One company memory. Every AI on it, at once.
Agentu Professional is the company workspace: encrypted memory repositories that many AIs work at the same time, across departments, projects and divisions. One brief goes to every AI the company allows, they compete or collaborate on it, and the work lands back in the repository with its evidence. Smaller teams ship faster because every AI works the same brief concurrently.
Several AIs, used one at a time, each holding a slice of the company
Companies now run their work through several AI apps. Each is used by one person, one window at a time, and each holds a different slice of the company's context, what was decided, which documents matter, what was drafted last week, mixed with everything else that person has ever told it.
Work waits in a queue: the brief goes to one AI, then its answer is pasted into the next. The context is invisible to colleagues, impossible to hand to the next person, and left behind in the vendor's systems when the employee leaves. A departing employee's company context stays inside three AI accounts the company does not control, which is now a data-protection problem as much as a knowledge problem.
What Agentu Professional does instead
It gives the company one encrypted memory, organised by department, project and division, and lets every AI the company allows work on it at the same time. A brief is sent once and every AI works it concurrently; what they produce lands back in the repository with its evidence; every person is an ID code outside the vault; and removing a person, and their data, is a single, evidenced action.
One memory, many AIs, at the same time
Every AI the company has connected, ChatGPT, Claude, Gemini, Codex, Cursor and anything else that speaks MCP, reads and writes the same memory repositories concurrently. Nothing is pasted between windows, nothing waits for a person to carry it across, and each AI sees only the categories the admin allowed it.
Departments
Finance, legal, marketing, engineering, operations: each department is a repository with its own keys, its own allowed AIs and its own retention. The AIs working for finance cannot see that the legal repository exists.
Projects
A supplier dispute, a year-end audit, a fundraising round, a product launch. Each project holds its memories, a work log, a documents index, its people and its status in plain words, and every AI on it draws on the same facts.
Divisions
Regions, subsidiaries and business units get their own cell: residency, retention and allowed AIs set once at the division and inherited by every project beneath it. Residency →
Concurrent, not queued
Four AIs on one project do four things at once from the same facts: one drafts, one researches, one checks the numbers, one produces the cut. The work log shows which did what, in what order, with what evidence.
Smaller teams ship faster
Because every AI works the same brief concurrently, the wait between one AI's answer and the next AI's turn disappears. The team sets the brief, judges the results and decides; the AIs do the parallel work, and the work ships sooner.
Evidence in the work log
Every AI session, every document used, every decision taken is logged against the project and the ID code that ran it. Metadata only by default; the content stays encrypted. The log is what an auditor, a client or a regulator sees.
One brief to all AIs
Write the brief once, in the project. Choose who gets it, every AI the company allows on that project or a subset, and choose how they work.
- Compete. Every AI answers the same brief independently. The team sees the answers side by side, picks the best or merges them, and the choice is recorded in the work log.
- Collaborate. The brief is split into parts, each AI takes the part it is best at, and the result is assembled in the repository with every step logged.
- Launch pipelines, including video. A product launch is one brief: positioning, page copy, the e-mail sequence, social posts, the launch video script and its cut, each produced by the AI best placed for it, at the same time, from the same facts, and reviewed in one place before anything leaves the building.
- Every answer carries its provenance. Which AI, which model, which categories it was allowed, when, and under which ID code.
One agentu.com login for every AI
A person signs in to agentu.com once, with a passkey. Every AI the company has allowed, ChatGPT, Claude, Gemini, Codex and the rest, works from that one identity and that one memory, and sees the person's ID code rather than their name. No AI holds a separate account to be forgotten when the person leaves, and no AI ever receives a key: it receives memories, on demand, for the categories it is allowed.
In build How AI apps connect today →
The admin portal
One place where the company sets the shape of its memory and the rules every AI works under.
Structure
Create departments, projects and divisions; set which repositories inherit which; move nothing to a less strict cell without a recorded decision.
AIs and categories
Which AIs each repository may use, which categories each AI may see, which model terms apply, and whether a project is Ask me first for sensitive categories.
People
Add a person, an employee or an outside adviser, with a role, a scope, the categories they may see and an expiry. Issue their ID code. Remove them in one action.
Residency and retention
Switzerland by default; the United States and Singapore as they go live; Dedicated, a single-tenant Agentu in the region the company chooses, on request. Retention and legal hold per project.
Work log and audit export
Every AI session, document reference and decision, by project and ID code, exportable for an auditor or a client without exposing content.
Seats and AIs
See who has which AI, give a person a seat from the Add people sheet, and watch which seats are still pending. Planned
Inside a project
Anyone added to a project, a new lawyer joining the matter, say, sees in one place all prior AI work, the data, the documents and what has already been completed, and their own AI apps draw on it through Agentu from the first minute.
Memories
The facts, decisions and correspondence the work needs, encrypted per project and per category, recalled by the AIs the company allows for that project.
Work log
What has been done: e-mails sent, documents used, AI sessions run, briefs issued, decisions taken. Metadata only by default; one click from Gmail, Outlook or Word.
Documents index
Where each document lives, in Drive, SharePoint, a file share or a vaulted copy, with who last changed it. References, not copies, unless the company chooses otherwise.
People
Employees and outsiders such as the company's lawyers, each with a role, a scope, the categories they may see and an expiry. Personal memories stay in the person's own Agentu vault, which the company cannot see.
Status
The state of the work in plain words: "Advice drafted, letter before action due 14 Oct". And the AI situation: which AI produced what, when, and whether it was saved.
Every AI, through MCP
ChatGPT, Claude, Gemini, Codex and Cursor recall the right project memory for a question, save what they produce back into the project, and see ID codes rather than people. Live for personal memory today.
The company sees every person as an ID code
Names, e-mail addresses and roles live only inside the encrypted vault. AI vendors, logs, exports and audit trails see AGU-7K3M-Q2X9. The same person has a different code in each company, so vendors and logs cannot link a person across companies. Apps and add-ins put the name back on screen locally, for the people who are allowed to see it.
When a person is removed, their mapping record is shredded. The project history stays attributed to the code, and the code can no longer be linked to a natural person by anyone, including Agentu. That is the pseudonymisation-to-anonymisation step the company's Article 30 record cites.
What leaves the vault carries the code
- Briefs and prompts sent to AI vendors
- Audit events and exports
- Erasure certificates
- Vendor API calls
Optional masking replaces names, e-mail addresses, phone numbers and identifiers in memory text before it reaches a vendor. Whether the shred satisfies the anonymisation standard in each jurisdiction is for counsel to confirm before any public claim; we do not make that claim here.
Remove a person in one evidenced action
When someone leaves, the company removes them in one action. Agentu revokes, rotates, shreds, and requests and evidences vendor deletion. It never shows "deleted" for a step the vendor has not confirmed.
In Agentu, always
- Sessions and tokens revoked, including the single login every AI used; memberships ended
- Project keys rotated, so the person's devices can open nothing new
- Identity mapping shredded: the ID code becomes anonymous
- Personal vault untouched and returned to the person
- A signed erasure certificate, company, code, date, every step with its evidence, and an updated Article 30 record
In connected systems, honestly
Capabilities as checked on 7 October 2026. A step the vendor cannot fulfil programmatically is shown as Requested, awaiting vendor with a due date, reminders and a place to attach the confirmation.
| System | Done directly | Requested and tracked |
|---|---|---|
| ChatGPT Enterprise / Edu | Deprovision via SCIM; delete the conversations, files and memories the Compliance API allows | Anything outside the API, until the retention window expires |
| Claude (Team, Enterprise, Platform) | Remove from the organisation and workspaces via the Admin API; deactivate API keys | Deletion of conversation history, filed through Anthropic's privacy route, with the request id |
| Google Workspace | Suspend, revoke tokens, transfer Drive and Calendar | Account deletion after the company's retention period |
| Microsoft 365 | Disable the account, revoke sessions, remove licences | Deletion under the tenant's retention policies |
| Other SaaS | SCIM deprovision where supported | Deletion request by e-mail, tracked to a confirmation |
Works where people already work
Gmail, Drive, Docs
The Chrome extension connects Agentu to Google Workspace. In Gmail the side panel recognises a thread as a project and offers Log to project; in Drive and Docs it saves document references, never copies.
Outlook, Word, Excel, Teams
A Microsoft 365 add-in puts Agentu, and the person's chosen AI, ChatGPT, Claude or Gemini, natively into Office: the project summary in a pane, a draft from the project's facts, the session logged.
Every AI, through MCP
ChatGPT, Claude, Gemini, Codex, Cursor, VS Code and any other MCP client connect as they do for personal memory today, with the single login and one button per app in build. Connect AI apps →
Who holds the company's keys: two options, chosen at sign-up
Managed by Agentu, the default. Company and project keys live in Agentu's hardware-backed key store in Switzerland. Memories are encrypted there and opened by the platform only to answer an AI the company allows, never stored in plaintext, every open logged in the work log. People unlock with their phone; no codes to type; admins recover access without any phrase. Nobody at Agentu reads memories. The honest limit: the platform can open memories to answer the company's AIs, so this is not zero-access; a compromised live server during the window of a request is the limit.
The company holds its own keys. Members' devices hold the keys, and a quorum of the company's admin devices can recover them, so a leaver can be removed without losing the work. Agentu's servers never hold a company key that opens anything. The company root key is never stored whole; it is split across the admin quorum's devices, with an optional extra share in an HSM or vault the company chooses. Either way, project keys are versioned, wrapped for each member device or for the key store, and rotated when a person leaves. How this fits the security model →
Personal vaults stay the person's own, whichever option the company chose, and each person decides how to hold their own key: managed by Agentu by default Live; "Hold my own keys", keys only on their devices with a 24-word recovery phrase Live; or the Agentu Sealed upgrade, the key sealed to their biometric In build. The company cannot see a personal vault in any of them. Biometric-Sealed Encryption →
The Dedicated option
A single-tenant Agentu operated by Agentu for one company, in the region the company chooses, including Swiss hardware later. The whole workspace, company repositories and personal vaults alike, runs on an instance that serves nobody else, with its own address for ChatGPT, Claude and every other AI the company connects. The same two choices of who holds the keys apply. Per deployment, on request. What Dedicated is →
Residency and compliance frame
- Switzerland is the home cell. Projects inherit the division's cell and cannot move to a less strict one without an admin decision recorded in the audit log. Residency →
- GDPR and UK GDPR: minimisation by design, erasure through removal, portability through encrypted export, an Article 30 record generated from the data map, DPAs with every processor including AI vendors.
- United States: CCPA/CPRA and state laws; sector rules declared on a project (HIPAA projects default to Ask me first and US residency).
- Retention and legal hold per project; legal projects can mark categories Privileged. Compliance →
The lawyer, the leaver, the launch
A. The lawyer joins the matter
Acme has project PRJ-0042 Supplier dispute. The owner clicks Add people, enters the lawyer's address, picks External adviser, this project, the categories she may see (Facts, Decisions, Correspondence, Documents, not Finance, not Private notes), expiry end of matter. She signs in with her own agentu.com identity and sees the Overview: status, what ChatGPT and Claude already produced, the documents and where they live, the decisions taken. Her own AI works from the same facts, and her work flows back in. When the matter closes, her access expires and the project key rotates.
B. The employee leaves
The admin opens People in the portal, selects AGU-7K3M-Q2X9 and clicks Remove person. The wizard shows what will happen in Agentu and in each connected system, the admin confirms with a passkey, and the job runs. Each step shows done, requested, awaiting vendor or failed. The result is a signed erasure certificate and an updated Article 30 record. The person's personal vault is untouched and theirs.
C. The launch
Marketing opens PRJ-0051 Spring launch and writes one brief: positioning, page copy, a five-mail sequence, social posts and a ninety-second video. She sends it to all four AIs in collaborate mode. Claude takes the positioning and copy, ChatGPT the e-mails, Gemini the posts, and the video AI the script and the cut, all at the same time, all from the same facts. By lunch the drafts sit in the project with their provenance; the work log shows who produced what; she picks, edits and approves. A small team shipped by lunch what used to wait a week in a queue.
Who it is for
| Person | What they need |
|---|---|
| Company admin / DPO | Set up the company in the portal, connect its AIs and enterprise systems, set policy (which AIs, which categories, residency, retention), add and remove people, prove compliance. |
| Project owner | Open a project, pull in the people, write the brief once, see the whole state of the matter in one screen, know what every AI has already done. |
| Employee | Work in Gmail, Word and ChatGPT as usual; sign in once at agentu.com; have the project's context follow them; log work with one click; keep personal memories private from the company. |
| Outside adviser | Be added to a client's matter and immediately see prior AI work, documents, correspondence and what is done; let her own AI use it; have her access end cleanly when the matter closes. |
| The AIs | Work the same repository at the same time, recall the right project memory for a brief, save what they produce back with provenance, and see ID codes rather than people. |
Giving people an AI
From the Add people sheet a company can give a person a native seat sourced through Agentu's partner channels (Google Workspace, Microsoft 365, Claude for Enterprise), with Agentu's partner discount passed on where the vendor allows, or an Agentu AI package on Agentu's own keys, billed to the company. The company stays the account holder of every AI tenant; Agentu acts as delegated admin through the agentu.com login, with SSO and SCIM in build. The seat shows as pending until the person accepts; the portal lists who still has no AI. Planned
Pricing
Agentu Professional is from $20 per person a month, or $200 a person a year, and a seat is charged only for the features that are live. Dedicated, a single-tenant Agentu operated by Agentu in the region you choose, is per deployment, on request. External-adviser seats: the proposal is that guests are free and the inviting company pays. Nothing on this page is an offer; the agreement is. Pricing →
How a company starts
Choose a home
Switzerland (Zurich) by default: storage, backups, keys and processing under Swiss law. A regional home in the United States or Singapore where a sector rule requires in-country storage, as each goes live. Or Dedicated: a single-tenant Agentu operated by Agentu in the region you choose, on request. Then choose who holds the keys: Agentu, or the company. Residency →
Sign the agreements
A data-processing agreement with standard contractual clauses where needed, a security annex that matches the security page, and the model-terms schedule for every AI vendor you connect.
Add people, connect AIs, send the first brief
Admins create the workspace, its departments and projects; people receive an ID code and one agentu.com login; the company's AIs connect; nothing is copied anywhere. Removal is one evidenced action.