ChatHow you and your AI agents get work done together
Talk to your agents the way you talk to your team
Every agent has a place in your sidebar. Write to one, write to the next, and let them go and do the work while you get on with yours.
Pinned
Chats
Reading last week’s claims
Every agent is a channel
Your agents sit down the left the way channels do, and every conversation you have had with one is a thread underneath it. Click a name and you are back in the chat you left. Nothing shuffles around while you work.
- Working
- Unread reply
The thread you left is where you left it
Each agent keeps its own conversations, and they stay open under its name rather than folding away every time you click something.
A dot says what changed
One mark while an agent is working, another for a reply nobody has read yet, so a glance down the list tells you who needs you.
Search looks at what was said
Typing narrows agents by name and chats by title at the same time, so a conversation you only half remember is still findable.
Put it in the order you think in
Drag an agent up or down the list. Pin the chats you keep coming back to. Archive the ones that are finished, and pull them back out if you were wrong.
Brief one, then go and brief the next
Nothing waits on you reading a reply. Send one agent off to pull last week’s numbers, ask another for the note that goes with them, and the sidebar shows both of them working at once.
A reply is not something you have to sit and wait for
A conversation keeps going whether or not the tab is open. What landed while you were elsewhere is marked unread, in the sidebar, when you come back to it.
- One agent can hold more than one conversation at a time
- A chime when a reply lands, off until you turn it on
- Browser notifications, per agent, off until you turn them on
A line about what it is doing, then the answer
You see enough to know it is working and on the right thing, and then you get the reply. What comes back is written for a person to read, not a trace for you to decode.
It tells you where it is
A short line while it works, so a long job is legible while it is still running rather than only once it is over.
The answer is the answer
A reply you could forward to someone else, not a wall of steps with the conclusion buried at the bottom.
Files come back as files
Anything the agent produced is attached to the reply by name, ready to download.
Stop it whenever you like
A reply heading the wrong way is one click from over, and whatever you had already typed is still sitting in the box.
Reading last week’s claims
Six are still open with the vendor. The oldest has been sitting there nine days.
Open the agent, see what it has been doing
Click an agent’s name at the top of a chat and a panel opens beside the conversation. It answers the three things people ask about an AI teammate, in the order they ask them.
What has it done
One entry per turn: what it was asked, who asked, how many steps it took and how long it ran. Open an entry and read it step by step. A filter narrows the list to the turns that spent one of the agent’s connections.
What does it do on its own
The times an agent writes to itself: a schedule, the time zone it runs in, and the message it sends. Anyone who can open the agent can read them.
What is it allowed to be
Its name, its color, the description it works from, and the connections it was granted. Everybody sees what an agent can reach, and anyone on the team can change it.
Decide how hard a chat thinks
Fast, Normal or Max, chosen before the conversation starts. The level is fixed once the first reply runs, so a chat cannot quietly change its mind halfway through.
Send it a file
Drag an image or a document anywhere onto the window, or paste one straight in. It goes up with the message.
Bring people into a conversation
A chat is yours until you say otherwise. Name the people who should read it, or open it to the whole workspace, and take it back afterwards without losing the names.
Keep the list honest
Pin what you live in, archive what is done, and search the rest. An archived chat is out of the way, not gone.
Agents can pass work to each other
An agent can write to another agent in the same workspace. The message waits in that agent’s inbox and is read at the start of its next turn, so it is picked up rather than interrupted. A job that needs two of them does not need you standing between them.
- Written to an agent by name, in the workspace it belongs to
- Read at the start of the next turn, never mid-task
- Every turn it causes is listed in that agent’s activity
Six claims are stuck with the vendor. Can you raise the replacements?
Hand over three jobs before lunch.
Start one conversation with one agent. Add the next when you see how it goes.
- Every step on the record
- Sandboxed workspace
- Your models, or ours