AI Chief of Staff
AI Chief of Staff
Everything your company knows. For everyone who works there.
Picture one memory for your whole company: it knows every document, every meeting and every project – and answers your questions in seconds.
One place for everything
Your company knowledge flows in
Documents, emails, tickets, presentations, customer data, meeting notes: everything that is scattered across many tools today ends up in a single Brain.
Always up to date
New knowledge arrives on its own
Nothing to maintain by hand. As soon as something new appears in your tools, it flows into the Brain automatically. At the Enterprise AI Summit, a talk could be searched 30 to 45 minutes after it ended.
Inside the Brain
Everything is connected
The Brain sees which people, projects and topics belong together and links them – just as you know who works on which project and which decision goes with it.
Ask
Ask a question, get an answer with its source
Ask in your own words. The Brain finds the answer and its context and shows you where it comes from: the document, the meeting, the exact moment in the conversation.
Security
Everyone sees only what they are allowed to see
The Brain is divided into areas such as sales, HR or finance. If someone has no access to an area, they get no answers from it – just like the folder permissions you already know.
Use
Your AI assistants work from the same knowledge
Chatbots, AI assistants and automations all draw on the same Brain, whichever AI runs them. Everyone in the company simply asks in Slack, Teams or by email.
The whole picture
A memory for your company – safe with you
The Brain runs in your own cloud or on your own servers, and your knowledge stays yours. In its first large-scale field test at IT Revolution’s Enterprise AI Summit, it made 1,617 talks and more than 21,000 pages searchable for 300 attendees. That was the first showcase—the approach is designed for use across the whole company.
AI Chief of Staff
A personal Chief of Staff for everyone in your company
Your Chief of Staff understands your goals, finds relevant knowledge and supports your day-to-day work—with a dedicated agent for each person instead of one chatbot for everyone.
The principle
Not one chatbot for everyone, but one agent per person
A single central assistant knows everyone and no one. An AI Chief of Staff inverts that: one container per person, with its own memory. It does not start as an empty prompt box but with a profile the person reads, corrects or deletes beforehand. Everything else follows from that person's goal: which sessions count, which contacts pay off, what belongs in the summary.
What the agent takes over
Six jobs nobody else gets around to
Every one of these is manual work today—which is why it is the first thing to go when two days are full.
Match goals against the programme
The agent scores every talk against its owner's role, interests and goals. You get a short list instead of a programme booklet.
Find the right conversations
It looks for people who have already solved your problem and negotiates a time with their agents. You accept, defer or decline.
Make the knowledge base queryable
Talks, speakers, books and papers sit in one shared knowledge base. The agent queries it on your behalf instead of leaving you to search.
Trace a slide back to its source
A photo of the slide is enough. The agent finds the original talk, the speaker and the chapter the idea came from.
Take notes after the conversation
You record a voice note; the agent files it against the person and the topic. Nothing depends on your memory or a stack of business cards any more.
Write the story for the manager
The agent maintains a running thread from company goal to your own challenge to what each talk contributed, exportable as a report.
First field test: event
The first field test, from onboarding to outcome
The event was the first large-scale showcase, with 300 attendees. Its workflow shows how the approach can be tailored to a specific use case; in a company, the Chief of Staff can support people continuously in their day-to-day work.
Claim it and review the profile
One link or QR code, work email, a password—done. No vendor account, no install, no setup work for the attendee.
Clarify the goal instead of filling a form
The agent presents its company research and asks about the personal challenge. The answer is spoken, not picked from a list.
Arrive with a plan
Meeting proposals, session picks and contacts are ready before the doors open. Accepting or declining stays with you.
Fold the content in as it lands
Every talk is transcribed and processed; roughly 30 to 45 minutes after it ends it is queryable. The agent asks whether to overlay it on your goal.
Take the report, take the data
Summary, contacts and follow-ups download in an open format. After that the environment is torn down.
The technology, high level
A dedicated environment for each person—with shared knowledge and clear safeguards
Each person’s private information stays separate, and every external action requires their approval.
The technical setup is deliberately straightforward and designed for up to 300 agents active at the same time. Each person gets a separate, isolated agent environment. Its memory is stored locally and backed up daily. When it is updated, the environment is provisioned again and the required context reloaded—running systems are not modified.
- A separate, isolated agent environment for each person, provisioned from a verified template.
- A shared knowledge base with information about talks, speakers, books and the programme. Access is controlled by a revocable permission.
- External actions such as scheduling and sending messages go through a controlled gateway. Each message is approved individually.
- Daily backups of agent memory, so it can be restored after an outage.
- Attendees may optionally use their own language-model account; its key stays in their environment.
The same approach can support other events and knowledge bases: the content changes, while the technical foundation remains.
What changes
The same two days, a different outcome
Without an agent
Choosing sessions
Contacts
Scheduling
Finding knowledge again
The report afterwards
Sending a team
With an AI Chief of Staff
Choosing sessions
Contacts
Scheduling
Finding knowledge again
The report afterwards
Sending a team
Data ownership
Why attendees take part at all
An agent that speaks in a person's name needs their consent at every single step. Otherwise nobody uses it.
The agent proposes messages, meetings and posts before sending them. The person approves or declines each one.
The research about the person is on the table before the event starts. Wrong entries get corrected, unwanted ones deleted.
Every agent runs in its own container with its own memory. There is no path by which one agent reads another's memory.
Anyone who wants to put their problem to the room releases an anonymised version with the company stripped out. The person decides the cut.
Summary, notes and contacts come as a download. No vendor keeps the outcome hostage.
The window is fixed up front: start four weeks before, teardown four weeks after, then a limited period to collect the export.
Frequent questions
What organisers and attendees ask first
Do attendees have to install anything or create an account?
No. A link or QR code, the work email and a password they choose is enough; access then arrives as a magic link. The agent runs on our side, not on their device. That is deliberate: at an event, every install step costs you a large share of the attendees immediately.
Where does the data live, and what happens to it afterwards?
Each agent has its own container with its own storage on European infrastructure. The window is fixed up front: after teardown there is an export in an open format and a limited period to collect it. Then it is deleted; there is no quiet continuation.
Can an agent write something in my name that I did not want?
Nothing is sent without the person's approval. The agent presents messages and posts as suggestions first. They are sent through a controlled gateway only after explicit confirmation, with a limit of about one message per minute.
Does the agent answer in real time during a talk?
Not to the second, and we would rather say so beforehand: between a talk ending and its content being queryable there are roughly 30 to 45 minutes for transcription and processing. The live feel comes from a separate display, not from the personal agent.
Can this be applied to our own event?
Yes. We tailor the knowledge base to your event, provision the number of agents you need, and help with access and onboarding. During the event, we support operations and handle incidents. Afterwards, we provide the agreed export.
What does it cost, and what can go wrong?
Effort depends on the number of attendees and how much knowledge is ingested; we cost that per event. Honestly, the biggest pitfall is not the technology but attendee attention: an agent nobody uses on day one will not be used on day two either. That is why activation is part of the project rather than an afterthought.
Next step
AI Chief of Staff for your company
Whether in day-to-day work, across teams or at an event, let’s discuss how a personal Chief of Staff could support your company.