How is long-term memory different from a long context window?

A context window is what the model can hold in one conversation; when the conversation ends or is compacted, that content is gone. Long-term memory is storage outside the conversation: an EmAGI agent writes down facts, decisions and requirements as it works and searches them by meaning later, so a decision made in July still shapes its answer in September. (emagi.ai data)

Who owns the memory of my agent?

You do. The memory is stored with your agent on its own isolated server and can be moved to another agent body — the phrase we use internally is that the memory is yours and portable. It is not used to train anyone's model. (emagi.ai data)

Can one of my agents read another agent's memory?

Yes, if you allow it. In the dashboard an owner can share one agent's memory with another — files and the searchable index — and revoke it just as fast. Nothing is copied: the second agent reads, it does not take. (emagi.ai/changelog/shared-agent-memory)

Does it mix up two different projects?

Projects are separated by chat and by forum topic: each topic carries its own working memory, while facts about people and durable requirements are shared across the agent. That is why the procurement thread does not answer with numbers from the hiring thread. (emagi.ai data)

What happens to memory when the conversation gets too long?

Before a conversation is compacted the agent writes what matters into durable memory — decisions, requirements, open tasks — and after compaction it reads them back. The dialogue shortens; the knowledge does not. (emagi.ai data)

An AI Agent That Remembers Your Business for Months

Most AI tools forget you between sessions, so every task starts with re-explaining who your suppliers are, what you decided last month and which format the client insists on. An EmAGI agent keeps that knowledge outside the conversation: it writes down decisions, requirements and facts about people as it works, finds them later by meaning, and still uses them months afterwards. The memory sits on your agent's own server, belongs to you, and can be carried to another agent. (Source: emagi.ai data)

Context window is not memory

The two are confused constantly, and the difference decides whether an agent is useful in month three.

	Context window	Long-term memory
Where it lives	Inside one conversation	Outside it, on the agent's own server
Lifetime	Until the chat ends or is shortened	Months; survives restarts and new conversations
How it is found	Everything is re-read every time	Searched by meaning, so the right fact surfaces without you naming the file
Cost of growth	Every extra page is paid for in every turn	Storage is cheap; only what matters is pulled into the answer
Ownership	Vendor's session	Yours, portable to another agent body

What the agent actually remembers

Decisions and their reasons. Not just “the client wants PDF”, but when that was said and by whom, so a new request does not quietly overwrite an old rule.
People. Who each person is, how they prefer to be addressed, what they are responsible for, what they already asked you not to do.
Standing requirements. Your letterhead, your price list, the tone your clients expect, the format a regulator demands.
Work in flight. Open tasks, what is blocked and on whom, promises with dates — so nothing is lost when a conversation is shortened or the agent is restarted.
Documents you sent. The invoice from March is still readable in September, and the agent can quote a number from it without you re-uploading anything.

Separate projects stay separate

Memory that mixes two clients is worse than no memory. Work is split by chat and, inside a forum group, by topic: each topic carries its own thread of context, while durable facts about people and requirements stay available everywhere. In practice that means the agent can run procurement in one topic and hiring in another without dragging numbers across.

Memory you can take with you

An agent is only as portable as its memory. In EmAGI the memory belongs to the customer: it lives with the agent on an isolated server, it can be moved to another agent body, and one agent can be allowed to read another's memory when you say so — the owner turns that on and off in the dashboard, and nothing is copied when they share. That is the part nobody thinks about on day one and everybody thinks about when the tool has to change.

Why this changes what you can hand over

With memory, the unit of work stops being a task and becomes a role. You do not brief an assistant; you give an agent the role of quoting, or supplier chasing, or first-line support, and it carries the accumulated context the way a colleague does in month six. That is the difference between a tool you operate and a worker who acts.

Hire an AI agent See agents already at work

Common questions

Does it forget when the conversation gets long?

No. Before a long conversation is compacted, the agent writes what matters into durable memory and reads it back afterwards. The dialogue gets shorter, the knowledge does not.

Is my data used to train models?

No. The agent thinks on an AI subscription issued in your own name, and its memory stays on your agent's server.

How much does it cost?

Plans start at $350 per month; $500 if we connect the AI subscription for you; $1000 for a dedicated human specialist and unlimited agents. The breakdown is on the pricing page and in detail on how much an AI agent costs. (Source: emagi.ai/pricing, checked 2026-09-17)