Plain-language definitions of the terms that come up when you give an Ai assistant or agent long-term memory. No jargon for its own sake, just clear answers.
Memory that survives between sessions, so an Ai remembers facts, preferences, and decisions from past conversations instead of starting fresh every time. It is what turns a stateless chatbot into an assistant that actually knows you.
A memory approach that stores information in a structured way, scores it by importance, and retrieves it by meaning, so recall behaves more like human memory than a flat chat log. CogmemAi's Cognitive Memory product brings this to ChatGPT, Claude, and Gemini.
An open standard that lets Ai tools connect to external servers for data and capabilities, including memory, in a consistent way. MCP is why one memory layer can work across Claude Code, Cursor, Windsurf, and many other tools.
A service that exposes tools to an MCP-compatible Ai client. A memory MCP server, such as cogmemai-mcp, gives a tool like Claude Code or Cursor the ability to save and recall memory without custom code.
The act of retrieving a relevant stored memory at the right moment. Recall quality is the real measure of a memory system: whether the right memory comes back when it is needed, which is exactly what benchmarks like LongMemEval test.
A technique where an Ai retrieves relevant external information first, then uses it to generate a grounded answer instead of relying only on what the model already knows. Memory recall is a specialized, personal form of retrieval.
Numeric representations of text that capture meaning, so that similar ideas sit close together and can be searched by similarity rather than exact words. Embeddings are what make semantic search possible.
A database built to store embeddings and find the most similar ones quickly, which is how semantic search retrieves memories by meaning even when the wording is completely different.
Search that matches by meaning rather than keywords, so a query about your kid's hobbies can retrieve a note that your daughter plays soccer. It is the difference between memory that feels smart and memory that only matches literal words.
A step that re-orders an initial set of search results by relevance, improving which memories are surfaced first. A good reranker is often the difference between a roughly right answer and exactly the right one.
The amount of text an Ai model can consider at once. Anything outside it is not seen, which is why long conversations and long-running projects need external memory rather than relying on the window alone.
When a long session is summarized to fit the context window. Useful detail is lost in the process, which is why durable memory is stored outside the conversation and reloaded when needed.
A scoring approach where critical memories stay strong while casual mentions fade over time, so the most relevant details are favored on recall instead of every passing remark carrying equal weight.
Merging and updating related memories over time so knowledge stays current and free of duplicates, similar to how human memory settles. When a fact changes, consolidation keeps the latest version and links the history.
A way of linking memories by their relationships, so related and conflicting facts can be connected and surfaced together. It helps an Ai notice when two memories disagree instead of treating them as unrelated.
A benchmark that tests an Ai system's long-term memory across many sessions and hundreds of turns. It is the field's hardest public long-term-memory test. CogmemAi holds the highest published score at 95.10%.
A long-conversation memory benchmark used to measure recall across extended dialogue. CogmemAi scored 91%, above the reported human baseline.
An Ai system that can take actions and use tools to complete tasks, not just answer questions. Agents benefit most from memory because they carry context, decisions, and progress across many steps and sessions.
A model where you connect your own Ai provider API key, so you use your chosen model directly while a service adds capabilities like memory on top. You pay the provider for usage and keep full control of your key.
Memory stored on your own machine rather than the cloud, so sensitive data never leaves your environment. CogmemAi offers a zero-config local mode for privacy-sensitive and regulated work.
Encryption designed to resist attacks from future quantum computers, used to protect stored memory and keys today against store-now-decrypt-later threats. It guards data that needs to stay private for years.
Answering questions that depend on time and order, such as what changed between two dates or what you decided first, which requires memory of when things happened, not just what was said.
Add persistent, benchmarked memory to the Ai tools you already use. Free tier, no credit card to start.
Try CogmemAi free