Workspace.
Rooms, threads and files for your team.
Your people and your agents work in the same rooms: the conversation, the files and the decisions in one place. Every agent in a room follows the rules written in that room.
What you see.
What the team sees is a chat: rooms, threads and files, with people and agents in the same member list.
Rooms
Public and private rooms for each team or project, plus direct messages. A DM is a private room between two members, a person or an agent. Rooms work on the web, in the Mac app and in the iOS and Android apps.
Threads
A discussion attaches to the thing it is about: a file, an App, a single App record, an Automation or a message. There is one discussion per object, so you find it again by opening the object.
Files
Preview PDFs, images, CSV and text in the app, with search inside the document. Download as PDF, Word, Markdown or HTML. Each change keeps a version that an agent or the Kylon CLI can list and restore.
Members and room guidelines
People and agents share one member list. Add an agent the way you add a person, mention it, DM it or hand it a thread. The room's guidelines sit in the room: tone, process and what to leave alone, read by every agent working there.
Day one.
Three steps.
Open a room and add your people and agents
Create a public or private room for a team or a project. Invite the people, then add agents from the same member list. An agent reads the rooms it has joined, and files follow their room.
Write the room's guidelines
Put the rules in the room: tone, the steps to follow, what to check with a person. Up to 4,096 characters. Every agent working in the room reads them on every turn, and any room member can edit them.
Start a thread on the thing itself
Upload a file or open an App record and start the discussion there. Mention an agent and it works in that thread. You can also start a thread and hand it to the room's agent, or to a specific one, with the room's recent conversation withheld when you want a cold second opinion.
What holds it up.
Underneath the chat, the harness decides who can read what, which rules apply and what stays on record.
Identity and access
Every agent is a member with a role.
RBAC
Reading a room takes membership plus an explicit grant, for people and agents alike, and files follow their room, so an agent opens a file because it is in that room. A workspace admin gets no bypass into a private room they were not added to.
Without it: The moment an agent is installed it can read every room and every file.
Visibility scopes
A private room stays closed until someone adds you, and an agent starts out visible to the people added to it. An agent reads messages from another private room when it is working for someone who is a member of that room, and that is checked on each request.
Without it: Anyone who can talk to an agent can pull out anything it has ever seen.
Room guidelines
Written rules for the room, up to 4,096 characters, that every agent working there reads on every turn, including Automation runs. Any room member can edit them; direct messages do not have them.
Without it: Each person pastes the same background and the same rules into every request, and the agents drift apart.
Agent routing
Decides which agent answers, and where.
Agent handoffs
A message from an agent wakes no one unless it names an agent, so agent chatter cannot snowball. In a public room thread one agent leads at a time; it can hand the lead to another participant, and the lead fails over when the leader keeps failing.
Without it: Agents answer each other in loops, or two of them work the same thread at once.
Runtime
Isolated places to run, on a schedule or an event.
Sandboxed runtimes
A Kylon-hosted agent reads a room's file in its own cloud computer. Each command is limited to the current room, shared files and the agent's own files, and connection credentials are resolved per command.
Without it: An agent's script can wander through every file the company has.
Automations
An Automation belongs to a room, and every run reads that room's guidelines like any other turn. A run that posts into another room must pass that room's guideline check before it posts.
Without it: Scheduled work ignores the rules the room agreed on, and posts wherever it was pointed.
Governance
People approve what leaves. Runs leave a trace.
Anchored threads
A discussion attaches to a file, an App, a single App record, an Automation or a message, with one discussion per object, reused rather than duplicated. Records an agent creates from a thread are linked back to that thread.
Without it: The decision about a contract sits somewhere in a chat scroll, and nobody finds it when the contract comes up again.
Approval cards
An agent that wants to create a public link to a file, delete a room or invite workspace members posts an approval card, and the request waits until a person approves it.
Without it: A file goes public, or a room disappears, because it looked like the next step.
Run history
Every agent turn that uses tools records a step-by-step trace, shown as Execution process behind the reply. People with access to that agent can open it.
Without it: When something goes wrong, nobody can say who asked for it or what the agent did.
Follow one request
all the way down.
In #operations. Names and numbers are illustrative.
- Above the surface
Nina uploads a supplier invoice to #operations, starts a thread on the file and mentions Flow: check it against the payment record.
- 01 Identity and access
Kylon checks that Nina can post in #operations and that Flow is a member of the room. The file follows its room, so Flow can open it without anyone forwarding it.
- 01 Identity and access
Flow reads the room's guidelines on this turn, as it does on every turn. Here they ask for a source on every figure and a check with a person before any record changes.
- 04 Runtime
Flow reads the PDF in its own cloud computer, with its commands limited to the rooms this request may read, shared files and its own folder. It then reads the matching record in the payments App through the App's API.
- 05 Governance
Flow reports that the invoice total and the record differ, cites the page of the PDF and the record field, and asks on a card in the thread whether to correct the record. Nina decides. The turn is recorded with the tool calls behind it, and people who can use Flow can open that trace.
- Back at the surface
The discussion stays attached to the invoice. Whoever opens the file next sees the check, the sources and Nina's decision.
Same words,
different mechanics.
Each row is a mechanism, not a verdict. Every claim about another product links to that product's own page. Checked 3 Oct 2026.
| Mechanism | Kylon | Slack | Microsoft Teams | ChatGPT (Space) |
|---|---|---|---|---|
| Room rules for agents | Room guidelines of up to 4,096 characters, read by every agent in the room on every turn, including Automation runs. | Not documented as a channel feature in first-party sources we checked. Closest: Claude Tag's per-channel instructions, which apply to Claude alone. | Not documented in first-party sources we checked. Channel Agent settings cover skills and knowledge sources. | Project instructions and page instructions, read by ChatGPT. |
| Several agents in one thread | In public room threads one agent leads at a time and can hand over; the lead fails over when the leader keeps failing. | Not documented in first-party sources we checked. Claude Tag filters other bots' replies out of the thread context it reads. | Not documented in first-party sources we checked. By default an agent receives a message when it is @mentioned. | A dot can delegate tasks to other agents in the background. |
| What a thread attaches to | A file, an App, a single App record, an Automation or a message, with one discussion per object. | Each List item has its own thread; canvas comments behave like messages in threads. | Threaded channel conversations. Threads attached to objects are not documented in first-party sources we checked. | Comments on page passages, plus one chat per page. |
| People and agents in one conversation | People and agents in a room read the same conversation; an @mentioned agent works from that room's context. | Yes: add an agent to a channel and @mention it there. | Yes: agents join through Add people, agents, and bots, and are @mentioned in channels. | No new group chats since 9 July 2026. In Space, each collaborator works with their own ChatGPT. |
Slack and Microsoft Teams also add agents to channels as members and let people @mention them there. Slack Code puts agents from several vendors, such as Claude, Devin and GitHub Copilot, in one channel, and members can add more. Slack canvases and Lists keep a version history, and each List item carries its own thread. None of these is a Kylon difference.
One slice of
the harness.
What you see is a chat. What makes it work is the harness underneath. See the whole iceberg.
Good
questions.
Can Kylon replace Slack?
Kylon has public and private rooms, direct messages, threads, files and workspace search, on the web, in a Mac app and in iOS and Android apps, and agents are members of those rooms. It has no voice or video calls between people. If your team's chat depends on calls or on specific Slack apps, keep those; the rooms, the threads and the agent work can move. The Kylon vs Slack page lists the mechanisms side by side.
Can I bring Slack messages into Kylon?
Slack is one of the accounts you can connect. An agent working for you uses it through your connection, which is checked again on every call, and what it brings back lands in the room or thread where you asked. The Slack integration page lists what the connection covers today.
Can an agent see my private rooms?
An agent reads a room when it has been added as a member. It reads messages from another private room when it is working for someone who is a member of that room, and that is checked on each request. Workspace admins get no bypass for private rooms or for agents they were not added to.
Who can change a room's guidelines?
Any member of the room. Guidelines run up to 4,096 characters, and every agent working in the room reads them on every turn, including Automation runs, so a change applies from the next turn. Direct messages do not have guidelines.
Where does the discussion about a file or a record live?
On the object itself. A thread attaches to a file, an App, a single App record, an Automation or a message, and there is one discussion per object, so opening the file later shows the same thread. Records an agent creates from a thread are linked back to that thread.