Your group produces it a piece at a time, on different days, without anyone noticing it adds up. GroupWisdom reads across all of it and says the thing no single piece of work said on its own, while it still matters.
A REST API and SDK, plus ready-made agents for Slack, Microsoft Teams and Buzz.
Conversation is mostly coordination, acknowledgement, questions and chatter. Something that responds to all of it is noise wearing a badge. GroupWisdom surfaces a finding only when every one of these holds:
It joins two pieces of work made separately. A message and the reply it answers are one conversation, not a combination.
Nobody has said it yet. Restating, agreeing, or narrating what just happened doesn't count.
Knowing it changes what someone does next. True but inert isn't worth the interruption.
It can point at what produced it. Specific contributions, named people, real numbers.
If all four aren't true it says nothing at all. That is the normal outcome, and the correct one.
An ordinary message. No card, no label, no preamble.
Two contributions, one finding. Neither of them could see this from their own work. It doesn't smooth the disagreement into agreement. The difference is the finding.
It hands over finished work. Leo isn't told to go and check with Dana. He gets Dana's actual finding, and starts two steps ahead.
One endpoint takes contributions, tagged with who made them. Analysis runs on its own moments later, and findings come back by signed webhook or whenever you ask for them. It is built around a group rather than around a question someone typed, and that difference shows up in everything below.
# Send a contribution, tagged with who made it curl -X POST https://testgroupwisdom.com/v1/projects/:id/ingest \ -H "Authorization: Bearer gw_yourkey" \ -H "Content-Type: application/json" \ -d '{ "title": "Users keep abandoning checkout on mobile", "content": "Support tickets up 30% this month.", "type": "note", "contributed_by": "Sarah" }' # Read the findings back whenever you like curl https://testgroupwisdom.com/v1/projects/:id/wisdom \ -H "Authorization: Bearer gw_yourkey"
Official JavaScript and TypeScript SDK, with no runtime dependencies. The full reference is in the docs.
A finding has to join something new to something the group already holds. Summarising one person's contribution is never wisdom, however well written, so a busy project produces far less output than you would expect.
Every item carries who contributed it, and every finding names them. Ask what the system believes each person has established, and it will tell you.
GET /gate-records returns what was considered and rejected, and the reason in full. Silence is checkable rather than mysterious.
An agent that researches or drafts is a contributor exactly as a person is. One person working with one agent is a group, and their work can be joined the same way.
A team owner uploads it and adds it to one team. It reads that team's standard channels — never private channels, never chats it wasn't added to, and nothing at all until someone enters a pairing code. No tenant-wide administrator is involved, and removing the app from the team stops it completely.
One zip. Your Microsoft 365 admin uploads it once and it appears in the Teams store for everyone. Or upload it yourself if you run your own.
It joins that team and posts a Connect this team button in the channel where it was added. Nothing is read yet.
The code fills itself in. Sign in, choose one project you own, and only then does it start reading.
As a reply under the work it came from, where the team already is — not in a dashboard nobody opens.
@GroupWisdom memoryEverything it believes this team has established, and whose work each piece came from. Scoped to the channel you ask in.
@GroupWisdom demoA one minute scripted example on invented messages, so a team can see it work before it has enough of its own.
@GroupWisdom muteSilence for now, or mute today for the rest of the day. It keeps reading and still answers when asked.
An admin adds the zip once and it appears under Built for your org in the Teams store, no approval step. Anyone can then add it to a team.
What it says is built only from what that team's standard channels can see. Private channels and chats are never read.
We're letting teams in a few at a time while we watch what it costs to run. Ask and we'll set you up.
Upload the zip, add it to a team, enter the code. About five minutes.
Add it to your workspace from our site, then /invite @GroupWisdom wherever real work happens. Each
channel it joins becomes its own project, with its own memory, so what it learns in one channel never shapes what
it says in another. It never reads private channels, direct messages, or channels shared with another company,
and it starts from the moment it is invited rather than reading back through history.
Sign in on our site, agree to what it keeps, and press Add to Slack. The workspace is tied to your account; there is no code to type.
It says hello, explains what it keeps and for how long, and starts reading. Nothing else in the workspace is read.
Under the message that completed it, also shown in the channel, with Helpful, Wrong and Late buttons underneath.
/groupwisdom memoryWhat it has picked up in this channel, with the message each fact came from. Only you see the answer.
/groupwisdom muteQuiet until unmuted, or mute today until midnight UTC. It keeps reading meanwhile.
/groupwisdom demoA one minute scripted example on invented messages. None of it is kept.
One press under any finding records whether it was worth saying. That is how we find out whether it works for you.
Separate memory, separate quiet period, separate sense of what is new. A busy channel never drowns out a quiet one.
We're letting workspaces in a few at a time while we watch what it costs to run.
Sign in, add it, invite it to a channel. About two minutes.
GroupWisdom joins your community with an invite link — the same one you would send a colleague. It appears in the member list, reads only the channels you tick, and is removed the way anyone else is. Nothing to install, nothing running on your machine, and no settings left behind when it goes.
Your community name, then Invite to community, then Copy link.
The link already says which community to join, so there is nothing else to fill in.
Every channel it can see, with a tick box. Untick anything you would rather it stayed out of. It never reads direct messages.
As an ordinary message in the channel where the work is happening — not in a dashboard nobody opens.
@wisdom memoryEverything it believes this channel has established, and whose work each piece came from.
@wisdom demoA one minute scripted example on invented messages, so a community can see it work before it has enough of its own.
@wisdom muteSilence for now, or mute today for the rest of the day. It keeps reading and still answers when asked.
Reacting to a finding records a verdict on it. Removing the reaction withdraws it.
What it says in a channel is built only from what that channel can see. Another channel's work stays there.
Buzz communities run agents beside people. An agent's output is a contribution on exactly the same terms.
One invite link is the whole setup.
An agent that researches, drafts or analyses is a contributor exactly as a person is, and its work counts the same. A research agent's output meeting a colleague's findings is wisdom on the same terms as two colleagues' work combining. What it needs is two pieces of work produced separately, not two people: one person and the agent they are working with is a group, and an agent that drafted a budget for one question and a schedule for another can disagree with itself across them. It never discounts a contribution for coming from an agent.
In Slack, only the channels you invite it to. In Buzz, every channel it can see is listed with a tick box. In Teams, one team, connected by its owner. In the API, the items you send and nothing else. What it says in one channel is drawn from what that channel can see.
Content is encrypted at rest. Raw messages are deleted after 30 days; what remains is the distilled memory it reasons from. Nothing read is used to train any model.
Every message it reads gets analysed. Pointing it at the two or three channels where real work happens costs a few dollars a month. Pointing it at a busy company-wide workspace costs a great deal more, for wisdom about lunch plans.
Connected and saying nothing is working correctly. It needs at least two related contributions before there's anything to connect, and it would rather stay silent than pad.
An API key, a Slack or Teams app, or a Buzz invite link. It stays quiet until the work adds up to something worth saying.