---
title: "Case: build an AI assistant on your team's memory"
url: https://memocore.ai/memos/6a587afcac52d245773487b9
project: "Memocore Knowledge Base"
tags: "use-cases, api"
updated_at: "2026-09-07T02:46:21.941Z"
---

# Case: build an AI assistant on your team's memory

A real example of using Memocore as the brain behind your own product: a chat assistant that answers from your team's knowledge.

How it works. Mint a company API key scoped to the projects the assistant may read. Your app then searches that memory over the API — search by meaning is built in, so there is no vector database to run. Take the most relevant results and hand them to an LLM, which writes the final answer in the user's language.

Memocore's own site assistant is built exactly this way. A company key defines which projects it can see, every question triggers a memory search first, and the model answers only from what it finds — if nothing matches, it says so instead of inventing an answer.

Why it is safe to expose. The key is project-scoped: it can only ever touch the projects you selected for it, and an id from anywhere else answers 404. Everything else stays invisible to it. Scope, not trust, is what keeps it contained — so give an assistant a key covering exactly the projects it needs and nothing more.

Writing back. The API can also create memos, which turns the same pattern into an assistant that remembers. A support bot can file what a customer told it; an agent can record what it learned on a run, so the next run starts from it. If you would rather the assistant never wrote anything, simply do not call the create endpoint — but keep in mind the key itself is not restricted to reading.

This is the same pattern any team can reuse to put a support or onboarding bot on top of its Memocore knowledge, in its own app or website.