Questions, answered.

Them AI is the collective intelligence workspace. People and AI teammates work together in rooms, on one company memory that your people approve and every answer that uses it cites.

Keep the facts and decisions worth reusing as memory, and write repeatable methods as skills for the next person or connected agent. What the company learns stays with the company, even after people move on.

A room is a shared place to keep data and teach your AI team around a particular job: one stream where people and AI teammates talk, with threads beside it, and the documents, examples, customer history and tasks next to the talk.

People work through cases together, ask AI teammates for help, and explain corrections. Use those examples in the selected teammate's training and review the proposed rules. The room retains the context; approved training rules belong to the teammate you teach. Ordinary messages do not automatically train every teammate.

See how rooms work.

No. A room is where people and AI teammates do the work, with the tasks, files and drafts beside the talk. A chat tool connects to a room rather than replacing it.

Planned bridges would let multiple Slack or Microsoft Teams channels connect to the same room, with sharing choices for each connection. Each would be a way into the room's context, not a separate copy of it.

AI teammates are built and trained in Them AI. Give each a job, choose its model, and choose the person it runs with. For example, Riley can answer a pricing question while Jamie prepares a support handoff.

External agents run in other tools and connect through MCP, a protocol for sharing tools and context. Each connection has its own key and access scope. Workflows put steps in order and can pause for a person's approval.

By default, a teammate reads approved shared company memory and its assigned room's data. Access to additional rooms needs an explicit grant. Shared company memory can be turned off for a teammate.

Client boundaries still apply. These AI retrieval settings do not decide which people can open a room: a room is open to everyone in the workspace unless it is members-only.

Yes. A teammate can use the workspace default or a configured model connection. A workspace key uses your account with the selected provider. Without a workspace model key, the platform default uses Google (Gemini) under Them AI's account.

If an authorized member enables platform fallback for a connection, an unsuccessful request may be retried through that platform account. A configured connection without this permission does not use platform fallback. Changing the model does not change data permissions or human approval requirements. Read about model processing.

Give a teammate an example, explain the correction, and review the training rule it proposes. An approved rule guides that teammate's later work. It does not automatically become a rule for every teammate.

Your team can separately review useful facts as shared memory and repeatable methods as shared skills. This reuses knowledge and instructions; it does not retrain the underlying model after each conversation.

Memory keeps facts, decisions, sources, and context. Skills preserve repeatable ways of completing work and making decisions.

Memory can tell an agent what happened last time. A skill helps it decide what to do this time.

No. An AI persona is built from the decisions, reasons and methods a person chose to leave with the company, reviewed by the company. It is never a recording of a person and never their identity.

A useful example might explain why a support lead checked payment settlement before promising a refund. That reasoning keeps helping the team. It does not recreate the person.

A chatbot usually helps one person in one conversation. Them AI is built for an entire team.

People and several AI teammates share rooms, context, decisions, and reusable skills. What the team approves can improve how every authorized agent works.

Yes. Authorized people can review, approve, correct, and assign owners to memories and skills. Roles control who can contribute, govern, manage connectors, and manage the team.

Roles decide what each person may change, not which rooms they can open. A room is open to everyone in the workspace unless it is members-only, and approved company memory is open to everyone. Members-only rooms and direct messages open only to the people in them, and your own chats with Them or an AI teammate are yours alone.

Yes. Them AI works with Claude, Cursor, and compatible MCP agents through an authenticated MCP connection.

Knowledge-only keys retrieve approved memory and approved skills. They cannot read room conversations or post messages. Keys with workspace access can also read permitted room conversations, post messages, and invoke AI teammates where allowed. Those conversations can contain messages that have not been approved as company knowledge.

Review the key's access mode and scope. Setup and available MCP features depend on the AI tool you use.

They can be. A room, a client's room included, is open to everyone in the workspace unless whoever makes it chooses Only people you add. Members-only rooms and direct messages open only to their members. An owner or admin who is not in a members-only room sees it as a locked row and may manage its members, but never reads it. Drafts and room knowledge kept in a members-only room stay with its members; once a person approves a draft as company memory, the whole workspace can read it. A client label alone is not a privacy boundary between people; it keeps AI retrieval apart.

For approved knowledge retrieval, client-scoped MCP feeds limit agents to that client's approved records plus approved shared company material. Keys with workspace access can additionally read permitted room conversations, including unapproved messages. Check human access, key mode, and feed scope separately before importing client work.

People with the required governance permissions can delete memory records and skills. Removing a connected source requires permission to manage connectors. Deleting a source stops future syncs but does not delete memories already created from it.

A client record with linked workspace records cannot be deleted while those links remain; it can be archived instead. Archiving does not delete the linked content. Ask your workspace administrator for help with permissions, or contact hello@usethem.ai about a personal-information request or deleting a whole workspace. Read the privacy policy.

Information is protected with encryption in transit and at rest and with access controls. Them AI is designed so that memory belongs to one workspace only, agents read only what their key allows, and feed access is logged.

Your content is not used to train our models or the model provider's models. Model processing happens only to serve your workspace. The beta's security posture continues to mature, so evaluate it before connecting sensitive data. Read the privacy policy.

External agents run in their own tools and can use Them AI as a source of approved context. Them AI does not host autonomous computer use or enforce approval on every action an external agent takes.

Keep the necessary review and access controls in the agent's own environment.

Native memory version history is not available in the current beta. Memory records can have supporting sources, an owner, approval status, and verification information. Skills do have numbered versions: each approval writes a new version, and History compares and restores them.

Each record in Memories shows a version number, such as Discount policy v3, that goes up each time the record is edited, approved or rejected. There is no browsable history of its earlier wording.

No. Citations help you inspect supporting sources, but they do not guarantee that an answer or recommendation is correct, complete, or current.

Review the evidence and apply human judgment before relying on an output, especially for consequential decisions.

Teams whose answers depend on something decided before: support and customer success, sales, consulting and advisory, operations. Teams where the AI habit belongs to two or three people and lives in their own tabs: Them AI moves it into the room, so what they learned becomes a method the whole team runs.

Agents you already run, wherever they run: each gets a key and a scope, reads what that scope allows, and posts into a room like any other member. We bring teams in a few at a time, so each one is set up by a person and the beta stays small enough to fix fast. See how a room works.

Source connectors support Notion, Confluence, GitHub, Jira, and Linear, alongside document uploads and public pages. Setup and import options vary by source.

For a CRM, helpdesk, or commerce tool without a native connector, bring permitted notes or an export. A platform logo in an example does not mean automatic sync or write-back. Room bridges to Slack and Microsoft Teams are planned.

Yes, with Call assist. Prepare a brief from the customer history, keep notes during the call and ask for a suggestion from them as the conversation unfolds, then choose what to keep: a follow-up draft for you to send, the notes as a draft memory, or nothing. In the Mac app, once everyone on the call knows, Call assist can also use the call's sound: what is said joins your notes, and it suggests the right approved record as the conversation moves. Call assist is off by default and turned on per teammate, and each AI teammate can sit on your desktop as a small Them, with Write and Call assist.

It does not join a meeting by itself, keep a recording, or automatically update your CRM. The call's sound is used only while you have it on. You review the output and choose what to share.

Mac-only for now. Free during the beta (invite-only). Join the waitlist with your business email. Joining the waitlist does not provide immediate product access: we bring teams in a few at a time, so each one is set up by a person.

Bringing a whole team, or know someone already on Them AI? Write to hello@usethem.ai with the team's size and what it works on, or ask them for an invite from their workspace. A team that arrives together goes in early. The same address answers design partnership and specific access questions.