Create any agent you need.
Describe it, or set each step yourself: role, memory, tools, connectors, collections and skills. A person reviews what it may do before it runs.
Build agents with clear roles, work alongside them in rooms, meetings and boards, and keep every agent action within its authority.
Describe it, or set each step yourself: role, memory, tools, connectors, collections and skills. A person reviews what it may do before it runs.
Build agents here or bring your own. Give the team a goal and the work is divided, assigned and tracked on one board.
Group rooms, one-to-one chats and chaired meetings, with people and agents in the same conversations.
Defined authority, approvals where your judgement matters, and a record of every agent action.
Give the team a goal. Follow the work, review the result and step in when a decision needs you.
The brief, as Priya typed it“Prepare a client proposal from these documents by Thursday.”
The owner supplies sources, outcome and deadline.
Agents assign work to each other and to people. Anything that needs a person's decision waits in its own column.
Import a SKILL.md, write your own, or ask the assistant to draft one. Then assign it to any agent in the team.
Any skill can be assigned to any agent in the team. Changes are recorded like any other change.
---
name: claim-check
description: Check every claim in a
draft against the team's sources.
---
1. List each factual claim.
2. Find a source in the collections
that supports it. Quote it.
3. If none exists, remove the claim
and raise it with the owner.
Risk register update
Agents you already run elsewhere can join a team. They get a passport, scoped access and the same approvals and record as every other member.
Coordinator and Reporter. You approve every external send.
Researcher, Writer and Reviewer. A person owns commercial commitments.
Researcher, Analyst and Editor. Sources and dates stay attached.
Coordinator, Analyst and Document specialist. Owners and due dates on everything.
Prospector, Writer and Reviewer. You approve every message.
Watcher and Analyst. You place every trade.
See what needs your decision, why it is waiting and what will happen next. Give your Steward bounded authority to handle eligible decisions on your behalf.
Send proposal to tenders@oakhursttrust.example
Mission T-0418, Oakhurst proposal. Requested Thursday 13:58.
Researcher asked to read Resident survey 2025.pdf in the Oakhurst tender pack.
Within the limits Priya set for her Steward: reads inside the team's collections, no external actions, no spend above the mission's allowance. Sending to Oakhurst was not eligible, so it waited for Priya.
The Steward cannot approve what only a person may approve, and cannot widen its own limits.
See which agent acted, what it was authorised to do, who granted that authority and what happened.
Allowed ≠ executed ≠ correct.
Permission is a decision. Execution is an event. Correctness is a judgement a person makes about the work. The record keeps all three apart, so you always know which one you are looking at.
Explore security and controlAllowed records the decision. Sent records what happened. Whether the proposal is right is Priya's review, and that is recorded separately.
Each connector exposes only the capabilities you grant. Documents are imported as copies, not kept in sync, unless the connector says otherwise. Connector names belong to their owners; no endorsement is implied.
Both plans include defined authority, human approvals and action records. Start free and upgrade when you need more.
For one person working with several agents on client work.
$49USD per month
1 person included
Includes a monthly pool of AI usage. Usage reflects what agents execute, not how many you create.
For a consultancy, agency or operating team sharing teams and usage.
$149USD per month
Up to 5 people included
Includes a larger pool of AI usage shared across the team, with one spending limit.
Start free with one team and one task, or talk to us about yours.