One shared room per project, one governed brain behind every answer. Every answer cited, every memory owned, nothing becomes canon until a human says so. The human + AI workplace.
Invite-only beta. See how the loop works
Like any good teammate.
Them learns from the work you already have: connect Notion, Confluence, GitHub, Jira, Linear, any docs page, or your own files. You choose every source it reads, and you can disconnect one at any time.
Everything it learns becomes governed memory: facts, entities, decisions, processes, templates, and skills, each with an owner, citations to source, and a status. Nothing becomes canon without a human.
The Coworker drafts from approved memory in your firm's own style, and the Feed hands that memory to every AI app over MCP. Acting inside your tools is next, and it will ask first too.
Your people and your agents, on one brain. Them puts what it learns to work on every surface they touch.
A shared room per client where your team and your agents work the same engagement. Everyone sees the thread, the agent answers from that client's memory, and any answer becomes canon.
Dana
They are pushing for a bigger discount. What did we agree?
Them
Capped at 20% without VP sign-off. Above that the deal desk owns it.
Notion, Confluence, GitHub, Jira, Linear, a docs page, or a pile of files. Each one gives up something different, and only the part worth keeping.
GitHub
why it was merged
Jira
how it was resolved
Notion
specs & policies
Confluence
runbooks & standards
SharePoint & Drive
on the way
Them clusters your sources and drafts memories with verbatim citations. Nothing becomes canon without a human.
Refunds over $500 need a signed renewal.
“anything over 500 we get the renewal signed first”
Policy v3 · Confluence
Cited answers from approved memory only. When the answer is missing, the question routes to the right person, and their reply is kept forever.
What's our refund window?
30 days for standard orders.
Your playbooks, templates, and methods, written once and reused everywhere. Approved skills are served to every AI app your team uses.
Owned by Dana O. · served to 4 agents
One governed memory, many scoped feeds. Claude, Cursor, and your own agents subscribe over MCP. Every pull is logged.
Claude read “Enterprise discounts cap at 20%”
Northwind feed · 2m ago
Cursor read “Deploys ship Thursdays, never Fridays”
Engineering feed · 14m ago
Organize the brain by client and engagement. An agent keyed to one client can never read another's memory. The wall is enforced, not promised.
Every memory has an owner, a citation, and a review date. Them asks if it is still true every 90 days, so the brain never quietly goes stale.
Enterprise discounts cap at 20%.
Owned by Maya R. · approved 12 Mar
Still true? Next check 10 Jun
Trial length is 14 days.
Dev P. said no at the 90 day check
Connectors
Just scattered across a dozen tools, and nobody is going to reorganise them. Them reads each one for the thing it is actually good for.
The standard, as somebody actually wrote it down.
Notion
Confluence
Any docs page
Your files
Why the change was made, not the diff. Git already has the diff.
GitHub
Jira
Linear
GitLab
The engagements your firm already delivered, in the folders they live in.
Google Drive
SharePoint
The thread where the exception was granted, never the small talk.
Slack
Teams
Gmail
Outlook
What broke, what fixed it, and what nobody should try again.
Sentry
PagerDuty
Datadog
Grafana
Everything above flows in. The Feed flows out, over MCP, every pull logged against the key that made it.
Claude, Cursor, ChatGPT, and your own agents read only the approved memory their key allows. Missing yours? Tell us and it moves up the list.
Who it is for
If the answer walks out the door when someone leaves, Them is for you. We go deepest in two places: advisory firms, where the knowledge is the product, and engineering teams, where AI now writes faster than anyone can remember why.
Not here to replace your people. Here so your people, and every agent they hire, finally work from the same brain.
When a partner rolls off, the framework leaves with them and the next team re-solves what the firm already knew. Them keeps it: every engagement becomes governed memory your people and your agents both work from.
Join the waitlist
An agent keyed to one client can never read another's memory. Enforced at the protocol layer, not promised in a policy.
Your team and your agents work the same engagement in a shared room, not a thousand private chats nobody else can see.
Proposals and readouts written the way your firm writes them, following your own approved skills, grounded in real precedent and cited back to it.
Your agents now write more code in a week than your team can review in a month, and the reasoning behind it lives in merged pull requests nobody will read again. Them keeps the why: the decision, who made it, and what it was based on, served back to the humans and the agents that come next.
Join the waitlist
Architecture decision records, README intent, and the reasoning attached to merged pull requests become governed memory. The code is already in git.
Claude and Cursor subscribe over MCP, so the agent writing today's change knows what last quarter's decision was, and every pull is logged against its key.
Who approved this, when, and on what evidence. Every memory carries an owner and a review date, so a decision cannot quietly go stale.
It works the same way for agencies, accounting and finance, support, sales, and ops: anywhere the answer is known by someone and written down by nobody.
Somewhere else entirely? Join the waitlist and tell us how your team works.
One question, one loop. Them learns from the work your team already does, compiles cited evidence, drafts a recommendation, and a human decides. The decision becomes memory the company keeps forever.
Them reads the sources you connect: Notion, Confluence, GitHub, Jira, Linear, any docs page, and the files you already have.
Cluster sources into cited evidence, with owners and provenance.
Draft a recommendation and a ready reply, never an invention.
The right human approves, and the decision becomes governed memory.
Search finds. Dashboards measure. Them governs. That is not a feature here, it is the product.
Nothing here is autonomous. Every memory and every skill is drafted, then approved by a named human before any agent can read it, and every pull an agent makes is logged against its key. Earned autonomy is where we are going, one proven skill at a time, and it will never skip the approval.
Your memory belongs to your workspace and never crosses into another. Answers are built only from what your team approved, every agent key is scoped to what it may read, and you can disconnect a source or delete a memory at any time.
Grounded drafting is live today. Next the Coworker works your actual tools: fill the form, file the ticket, update the record. You approve, Them clicks, and every action lands in the audit log cited back to the memory it came from. It will ship the way everything else here did, with your approval, always.
The examples rotate. The loop does not. Every team runs on questions it has answered before.
A proposal drafted from your templates and past case studies, cited back to each source. You edit and send.
Who approved this last time, and why? Them surfaces the governing rule, the past exceptions, and the owner.
Day one, a new hire asks anything and gets the cited answer. Nobody gets interrupted.
Every team runs on questions it has answered before. Them keeps the answers, cited and owned, and hands them back wherever the question shows up next.
“Will we still know this after people leave?”
“Who approved the last refund exception, and why?”
“How do we handle enterprise discounts here?”
“What is the approved policy, and who owns it?”
Invite-only beta. Onboarding design partners now.
