Access Evo Researcher has two features called memory, and they do different jobs. AI Memory lets the AI search the documents in your project. Observational Memory keeps a long chat consistent. This article explains what each one holds, so you know what the AI can and can't recall.
AI Memory
AI Memory is a searchable library of the documents in your project. When you ask a question, the AI searches this library and pulls in the relevant parts to inform its answer.
How AI Memory is filled
You add documents to AI Memory by uploading files to a project. Files that come in when you create a project from a template are added too. Everything you upload this way becomes searchable.
Where AI Memory applies
AI Memory is project-wide and lasts across sessions. Every chat in the same project can search the same library, including chats you start on another day. Each project has its own library, so AI Memory isn't shared between projects.
đ Note: To search beyond a single project, use the Evo Researcher connector. It can search across the projects you own or collaborate on, so you can pull together information from several projects at once.
For example, you upload last quarter's sales report and a pricing policy to a project. In one chat, you ask the AI to summarise the report. The next day, in a new chat in the same project, you ask "what's our discount limit?" The AI answers from the pricing policy, because both files are in the project's AI Memory and any chat in that project can search them.
What AI Memory doesn't hold
AI Memory doesn't store:
Your conversations. It has no record of what you asked or what the AI replied in an earlier chat.
Facts you type into a chat. Typing a fact into a message doesn't add it to the library. Only uploading a document does.
Documents the AI creates for you. These are saved separately as artifacts and aren't part of the searchable library.
đ¤ Tip: If you want the AI to use a piece of information in future chats, upload it as a document.
Observational Memory
Observational Memory keeps a single chat consistent as it gets longer. It runs automatically in the background, so there's nothing to switch on or configure.
How Observational Memory works
As a conversation grows, Access Evo Researcher condenses the earlier messages into short, dense notes rather than re-reading every message each time. This keeps the AI consistent over a long session. It holds on to the constraints and decisions you set earlier, doesn't lose the thread, and doesn't repeat work it has already done.
Where Observational Memory applies
Observational Memory is scoped to one chat. It applies only to the conversation you're in and doesn't carry over to other chats. If you start a new chat, the AI doesn't remember the previous one.
For example, in a long chat you refine a brief over several messages: "keep it under two pages", "British English" and "no jargon". Twenty messages later, you ask for a final draft, and the AI still applies all three rules. If you open a separate chat and ask for "the final draft", the AI doesn't know which brief you mean.
How the two features differ
Aspect | AI Memory | Observational Memory |
What it holds | Your project's uploaded documents | The history of one conversation |
How it's filled | You upload files | Automatically, as you chat |
Scope | The whole project, across every chat | A single chat only |
Lasts across sessions | Yes | Only within that chat |
You control it | Yes, by uploading and managing files | No, it runs automatically |
Purpose | Lets the AI reference your documents | Keeps a long chat consistent and affordable |
What to expect across chats
Within a project, any chat can use the documents in AI Memory. Uploaded material is available in every chat for as long as the project exists. The AI doesn't carry a conversation from one chat into another. That continuity comes from Observational Memory, and it stays inside the chat where it happened.
đ¤ Tip: If something matters beyond the current conversation, don't rely on telling the AI. Put it in a document and upload it, so every chat in the project can use it. You can also create a skill and attach it to a chat or project.