GuardianMemory

Guides to AI memory

Plain-English guides to giving AI assistants a memory that lasts: what agent memory is, how a company brain works, what an MCP memory server does, and how to share one memory between ChatGPT, Claude, Cursor and the rest.

Already using GuardianMemory? The docs cover connecting each assistant, the instructions to give them, and exactly how memories are kept clean.

AI memory glossary

The terms these guides use, in a sentence or two each.

AI agent memory
What an AI agent keeps beyond a single conversation: facts, preferences, decisions and events, stored outside the model and brought back into its context when they matter. Read the guide.
Agent amnesia
The way AI agents start every conversation knowing nothing, because the model itself keeps nothing from one conversation to the next.
Context window
Everything a model can read at once: its instructions, the conversation so far and any tool results. It is emptied when the conversation ends, which is why an agent needs memory outside it.
Context engineering
Choosing what goes into a model's context window, and when: instructions, documents, tool results and memories. Long-term memory is the part of it that outlives the conversation.
Long-term memory
A store that an agent writes to and searches, and that outlives the conversation. In GuardianMemory, each long-term memory is one self-contained sentence.
Embedding
A list of numbers that captures what a piece of text means, so that texts with similar meanings sit close together. GuardianMemory stores a 768-number embedding with every memory.
Search by meaning rather than by matching words, using embeddings. “What database does billing use?” finds “The billing service runs on Postgres and Stripe”.
Retrieval-augmented generation (RAG)
Looking things up in a set of documents and putting what is found into the model's context. Agent memory is read the same way, but it is written as work happens and changes over time.
Model Context Protocol (MCP)
The open standard AI assistants use to connect to outside tools and data. An MCP server offers tools, and the assistant, the MCP client, decides when to call them. modelcontextprotocol.io.
MCP memory server
An MCP server whose tools save and search memories, so that every assistant that speaks MCP can share the same memory. Read the guide.
Memory bank
In GuardianMemory, where memories are kept. Every account has a personal bank, and team banks are shared.
Team bank
A GuardianMemory bank that several people share, with owners, members and readers. An assistant that writes to a team bank cannot read anybody's personal bank. How team banks work.
Company brain
One shared memory of what a company knows, such as decisions, systems, owners and conventions, that people and their AI assistants read from and add to. Read the guide.
Memory profile
What GuardianMemory returns when an assistant searches with no query: pinned memories, then the most used, then the newest. It gives an assistant the basics before it knows what to ask.
Pinned memory
A memory that leads the profile every assistant reads first. Pin the handful of facts that should shape every answer.
Near-duplicate merging
Treating two memories that mean almost the same thing as one. GuardianMemory merges a new memory into an existing one when their cosine similarity is 0.95 or more.
Superseded memory
A memory retired because a newer one replaces it. People can still see it as history, and assistants are no longer shown it. How memories are kept clean.