Memory
Your agents remember, so you do not have to repeat yourself.
Everything your team learns about you and your business is kept, organised and found again when it matters. Ask once; from then on every agent, in every chat, already knows.
What it means for the work.
Most AI tools start every conversation blank. You re-explain the client, paste the brief again, remind it what you decided last week. Easy's agents check the memory before they start, so the researcher already knows which markets you care about, the writer already knows your house style, and the coordinator knows what was tried before and what happened.
That memory is shared. What one agent learns, the others can use. A note the support agent saved on Tuesday is there for the account manager on Friday, in a different chat, without anyone forwarding it.
It finds the thing, and what goes with it.
Ask "what did we agree with the late supplier" and it does not need the supplier's name, the date or the words you used at the time. It finds the note by what it means. Then it brings back the material around it: the invoice the note refers to, the email that followed, what was decided. An agent sees the decision in context, not one sentence on its own. Fold
The same works for documents and code. A contract, a price list or a whole repository is kept intact and searchable by section, so a question lands on the right clause or the right function, with a summary of what sits around it. Ask a coding agent how to add a feature to a large system and it answers with exact files and lines in one search, instead of reading its way through for half an hour.
Filed by meaning, while you sleep.
Notes do not pile up in a heap. Between conversations, a curator called Engram reads what the agents have saved and files it where it belongs: the same client together, the same project together, decisions separate from facts, events on a timeline. It links related notes to each other and keeps a short index of what matters most, so an agent can find its way in seconds. Engram
Nothing is thrown away. A note the curator is unsure about is kept and marked, not deleted. You can open the memory at any time and read, edit or remove what is there. Memory
What you use stays close. What you do not never disappears.
Memory in Easy behaves a little like your own. Things you keep coming back to stay near the top. Things nobody has needed for months fade towards the background, so a year of notes does not drown the recent ones. But faded is not gone: an old fact is still there the day it is needed again, and using it once brings it back to the front.
That is what keeps a long-running memory useful. Without it, the tenth month of notes would bury the first, or you would have to prune by hand. Memory, explained
It lives with you.
The memory runs on your own Easy installation, beside the rest of your team. Nothing is copied anywhere you did not put it. Indexing reads your material through the AI account you connected, once, and again only when something changes: a small one-off cost per document, not a running one.
You can see what is remembered, what it has cost and how healthy each index is, and you can pause, refresh or rebuild any of it. What it costs
For the technically curious.
The memory store is called Fold, a separate open project Easy runs beside itself. It extends ordinary semantic search in three ways. Every memory is linked into a knowledge graph automatically, by file relationships, similarity of meaning, links a language model suggests and the commits that touched it, so a search returns a neighbourhood rather than a point. Each memory carries a retrieval strength that follows the ACT-R model from cognitive science, fading with a thirty-day half-life towards a floor and boosted by use. And results come back grouped under their parent document with related sections nested beneath. We describe the combination as proto-holographic: any fragment reaches the context around it.
Fold is written in Rust, keeps vectors in Qdrant, splits code and prose with tree-sitter, and speaks HTTP and MCP so tools such as Claude Code and Cursor can query it directly. Its lineage runs through A-MEM (NeurIPS 2025), Kanerva's sparse distributed memory and Plate's holographic reduced representations.