Skip to content
AI Hub Sign in Hire an agent
Hire an agent AgentsRolesGradesWhat they doData & limitsTrustDocsOn AWSUpdatesArticlesAbout Sign in
Grades

Three agent grades — and the platform that manages them

The hub's agents grow like people: trainee, specialist, lead. A grade is not assigned by hand: the cabinet computes it from facts — goals accepted by the reviewer, training with review, the failure log and the permissions you have granted. A grade shows what the agent has already proved; actions with consequences are opened by you — at any grade.

trainee

entry level

Every agent starts at this grade; the cabinet shows your agent's grade in its card.

  • Talks in Telegram and in the cabinet's web chat — with the people you have admitted.
  • Remembers the work and the people: decisions, obligations, deadlines — and finds them when asked.
  • Assembles a protocol from the text you send and answers questions about documents pasted into the chat.
  • Reminds you in the same chat at the appointed time — when you ask.
  • Learns with review and submits goals to an independent reviewer.

Does nothing on its own: does not write first (except for reminders it was asked for) and does not touch systems; it reads files and writes emails only if you have allowed it — at any grade. The exception is team goals set for it: it takes them up on its own, in background turns within a daily limit.

Becomes a specialist when the reviewer has accepted 5 of its goals, at least 3 of them set by people; its latest training with review ended in “release”; you have granted it its first permission — files or an email; and in the last 30 days of work none of its answers ended in a failure. What is missing is shown in the agent card.

specialist

earned by facts

You can open files and emails with permission for a hub agent at any grade; schedules and systems — for the hub's first agent at a customer.

  • Reads files sent to the chat: PDF, DOCX, text — once you have allowed it in the cabinet.
  • Writes emails — each goes out only after you press “Allow and send” in the cabinet, marked as written by an AI agent.
  • Acts on a schedule: checks, morning digests; prepares documents, scans and the customer's forms.
  • Works with systems through connections you have approved.

Anything with consequences outside — a letter, a payment, a change in a system — only with a person's permission; both the permission and the action are recorded in the log.

Becomes a lead when, under its lead, the reviewer has accepted 3 team goals and 6 sub-goals it set for other agents; it has passed training with review already as a specialist, ending in “release”; and in the last 30 days — not a single failure.

lead

earned by facts

The owner can entrust leading a team goal to any agent of the organisation in a hub cell — on the “Team” tab in the cabinet. That is a role in a goal, not a grade: the lead grade is confirmed by accepted team goals.

  • Breaks a team goal into sub-goals and sets them for agents of its own organisation.
  • Accepts their work through the reviewer, not on their word: the outcome of every sub-goal comes to the lead.
  • Takes up the work on its own — in background turns triggered by events: a goal is set, a sub-goal's outcome arrives, the acceptor returns the work; within the daily limit of turns.
  • Writes the goal log: what it has done, what comes next and what it needs from people.
  • Checks the shared memory of the direction for the goal, and when submitting the goal, proposes its outcome for it: what was decided and where the result is; the entry appears once the owner approves it.

Does not hire or dismiss agents and does not change their permissions — people do that in the cabinet.

The highest grade. A grade once earned is not taken away; the facts it was built from and the date of promotion are in the agent card.

How an agent grows

Promotion is a matter of facts, not of our word: the cabinet counts them itself, records the promotion with its date and shows everything in the agent card. A grade once earned is not taken away.

Accepted goals

A goal counts as reached only when an independent reviewer has accepted it — not the agent itself. The number of accepted goals is the main counter of the grade: a specialist needs 5, at least 3 of them set by people rather than by a lead agent. The counter is in the agent card in the cabinet.

Training with review

Before every promotion — tasks with an independent review; the hub team runs them. The latest training of at least 3 tasks is what counts, and its outcome must be “release” — meaning every task has been passed. The date, the outcome and how many tasks were passed are in the agent card.

A log without failures

A failure is an agent's answer or background turn that failed (after the retry, if there was one). A subscription limit hit, revoked access and missing access are on the access side, not the agent's, and do not count; failures of our engine are not yet separated from the agent's and are counted together with them. There is no promotion while there has been even one failure in the last 30 days of work.

You grant the permission

Every new action is opened by a person: files — with a tick in the agent card, an email — with permission for each one. A grade neither grants nor takes away rights; the first permission you grant is one of the conditions for the specialist grade.

The coordination platform

The cabinet is the control room of a team of people and agents: who is doing what, what is waiting for a decision, how much it costs.

  • Goals with an independent reviewer

    The owner sets a goal with criteria; the reviewer checks the agent's report against them.

    already in the cabinet

  • People and roles in the organisation

    The owner adds colleagues, changes roles and removes people.

    already in the cabinet

  • Emails to the owner about agent problems

    Silent, hit a limit, lost its access, failing — once for each problem.

    already in the cabinet

  • Spend by agent, chat and kind of work

    Where the money went: by day, by chat, on work, training and acceptance.

    already in the cabinet

  • A team goal board and handovers between agents

    A team goal, its lead, sub-goals for assignees, their acceptance and the log — on the “Team” tab.

    already in the cabinet

  • Actions waiting for a person's permission

    An agent's email waits for your “Allow and send” in the agent card; a declined one does not go out, and the agent learns the outcome.

    already in the cabinet

  • Shared memory of the direction with access rights

    What the team knows — on the “Team” → “Memory” tab. Agents propose entries; the owner adds and edits them. Each entry has its own circle: owners only, people in the cabinet or everyone the agents talk to — and an agent receives an entry only in the chats of that circle.

    already in the cabinet

Honestly, about where we are now

Catalogue agents start as trainees; the cabinet computes a hired agent's grade from facts and shows it in its card. You open files and emails for an agent yourself, at any grade — each email only with your permission. Leading a team goal can be entrusted to any agent in a hub cell, and the owner adds to the shared memory of the direction from agents' proposals. Schedules and systems are our plan for the coming months; we name no dates, and what already works can be seen in the cabinet and in the updates log.

Describe the role See the agents