glossary · What it keeps

Agent memory

definition

Agent memory is persistent, retrievable knowledge that carries across AI sessions: what past runs learned, what was decided and why, and where things live, recalled before acting rather than rediscovered.

in one line: What past runs learned, recalled before the next one acts.

in context

Where does agent memory fit?

Look first. Work. Keep what you learned. In ConvOps, every run asks before it acts and saves on the way out, so the next run starts ahead.

recallask shared memory
workright files
decideapproval
capturememory + reason
next runstarts ahead
two ways to say it

What is agent memory, in plain words?

Same idea at two depths: the plain version, then what the engine actually does.

Every new chat starts from zero. Memory keeps what your team decided, and where things live, so the next chat starts knowing it.

Teach it once. Every AI remembers.

for engineers

Memories (the what) carry the situations they apply in. Recall ranks by meaning, situation, recency and use. Routes (the where) map files and docs.

context_query returns routes, memories and related tasks in one call. Forgetting is soft and can be undone.

the longer answer

Why does agent memory matter?

It is distinct from two things it is often confused with. Context windows are working memory: everything in them evaporates when the session ends. And retrieval over documents (RAG) recalls what the organization wrote, but not what the agents themselves learned by doing the work.

Useful agent memory splits WHAT from WHERE. Knowledge (a decision, a correction, a learned constraint) ranks by relevance, recency and how often it proves useful. Location (which file, which module, which doc) is a map that spares the agent re-exploring territory it has already covered.

The capture side matters as much as recall: memory written automatically from everything becomes noise, and memory that depends on someone remembering to save it becomes empty. The durable pattern wires capture into the process itself, at the steps where decisions actually happen.

Memory also needs upkeep to stay useful. Knowledge goes stale: a decision is reversed, a file moves. A good memory system lets the next run correct an entry rather than add a contradicting one, and lets a forgotten entry come back when the forgetting was a mistake. Without that upkeep, the same question is answered again and again, and the answers drift apart.

in convops

How does agent memory work in ConvOps?

In ConvOps, memories carry a category and the situations they apply in, and recall ranks them by meaning, situation, recency and use. Routes map where things live: files, modules, docs. One call returns the relevant routes, memories and related tasks together, so the agent looks first and acts second. A memory can be captured with the task step that produced it, recall spans every workspace you belong to, and forgetting is soft and can be undone. A route points the agent at the right file before it starts searching.

questions

What do people ask about agent memory?

What is agent memory in ConvOps?

Agent memory is knowledge that carries across AI sessions: what past runs learned, what was decided and why, and where things live. In ConvOps, memories carry a category and the situations they apply in, recall ranks them by meaning, situation, recency and use, and routes map where things live, so the agent looks first and acts second.

What is the difference between agent memory and a context window?

A context window holds what the model sees in one session and is lost when the session ends, while agent memory persists across sessions. In ConvOps, decisions, corrections and constraints are stored as memories that later runs recall, so the next session starts from what the last one learned. Memory feeds the context window only the parts worth knowing.

What should an agent save as a memory?

Knowledge a later session would otherwise have to rediscover: a decision and its reason, a correction from a reviewer, a constraint, an approach that failed. In ConvOps each memory carries a category and the situations it applies in, and where things live goes into routes instead, so recall returns short, specific facts rather than whole documents or transcripts.

Can agent memory be shared across a team?

Yes, when it lives outside each person's client. In ConvOps memory belongs to the workspace, so every member's agent recalls it over MCP, and recall spans every workspace you belong to. Forgetting a memory is a soft delete that can be undone.

Whose work improves most with agent memory?

People who run AI agents on the same codebase or the same operations again and again, where a correction made today should hold tomorrow. ConvOps memory serves engineers, operators and content teams alike, because decisions, constraints and fixes are recalled in later sessions instead of being explained again at the start of each one.