Goooy Get more out of AI, and keep what your company learns
A company's knowledge usually sits in the heads of its employees. It walks out of the door with every departure, and every new colleague starts the climb again. In Goooy your people work alongside AI colleagues on the real work, and what gets confirmed along the way is written back into a company knowledge base that keeps growing. Nobody has to run a knowledge-management project for it, because it happens as a byproduct of the work.
Knowledge management has always failed for the same reason
It was never that nobody cared. Capturing knowledge cost effort nobody had spare, on top of the work that actually pays.
It leaves when people leave
A resignation takes years of context with it: why this customer is handled that way, which supplier to call, what was tried two years ago and why it did not work. What stays behind is a wiki nobody has updated since onboarding.
Every new colleague starts at zero
The ramp-up tax is paid again with every hire, and it is paid by your most experienced people, who spend their week answering questions they have already answered.
Writing it down is the first thing to go
Documentation lives outside the work, so it competes with the work. Under pressure the work wins every time, and the base drifts further from reality until nobody trusts it.
Capture becomes a byproduct of doing the work
Goooy's AI colleagues are provisioned as users of the suite, with their own mailbox, calendar and tasks, rather than sitting in a chat window bolted onto the side. They take part in the work instead of hearing about it later.
- 1
Work happens where it always did
Mail, chat, the shared helpdesk inbox, tasks, meetings and scheduled business processes. Your team does not change how it works, and nobody files anything into a second system.
- 2
AI colleagues do a share of it
An agent triages the support inbox, drafts a reply, chases an invoice, prepares a meeting or advances a recurring process. It reaches only the mailboxes, calendars and files its role has been given.
- 3
What gets confirmed is written back
When an agent establishes something durable and reusable (a procedure, a policy detail, the outcome of a process) it records it as a knowledge concept with its author and source attached. Transient, personal or unverified material stays out of scope.
- 4
The next answer starts from it
Every later run, and every colleague searching the base, grounds the answer in what is already known. The base gets denser as the work goes on, which is why it compounds instead of just sitting there.
Institutional memory, in a format you can read
Every concept is a markdown document with structured metadata, cross-linked to the concepts it depends on. You can open one in a text editor and understand it, and so can any tool you ever want to move it to.
Your processes, as they are actually run
Goooy ships a department-aligned catalogue of business processes (order to cash, procure to pay, incident management, recruitment and more) seeded as concepts on day one, so agents know the steps, the handoffs and the owner before they touch anything.
Confirmed facts and decisions
Which supplier we use and why, what the escalation path is, what a customer agreed to. The kind of thing that currently lives in somebody's sent items.
Procedures agents write for themselves
After a successful run an agent can distil what it did into a reusable procedure. These are text only, they can never grant an agent new access, and a person approves them before they take effect.
Your own documents, indexed alongside
Ingest the handbook, the policies and the contracts straight from the Drive, including Word, Excel, PowerPoint, OpenDocument and PDF. They become part of the same searchable base.
A graph you can follow
Concepts link to each other, so an agent asked about invoicing can follow the link to the approval policy instead of guessing. Search returns the passage, and the agent can then pull the whole concept.
Captured with permission, under rules you set
The promise is that knowledge is captured as a byproduct of collaboration. Software does not read your employees' minds, and human judgement still decides what counts.
- Writing to the knowledge base is a separate capability, off by default, granted per agent role.
- Roles set to suggest rather than act have every write queued for a person to approve.
- An agent may correct only the concepts it wrote itself. Anything a human authored is protected, and the agent is told to propose the change to a person instead.
- Every write is attributed and audited, with the agent and the timestamp on the record. Admins can edit or delete any concept.
- Agents are instructed to record only confirmed, reusable knowledge, never transient, personal, secret or unverified material.
- Each agent reads only the mailboxes, calendars and files its role has been given. There is no blanket access to the organisation.
The asset you are building belongs to you
This is the part a hyperscaler cannot offer: the knowledge accumulates on infrastructure you run, in a format you can read, and it is indexed by a model you picked yourself.
- Export the whole base as a bundle of markdown files, one per concept, with the cross-links intact. Import it back, or take it somewhere else entirely.
- Point the language model and the embeddings at an EU-resident or self-hosted endpoint, so the documents never have to leave your boundary.
- Goooy is open source and runs on your own infrastructure, so nobody else can charge you for access to what your company knows.
- What keeps you is the value the base has accumulated, and you can walk away from it whenever you like. The data is yours and stays legible if you ever leave.
Answers, including the limits
Does this mean the AI reads everything my employees do?
No. Each agent is a user with least-privilege access and reaches only the mailboxes, calendars, files and channels its role has been given. Nothing runs until an admin switches it on, and the knowledge-writing capability is separate and off by default.
What stops the base filling up with noise?
Agents are instructed to record only confirmed, reusable knowledge. Roles set to suggest have every write approved by a person first, and admins can edit or delete any concept, with a source column showing exactly what came from an agent.
Can we start from what we already have?
Yes. Ingest existing documents from the Drive (Word, Excel, PowerPoint, OpenDocument, PDF and markdown) or import an existing bundle. Goooy also seeds a starter pack of department-aligned process and policy concepts, so agents have on-policy grounding from day one.
Which model does this run on?
One you choose. Point Goooy at any OpenAI-compatible endpoint, including EU-resident providers and models you host yourself, for both the language model and the embeddings. Every run records which provider handled it.
Do we need the AI agents to use the knowledge base?
No. It works as a curated, searchable company knowledge base on its own, and admins can author and maintain it by hand. The agents are what make it fill itself.
Start the flywheel with the work you already do
Bring your suite in-house, switch on the agent roles that match your departments, and let the base build itself while your team gets on with the job.