. These controls apply before any deletion or outbound message.
Every new group logs rule matches before it can delete. The minimum observation period is 24 hours.
The group must leave observation mode and a local operator must open the enforcement lock outside WhatsApp.
Hard action budgets, timeouts, duplicate protection, and a shared failure circuit stop uncertain work from repeating.
Local guards need no Docker or AI key. The Claude agent, scheduled tasks, DMs, and typing indicators start disabled.
GroupGuard uses WhatsApp's unofficial linked-device protocol because Meta's official API does not expose consumer-group moderation. WhatsApp may restrict automated accounts. No tool can promise otherwise.
GroupGuard evaluates 14 local guards without sending messages to an AI. A fresh installation logs matches in observation mode. Deletion and private notices stay off until a local operator enables them.
Choose how many messages one person may send in a short window. GroupGuard logs every match first. After you unlock enforcement, it can remove the matching message and, if enabled, send a private explanation.
Set a minimum delay between messages from each person. In observation mode, GroupGuard records who would hit the limit. Unlock enforcement only after the timing fits the group.
Match messages that contain URLs while exempting admins by default. Review the matches first, then unlock removal if a no-links policy fits the group.
Choose the hours when members should not post. GroupGuard first records matching messages, so you can check the schedule and timezone before enabling removal.
Detect forwarded messages such as "send this to 10 people for good luck." GroupGuard logs them first. You can unlock removal after checking that legitimate forwards will not be caught.
Choose which content types are allowed, such as text only, no stickers, or videos only. You can also limit message length. New policies remain in observation mode until you review their matches.
Match specific words or phrases while respecting word boundaries, so "class" does not trigger on "classic." Review the matches before enabling removal.
Define who may post in an announcement-style group. Messages from everyone else are logged first and can be removed only after you unlock enforcement.
The optional Claude agent can schedule reminders and research updates in plain language. It stays disabled by default and requires Docker plus your own AI credentials. Core moderation does not depend on it.
A public GroupGuard Playground is in development. When it opens, you will be able to run a few fixed moderation scenarios inside a dedicated WhatsApp group and see the result for yourself.
The Playground is not live yet. Leave your email and we will send one update when it opens.
Follow Development on GitHubSelf-hosting is available now. A managed service is still in development.
Run GroupGuard on hardware you control and keep the account state local.
A future option for admins who do not want to maintain their own runtime.
Everything you wanted to know (and some things you didn't)
The self-hosted code is open source under the MIT license. You may still pay for a separate WhatsApp number, an always-on VM, or the optional AI provider. The managed service is not available yet and its pricing has not been set.
A self-hosted installation processes messages on the machine you control. The GroupGuard maintainers do not receive them. The optional AI agent can send selected triggered context to your configured AI provider, so leave it disabled unless you need it.
The same way WhatsApp Web works. You scan a QR code, and GroupGuard appears as another linked device on the account.
There is no ban guarantee. GroupGuard uses an unofficial linked-device protocol because Meta's official API does not support consumer-group moderation. Use a separate number you can afford to lose, keep the default limits, and stop if WhatsApp shows a warning.
At first, GroupGuard records the match and leaves the message alone. After at least 24 hours, a local operator can unlock deletion for one group. Private notices remain off unless the operator enables them.
Yes! By default, group admins are exempt from all moderation rules. You can turn this off if you want everyone treated equally, but most groups prefer letting admins post freely.
You need Node.js 20 or newer and a separate WhatsApp number. Use your computer for linking and a supervised test. For continuous operation, move the installation to a dedicated low-privilege VM. Docker is only needed for the optional agent.
Yes. The word filter accepts terms in any language. Other guards inspect message properties such as links, forwards, and media types, so they do not depend on the message language.
Very. Every protection has settings you can tweak (how many messages are allowed, during which hours, which words to block, etc.). Each group gets its own separate settings. And if you're technical, the code is open source so you can change anything.
Absolutely. One GroupGuard manages multiple WhatsApp groups at the same time, each with its own rules and its own scheduled tasks. Groups are completely separate from each other.
It is still in development. The site does not promise a launch date, feature list, or price. Join the waitlist if you want an update when there is something concrete to test.
Not yet. The Playground will be a dedicated public WhatsApp group with fixed, rate-limited demo commands. It will not expose free-form AI or delete visitor messages. We will add an invite only after the account, moderation coverage, and incident process are ready.
Leave your email and we will contact you when there is a concrete managed-service test. This list is separate from the Playground.
Link a separate number, choose one group, and review what GroupGuard would catch before it changes anything.