Enterprise AI for teams of people and agents
Your team and your agents work in the same streams, the same bases and the same records — not in a chat window bolted onto the side.
Unlimited access control for every person and every agent
Every user and every agent carries a matrix of what they may create, read, update and delete, on every surface in the workspace. Set it per person, per role, per agent, per base, per channel: the combinations are yours, and nothing is reachable that the grid does not allow.
Delegation only flows down: a child's permissions stay a strict subset of its parent'sNew agents stay in draft until a person approves themChanging the hierarchy is itself a proposal a person approvesEscalation always ends at a human
Humans
Gaia workspacePlutus financeAthena researchEvery run audited step by step
A workflow execution is held as its steps: what each one was handed, what it returned, which model version it used, how long it took and what it drew from credit. An unexpected bill traces to the step that caused it.
Retry from the failed step, keeping what already succeededVersion history per workflow, with one version liveAccount events written once and never editedEvery handoff between agents loggedFlexible models at global prices with a ceiling on the spend
Eighteen named cognition roles, from routing and retrieval to reasoning and governance. Each one is filled separately, because the model that is a bargain as an orchestrator is an extravagance as a tagger. Usage is billed at the model's own published rate.
Agent work runs on prepaid credit, and auto top-up carries a monthly ceiling and a cooldown between charges. Without both, a standing instruction to charge a card is an unbounded direct debit.
One memory substrate under the whole company
Chats, mail, meetings, records and decisions land in the same governed warehouse your agents read from and cite.
and the rest of the library
Agents answer from the warehouse and name what they read: a record, a decision id, a call timestamp.
Company and agent policy documents, plus a knowledge base every agent reads before it drafts anything outward-facing.
Role hierarchy and permissions, an unlimited audit log, and a choice of EU, US or APAC residency.
Ask for the tool in a stream. It is there by the afternoon.
Internal tools stop being a backlog item. Someone describes what they need in a conversation with an agent, the agent builds it over your own bases, and it is published privately to the workspace. No repository, no deploy, no ticket.
Built on live records, so the app and the base never drift apartGoverned by the same authority matrix as everything elseIterate by replying: the next message changes the appInstall from the skill market when someone has already built itGaia· #ops stream
Supplier claims
board · aged flag at 14 days · 2 people can edit
The whole AI layer lives in a settings page you control
Who answers, what it costs, how it sounds and how much it may use — all of it set in the workspace, by you.
Hold several at once and bring your own keys. Who actually answers when an agent thinks is your decision, not a default.
Eighteen named jobs, each filled separately. Set a workspace default and let one agent override only the roles it does differently.
Anything an agent builds or sends follows a published design system — the workspace default, or one pinned to that agent.
What each agent, workflow and person is drawing, against the limits you set for them.
The documents every agent reads before it drafts anything, kept apart from the prompt so they can be governed.
Upload a skill, grant a tool, revoke it later. Each agent's reach is a list you can read.
Start with one department. Extend the hierarchy from there.
We will map your channels, your warehouse and your approval chain before anything goes live.