how the product works

How Hermes’s lasting memory works

Memory is the promise behind intakto, so here is how it works, without marketing. In plain terms: where the memory lives, how Hermes sorts it overnight, and why it stays with you when you switch the AI model.

three memory layers · nightly consolidation · a memory graph in your account

Memory is not a chat log. It is a second brain for your work

not a chat log

A chat log is a feed of messages: you scroll past it and lose it. Hermes works differently. She turns every conversation, note and document into a meaning, then links it to what she already knows about your work. So she answers not “what you typed on 3 March” but “what you decided about this client, and why.”

Ordinary search matches exact words, like reading the spines of books on a shelf. Hermes searches by meaning, the way a person recalls the gist of a conversation even when the words were different.

Three layers of memory: Hot, Warm, Cold

three layers

Memory is sorted into three layers, like the space around your desk. Hermes moves records between them herself: what you keep returning to rises to the top, the rest folds into the archive. You never sort it by hand.

Hot

Hot layer

What you are working on right now. Close at hand, answered instantly.

Warm

Warm layer

Active topics and clients from recent weeks. Nearby, but out of the way.

Cold

Cold layer

The archive. Nothing is lost; it surfaces again when you need it.

The memory is rebuilt overnight

In sleep, a person files away the impressions of the day. Hermes does something similar. Each night she goes back over what you added: she links the new to the old, lifts what matters, folds away the rest. By morning the memory is not just larger, it is better ordered. This is not a roadmap promise. It runs today.

You can watch your memory graph grow

memory graph

The memory is not a black box. Your account has a Memory Map: a plain diagram of your clients, projects and the links between them. You can see what Hermes knows about your work, and how the graph grows from month to month.

Hermes's Memory Map in your account: a knowledge graph of nodes and links on the Memory tab.
the Memory Map in your account. Each node is a fact about your work, each link is the meaning between them. Yours fills up with your own clients and projects.

You can fill the memory from day one

You do not have to start from a blank page. On the first day Hermes walks you through a short conversation about your work and helps you bring over what you already have: notes, threads, your client list. The memory fills with your own context straight away, not after months of work.

Choose the AI model, and switch it later

the model is yours to pick
models

Pick the model that fits your work

Hermes is not tied to one model. Choose the model that suits your tasks, and change it whenever you like. The assistant stays the same; only the engine behind her changes.

memory

Your memory stays when the model changes

The AI model is like an engine: you can swap it. The memory is the body and everything inside it: it is yours and it stays put. Switch the model, and Hermes keeps remembering all of your work.

The memory lives on your server, not inside the model

This is the part that makes the rest hold. The memory is a separate store on your own server. The AI model connects to it from the outside. That is why you can change the model and the memory cannot be taken away: it was never inside the model to begin with.

One plan covers it: €29 a month, the running server, the memory, nightly backups. No setup fee, no hidden tiers. The exact figures are on the price page.

How this differs from ChatGPT under the hood

With ChatGPT the memory lives inside the provider’s service: limited, and not yours. With Hermes the memory is a separate store on your own server, and the AI model connects to it from outside. So the model can change while the memory cannot be taken away. The difference is not how clever the model is. It is where the memory lives and who owns it. More in the comparison with ChatGPT on memory.

Where your data lives

ownership

Your memory, settings and backups sit on your own server, hosted in the EU, under EU rules. The access keys are in your account. The data is not with intakto, not with OpenAI, not with Google. Decide to leave, and you take everything with you. intakto keeps no copies.

questions about the product
What is Hermes's memory made of?

The memory has three layers: a Hot layer for what you are working on now, a Warm layer for active topics, and a Cold archive. Each night Hermes consolidates it: what matters rises, the rest folds away.

Does the memory survive a model switch?

Yes. The memory lives on your server, not inside the AI model. You can change the model and Hermes keeps remembering all of your work.

Where is my data stored?

On your own server, hosted in the EU, under EU rules. The keys are in your account, intakto keeps no copies, and if you leave you take everything with you.

How is this different from chat history?

A chat log is a feed of messages you scroll past. Hermes turns each conversation into a meaning and links it to what she already knows, so she recalls decisions and context, not just what you typed on a given day.

Memory you can see, and memory you own

Get started with Hermes and watch your memory graph grow from the first day.