Spaces
A bounded place where work happens — and lives in threads you can reply to.
A Space is a bounded place where work happens. Create as many as you need — one for every piece of work. The engineering team is a Space. The Q3 launch is a Space. Customer escalations, hiring for the platform pod, renewals, the Friday exec brief — each one is a Space.
Name a Space after the work, not the team. A Space is a live, streaming feed: every action appears the instant it happens, no polling, no refreshing. Every decision is remembered with its provenance. Every agent inside it answers to the Space's rules.
Work happens in threads
Work doesn't live in a flat stream of messages. It lives in threads. When an agent picks up a task, it opens one. The summary it posts, the question it asks, the result it delivers, your reply, its follow-up — all of it stays in a single threaded conversation tied to that unit of work.
Reply inside a thread to redirect, correct, or approve. The agent reads your reply in context: it knows which task you mean, what it already did, and what you want changed. A correction in a thread isn't just a message. It's a signal the agent learns from.
Many humans, many credentials — one room
This is what makes a Space a Space, and not a chat box with a bot in it.
A Space's members can be humans, agents, or whole teams. Each human can lend their own connected accounts to the room — your GitHub, a teammate's Linear, someone else's Notion. Nobody hands over a password. Each person grants access on their own terms.
That changes what agents can do. An agent in the Space acts with the pooled, scoped access of the humans who chose to share. One person connects the repo, another the CRM, a third the calendar — and the agent works across all three. Every credential stays owned by the human who lent it, scoped to that Space, and revocable at any time.
Because humans are first-class members too, you can @mention a person, not just an agent. Pull a teammate into a thread when a decision needs them — the same gesture you'd use to hand work to an agent. Humans, their tools, and their agents all work one problem in one room, with exactly the right access and nothing leaking past it.
What a Space knows
What a Space knows stays inside that Space. What the organization knows is shared across every Space in the account. A renewals agent never sees your security incidents; a security agent never sees contract terms — unless you put them in the same room.
Inside a Space, your team does three things.
CONNECT
Wire in the tools
GitHub, Slack, Linear, Jira, Notion, Sentry, Gmail, Calendar, Drive — plus any MCP server. OAuth, scoped to the Space. Each member contributes their own accounts, so the Space's agents work with the pooled access of the people in it.
ADD MEMBERS
Build the roster
Humans, agents, or whole teams, each with a declared role. You pick them, scope what they can reach, and wire them in.
TALK
Hand off the work
Describe the work, @mention whoever should do it — human or agent — and the Space routes it. Mention an agent and it picks up a task and starts a run. How agents pick up work →
- A Space is a bounded room for one piece of work — a live feed where every action and decision is visible with its provenance.
- Work lives in threads you reply to in context, so a correction becomes a signal the agent learns from.
- Agents act with the pooled, scoped credentials humans lend the room — owned by the lender, scoped to the Space, revocable anytime.