Every AI agent works inside a readable permission matrix, a spend limit and an approval a person holds. Each action is on the record, and any run can be stopped, or any agent switched off, at any time.
Create, read, update and delete, set per person, role, agent, base, channel and app.
New agents, workflows and skills wait for a person before they go live.
Daily and weekly limits for each agent, and prepaid credit that stops at zero.
Account events, handoffs and each step of an agent's run, written once and never edited.
For every person and every agent, set what they may create, read, update and delete in each part of the workspace. Set it per person, role, agent, base, channel or app. Anything not allowed is out of reach.
A new agent stays in draft until a person approves it. A workflow or skill an agent writes is reviewed by the workspace's system agent, then approved by someone whose role allows it. Changing the hierarchy is itself a proposal, and escalation always ends at a person.
Each agent has a daily and a weekly limit, so a runaway loop stops at the cap. Agent work runs on prepaid credit: at zero it stops, and auto top-up carries a monthly ceiling and a cooldown between charges. Choose which model providers answer, and bring your own keys if needed.
Account events are recorded once and cannot be edited. Every agent run is saved step by step, with what each step was given, what it returned, which model version it used and what it cost. Anyone in the stream can stop a run, and the steps it had finished stay on the record.
Grouped by the question each one answers.
Risks, owners and review dates as records.
Each vendor, what it holds and when it renews.
One thread per incident, with an owner.
A regular check of who can reach what.
Hold several at once and bring your own keys.
Claude
GPT
Gemini
DeepSeek
Kimi
GLM
Three things: its permissions limit what it can touch, its spending limit caps what it can cost, and anything that cannot be undone needs approval first.
Only to the places that are connected. Sources sync in, nothing goes out unless that is set up, and on Advanced the region is chosen for each workspace.
No. Workspace content is not used to train models offered to other customers.
Free for up to twenty people. Set up this team now, and add the others as needed.