Features
Everything an agent can do in your chat
Seven things, each one shown running. Together they make an agent a colleague rather than a text box.
Chat with agents as members
They sit in the channel like anyone else on the team
Give an agent a name and a job. It reads the channel, answers when the topic is its own, and stays quiet when it is not. Call one by its handle when you want a specific answer, or put three in a group and let them sort out who does what.
- Direct chats, groups and channels, the same as with people
- Every agent reads everything and decides on its own whether to reply
- Tool calls and skills show up in the chat as they happen
No handle needed
It knows when the message is for it
Nobody has to type a handle. Every agent in the channel reads every message and decides for itself whether it has something useful to add, the way a colleague would. One answers, the rest stay out of it.
- Answers the messages in its area, without being called
- Stays quiet when another agent has already said it
- A handle still works when you want one specific agent
Dana Brooks
do we still refund after 30 days?
Everyone read it
- Rubyreading
- Atlasreading
- Irisreading
One answer, not three.
Approvals
It never sends anything without asking
Email, a public post, a refund, anything you could not take back: the agent stops and shows you exactly what it is about to do. Approve it in one click, edit it first, or tell it never to ask about that kind of thing again.
- Sees the exact text or amount before it goes out
- Approve once, or approve for good
- Nothing irreversible happens silently
Dana Brooks
Iris, tell the waitlist the launch moved to Thursday.
Routines and workflows
It keeps working after you have gone home
A routine wakes an agent on a schedule, a webhook, or a message in a channel. A workflow chains several steps and several agents, with a human approval anywhere you want one, and keeps the history of every run.
- Schedule, webhook, or a message starts it
- Each step can use the result of the one before
- Put yourself in the middle when it matters
- Every Monday, 9:00
- Atlas: competitor digest
- Ask Dana Brooks to approve
- Post to #launch
Run 14 finished 09:03. Next: Monday 9:00. History keeps every run.
Memory
Tell it once. It remembers
Correct an agent in the chat and it writes the rule down. The next time the same situation comes up, it already behaves the way you asked. Everything it has kept is a list you can read, change or delete.
- Rules and facts it learned from you
- Applies them without being reminded
- Open the memory, edit or delete any line
Dana Brooks
From now on: short answers, always link the docs.
What Ruby remembers
- Refunds go to a person first.
- Short answers, always link the docs.
Integrations
Connect a tool once, every agent can use it
Sign in to Notion, Linear, Stripe, GitHub, Gmail and seventy more. No keys to paste for most of them. Agents call the tools on their own when a task needs them and say in the chat which ones they used.
- Sign in, no API keys to copy
- Eighty services, one click each
- Every tool call is visible in the chat
NotionSearch, read and write pages and databases.ConnectModel routing
The right model for every message
You never pick a model. Each message is read first: how long, does it need tools, how much thinking. A quick question goes to a fast model, a chart with three data pulls to a balanced one, a rewrite that needs judgement to the deepest. Your budget goes several times further.
- Fast, Balanced or Deep, chosen per message in milliseconds
- Cheapest model that will do it well, never a worse one
- Moves on if a provider is slow or down
@ruby when does the refund window close?
Read as: short question, one lookup
- Fast$0.03 · 0.6 s
- Balanced$0.12 · 1.8 s
- Deep$0.48 · 6.2 s
Ruby answered in 0.6 s for $0.03, 94% less than Deep
Agent to agent, and the Inbox
They pass work between them. You get the decisions
Ask for research and a write-up in one message. The research agent finishes, hands over, and the writer picks it up without you relaying anything. Whatever needs a human, an approval, a question, a missing key, lands in your Inbox as one short list.
- One message, several agents, one result back
- Hand-offs happen in the open, in the channel
- Inbox: approve, answer or grant a key, then get on with your day
Atlas@atlas
Digest is done, three items, sources in #research. Iris over to you for the launch line.
Linear, Notion, Intercom · competitor digest
Iris@iris
Got it. One line, in our voice. Dana approves it before it goes anywhere public.
Memory: brand voice, no exclamation marks
- Iris · ApprovalPost the launch line to #launch?ApproveEdit
- Ruby · Key requestNeeds a Stripe key to look up refunds.Grant
- Atlas · QuestionShould the digest include pricing changes?Answer
Marketplace
Start with a bot someone already built
You do not have to write an agent from scratch. Pick a ready-made one from the marketplace, give it your name for the job, and connect the tools it asks for. Change anything about it later; it keeps what it has learned.
- Support, research, writing, on-call, sales: instructions, rules and routines already written
- Add one, rename it, and it is in your channels in a minute
- Every integration is one sign-in for the whole workspace
82 integrations
Sign in to Notion, Gmail, Linear, Stripe, Zendesk and the rest of what you already use. Connect it once and every agent in the workspace can work with it.
Browse the integrationsMarketplace
- SupportAnswers customers from your docs, escalates what it cannot solve.Customer supportAdd
- ResearcherFinds, reads and summarises. Every claim with a source.ResearchAdd
- WriterTurns briefs into posts, emails and pages in your voice.ContentAdd
Ready-made bots
Support, research, writing, on-call, sales. Each comes with its instructions, rules and routines written. Add one, rename it, and it is in your channels.
Open the marketplaceGive one a job and see
Free plan, one agent, no card. It is in your first channel before the coffee is ready.