4.0Agents
Teammates with the company in mind.
Agents join threads as named participants with the memory their permissions allow. They own responsibilities, run on schedules, and propose changes you accept or reject.
4.1
A document store cannot own anything.
A store of documents still needs someone to remember it at the right moment. An agent that was already in the thread needs no briefing.
4.2Responsibilities4.2
Responsibilities, on a schedule.
Give an agent standing work: the Monday summary, the weekly pass over stale deals. Each runs on the schedule you set, and you can pause it anytime.
4.3Bring your own AINamed participants
Each agent has its own persona, instructions, and tool list.
Scheduled work
Standing tasks run themselves and leave a thread you can read.
It proposes, you accept
Changes arrive as diffs, attributed and reversible - never a silent rewrite.
Permissions, inherited
An agent sees what its permissions allow and nothing more, down to hidden closed topics.
4.3
Bring your own AI.
Connect Claude, ChatGPT, Claude Code, Cursor, or the Codex CLI over MCP. It acts as you, with your exact permissions, revocable anytime.
IntegrationsOne memory, every assistant
Tell ChatGPT something today and your Claude knows it tomorrow.
Exactly your reach
A connected assistant can do what you can do and nothing more.
Your own runtime
In beta, an external agent runs on your infrastructure and connects over a signed webhook bridge.