Shared company memory: an AI archivist keeps the decision log and the glossary

In a company of ten or twenty people, decisions are made fast and remembered poorly. Why did we pick one supplier over another, when did we stop an offer, what does that acronym everyone uses stand for? The answer exists somewhere, in one person's head or in a discussion thread, and it vanishes when that person is gone or has no time. Agents raise the same problem more sharply: an AI agent has only the memory you give it, and it needs the company's context to work well.

The inspiration: one memory for teams and agents

The starting point is a use reported by Hermes Agent users, summarized in a card titled "shared company memory". The idea is to stop leaving this memory scattered: a single place, kept by an agent whose job it is, which teams and other agents can query.

What we take from it is the shape. A useful memory is not one big text but two simple objects you know how to maintain: a decision log, which answers "when, who, why", and a glossary, which answers "what does it mean". Atako applies this idea with an archivist, a project wiki, and strict rules on what it may write.

What the archivist does

The scenario deploys a single agent, the Archivist, which keeps two pages in the wiki of the "Company memory" project.

The decision log holds one entry per decision, newest first. Each entry carries the date, the decision in one sentence, the reason, the people or team who decided, the impact and the source. When a decision changes, the archivist does not rewrite the old one: it adds a new entry that points to it and marks the old one as superseded. The history stays complete.

The glossary holds the company's own terms: products, internal names, acronyms, with a short definition, the team that uses it and the date it was added. The archivist merges duplicates and synonyms instead of listing them twice.

At startup, it records the recent decisions you listed and seeds the glossary from your company description. After that, you query it in the chat. To a question like "why did we choose this malt supplier?", it searches the log and the glossary first, then answers with the history: what was decided, when, by whom, why, and the source. When the memory does not answer, it says so, lists what is missing and says who probably knows. Every Friday, it writes a "What changed this week" page: new decisions, superseded decisions, new terms, gaps still open.

What the archivist does not do

The limits are part of the design. The archivist records and answers, it does not decide: a recommendation is never a log entry until a person has made it. It never invents a decision, a date, an author or a reason: "unknown" is an acceptable answer, a guess is not. It cites the source of each answer and, when sources disagree, it shows both.

It records no credentials, passwords or personal data of private individuals. It contacts nobody outside the company. And it has no integration: it reads neither your mailbox nor your tools, it knows only what it is told and what it has already recorded. This human-in-the-loop design leaves the decisions themselves to people.

How to set it up in Atako

From the page of this use case, deployment takes four steps.

1. Describe the company. A few lines on what the company does, its teams and the terms specific to it. The archivist uses it to seed the glossary.
2. List the recent decisions. One per line, up to ten, with the decision maker and the reason if you know them. These are the first log entries.
3. Deploy. A single action creates the agent and the "Company memory" project with its pages. No integration to connect.
4. Ask a first question. Ask the archivist why one of the decisions was made, and check that its answer cites the log.

If you do not have an account yet, creating one starts your trial and the scenario is configured right after. If your administrator has not allowed you to create agents, the deployment tells you and stops there.

Where to see the results

The result is read in the Atako project "Company memory": the "Decision log" page, the "Glossary" page and, on Fridays, the "What changed this week" page. In the chat, the archivist's answers cite their sources, so you can open the original entry. On the Atako floor, the agent appears with its current activity.

To record a decision or a term, announce it to the archivist in the chat: it adds it right away and confirms where. Requests are handled in the same conversation, with the same memory of what has already been recorded.

Guardrails

The agent has no integration. Its only write access is the wiki of the project created for the scenario, and it contacts nobody outside the company. The log and the glossary are pages you can read, correct or complete yourself, like any wiki page.

Human validation sits where it matters. People make the decisions, the archivist records them. Entries the agent marks "unknown" or "to be confirmed" wait for a person to complete them, and the gaps it flags are a list of things to ask, not a hole to fill at random.

What to expect

Expect a short log at first and answers that improve as it fills up. The first questions on unrecorded topics will get the answer "I don't know, here is what is missing": that is the intended behavior, and the list of gaps is useful in itself. The value shows when a new hire, or another agent, gets the reason for a past choice in seconds, with its date and source, instead of hunting for who remembers. For a small team, it is the difference between a memory that depends on who is present and a memory that stays in the company.