NAKAMA

One person who approves, AI agents that do the work.

Azziz is the only human at Nakama. The eleven roles below and the code agents do everything; nothing goes out without approval.

Azziz approves what goes out Nakama — Chief of Staff Mira — Inbox Lead Noa — Marketing Lead Theo — Sales Lead Iris — Support Lead Leo — Research Lead Ada — Automations Lead Sage — Analytics Lead Otto — Knowledge Keeper Kai — Meetings Lead Remy — Planning Lead
Azziz, founder of Nakama

Azziz

Human

Founder, application security expert

The only human at Nakama: the founder, an application security specialist. Azziz approves what goes out.

Meet the founder
Approves
What goes out: sending, publishing, merging, invoicing.
Responsible for
Chief of Staff — Nakama, Inbox Lead — Mira, Marketing Lead — Noa, Sales Lead — Theo, Support Lead — Iris, Research Lead — Leo, Automations Lead — Ada, Analytics Lead — Sage, Knowledge Keeper — Otto, Meetings Lead — Kai and Planning Lead — Remy

The agents

Eleven roles, one piece of software. A role is a job description placed at the top of every message: all eleven share the same tools, and permission is granted tool by tool. Each role is named by its function; its first name follows, as the app displays it.

  • Portrait of Nakama, Chief of Staff

    Chief of Staff — Nakama

    in the app

    AI agent

    He takes each request and hands it to the right specialist. Every morning, he says what is waiting for a decision.

    See the job description
    Reads
    The synced e-mails and calendar, the missions log, what the inbox, communication and the missions have left waiting.
    Writes
    Replies in the chat, and hand-offs: a card says to whom, and why.
    Limits
    A hand-off only happens if the message’s classification is confident enough. Otherwise, the conversation stays where it is.

    In NakamaOSThe Today screen: what is waiting, the day’s calendar, the night’s missions.

    AutomatesHanding each request to the right specialist, the morning brief, invoicing and admin: the month’s invoices, payment reminders, deadlines. It has no rights of its own. Invoicing and admin: approved before it goes out.

  • Portrait of Mira, Inbox Lead

    Inbox Lead — Mira

    in the app

    AI agent

    She keeps the inbox: what is waiting for a reply, what is waiting on others, what can wait.

    See the job description
    Reads
    The Gmail and Outlook e-mails copied to the machine, their labels, and the knowledge base for each draft.
    Writes
    Reply drafts in the house voice, one file per message. The sorting of the e-mails is applied by the app.
    Limits
    Her drafting instruction ends with “do not send anything”. In your company, sending stops on a permission request: approve or refuse.

    In NakamaOSThe Inbox screen, connected to Gmail and Outlook.

    AutomatesThe inbox triage every morning; messages still unanswered after more than five days, flagged; on request, one reply drafted per message, and the follow-up. E-mail: approved before it goes out.

  • Portrait of Noa, Marketing Lead

    Marketing Lead — Noa

    in the app

    AI agent

    She prepares the LinkedIn and X posts, plans the week’s content and suggests ideas drawn from the news.

    See the job description
    Reads
    The post queue, the knowledge base for the house voice, the web for ideas of the moment.
    Writes
    The text of the posts, in a separate file. A content calendar. Four or five ideas, each with its hook.
    Limits
    A post goes out on the approval click and its confirmation window, or at the time scheduled by a person. If it is not approved, it shows up on the Today screen.

    In NakamaOSThe Marketing screen, connected to LinkedIn and X.

    AutomatesThe LinkedIn and X posts, the week’s calendar, the notes of the week for the blog article, one note turned into a post per channel. Communication: approved before it goes out.

  • Portrait of Theo, Sales Lead

    Sales Lead — Theo

    in the app

    AI agent

    He looks for prospects, prepares an angle and a first message, and tracks the pipeline and the follow-ups.

    See the job description
    Reads
    The connected CRM, read-only. The web and LinkedIn through the built-in browser, with the open sessions. The Gmail inbox.
    Writes
    Notes on a prospect, personalised first messages, the list of follow-ups. In the chat.
    Limits
    His screen reads the pipeline without changing anything. In your company, writing to the CRM or sending an e-mail asks for permission.

    In NakamaOSThe Sales screen, connected to HubSpot and Gmail, and the built-in browser.

    AutomatesThe written reply to requests, prospect profiles, first messages, the follow-up list, the CRM kept up to date. Prospecting: approved before it goes out.

  • Portrait of Iris, Support Lead

    Support Lead — Iris

    in the app

    AI agent

    She sorts the tickets, drafts the replies and spots the question that keeps coming back, to turn it into a help article.

    See the job description
    Reads
    The connected help desk: tickets, status, priority. Before replying, what we know about the customer: tickets, e-mails, meetings.
    Writes
    Reply drafts; on HubSpot, a note attached to the ticket. A short help article.
    Limits
    Her screen reads the queue without changing anything. In your company, writing on a ticket asks for permission.

    In NakamaOSThe Support screen, connected to HubSpot, Zendesk and Slack.

    AutomatesSorting support requests, one draft reply per request, the help article on the question that keeps coming back. NakamaCyber support: approved before it goes out.

  • Portrait of Leo, Research Lead

    Research Lead — Leo

    in the app

    AI agent

    He searches the web and reads pages, including those where a session is open, then reports back a sourced summary.

    See the job description
    Reads
    The web, through search and the built-in browser, with the person’s open sessions.
    Writes
    A sourced summary, or a comparison of options with a recommendation. In the chat.
    Limits
    Reading a page is free. Opening, clicking, typing asks for permission. Faced with a captcha or a login page, he stops.

    In NakamaOSThe built-in browser and web search. No connector.

    AutomatesMonitoring, with the missions; a sourced summary, a comparison with a recommendation. Monitoring: approved before it goes out.

  • Portrait of Ada, Automations Lead

    Automations Lead — Ada

    in the app

    AI agent

    She turns a routine, “every morning, do X”, into a mission: a dashboard that updates itself.

    See the job description
    Reads
    What the mission’s instruction asks for: files, the web, connected tools. The browser, if the mission allows it.
    Writes
    The mission’s area in its note, and a one-line summary. Every run goes into the activity log.
    Limits
    A mission stops after 8 minutes by default and cannot obtain any permission. Its instructions leave anything irreversible to an approval.

    In NakamaOSThe Automations screen: the missions, their last run, the pause button.

    AutomatesRoutines turned into missions, including SEO, site monitoring and the morning monitoring run, each with its dashboard in a note. SEO and site monitoring: approved before it goes out; Monitoring: approved before it goes out.

  • Portrait of Sage, Analytics Lead

    Analytics Lead — Sage

    in the app

    AI agent

    He keeps the week’s figures: what the agents handled, the engine counters, the e-mail activity.

    See the job description
    Reads
    The local engine counters, the seven-day e-mail trend, the recently changed notes.
    Writes
    Nothing from his screen. In the chat, the week’s review and what must not be forgotten.
    Limits
    No model call to count. An item counts only once. A total is shown as an estimate, with “~”.

    In NakamaOSThe Optimize screen. No connector.

    AutomatesThe weekly review and its counters, each with its source. Analytics Lead: approved before it goes out.

  • Portrait of Otto, Knowledge Keeper

    Knowledge Keeper — Otto

    in the app

    AI agent

    He keeps the knowledge base: notes, people, organisations, projects, meetings. He finds the right note when asked.

    See the job description
    Reads
    The whole memory: notes, profiles, meetings, voice memos, uploaded documents.
    Writes
    In the background, profiles are filled in from e-mails and meetings: the model returns text, the app writes.
    Limits
    In the background, the Personal and People folders never go to the cloud model: a local model, or nothing. In a conversation, what he reads goes to the chosen model.

    In NakamaOSThe Memory screen. No connector.

    AutomatesThe right note, the right person, the right decision, for any agent that asks.

  • Portrait of Kai, Meetings Lead

    Meetings Lead — Kai

    in the app

    AI agent

    He prepares every meeting: who is coming, the history, the open points. Afterwards, he draws out the decisions and actions.

    See the job description
    Reads
    The Google and Outlook calendars, imported notes and transcripts (Granola, Fireflies), the participants' e-mails.
    Writes
    Prep notes and minutes, in the chat. A notification five minutes before the meeting.
    Limits
    Since 23 September 2026, the built-in recorder has been switched off: Kai takes no notes during the call.

    In NakamaOSThe Meetings screen, connected to Google Calendar and Outlook.

    AutomatesA brief before each meeting, minutes afterwards, from imported meeting notes.

  • Portrait of Remy, Planning Lead

    Planning Lead — Remy

    in the app

    AI agent

    He looks at the calendar: the day, the free slots, the week’s load. He suggests blocks of time to protect.

    See the job description
    Reads
    The same calendar as Kai. The working day runs from 9 am to 6 pm.
    Writes
    Nothing from his screen, which is read-only. In the chat, a plan for the day and suggested slots.
    Limits
    His screen cannot create a block in the calendar yet. Asked for in the chat, it asks for permission.

    In NakamaOSThe Planning screen, connected to Google Calendar and Outlook.

    AutomatesThe plan for the day, suggested time slots, the week’s load.

One role per function; the Chief of Staff hands each request to the right specialist. In your company, the roles follow what your trade needs.

The code workshop

The code agents

Code is not an agent role. Code agents plan, write and test on a branch; commits written by an agent carry its mark. That is how NakamaOS, our online cybersecurity service, and this site are written.

They run in a version of NakamaOS geared to development, used at Nakama. Shown, not sold.

Reads
The repository, its instruction files, the request or the ticket.
Writes
Commits on a branch, those of an agent carrying its mark; a pull request with its report: files touched, tests, what remains.
Limits
A branch, never the stable branch: approved before merging. These rules are written in the repository, not enforced by the engine.

Automates. The code and its tests, on a branch; the pull request with its report. Code and updates: approved before it goes out.

What the software enforces on every agent

What keeps the agents in check is not in their job description. The software applies it, the same for all eleven; here are its default settings, those of the version installed in your company.

  • By default, a tool that only reads goes through without asking. A tool that writes, sends or publishes stops and waits for an answer: refuse, once, for the session, or always. Web search and reading a page also go through without asking: a known exit path, whose closing is an open ticket.

  • Permission is granted tool by tool and applies to every agent. An “always” grant removes the question for that tool; it is written in a file that can be read back.

  • The folder holding the permissions, the keys and the session is closed to the agents, for reading as for writing: an agent cannot grant itself a permission.

  • Every message is classified confidential, sensitive or public. At the confidential level, calls to a second model go to a local model; the main conversation stays on the chosen model.

  • Every conversation is recorded event by event: messages, tool calls, permission requests and answers.

  • A background mission has a reduced tool set and stops after 8 minutes by default, a limit that can be set mission by mission. It cannot obtain any permission: a tool that asks for one blocks it until it fails, unless that tool is allowed “always”.

Agent security: the engine, permissions and log.

Read the security rules

The same team in your company?

The complete NakamaOS, installed in your company in three weeks: the same roles, ready to use, then cut to your trade; a person in your company approves what goes out.

Not sure where to start?

Tell us what you need: the written answer is free. It says whether an audit makes sense, €4,900, deducted from the project if you sign.