| Address | A short link that starts with memo://, pointing at exactly one saved item. It keeps working even if the item is later forgotten |
| Agent | An AI assistant, such as Claude, that uses tools on your behalf. Also the lowest trust level: “Claude saved this” |
| Browser view | The read-only web page that shows your memory: memo-mcp ui |
| Curated | The highest trust level: checked and approved as a reference source |
| Degraded | Searching works, but only by matching words, because the meaning-matching model is not ready |
| Document | Anything longer you saved: a note, a guide, a web page, meeting notes |
| Embedding | The way a language model turns text into numbers so it can match meaning rather than exact words. You never need to handle them |
| Fact | One short statement worth keeping on its own, with a date range and a source |
| Forget | Remove an item from searches, keeping a record that it was removed, when and why |
| History | Earlier versions of documents and facts, kept when something changes |
| Language model | The downloaded file that lets memo-mcp match by meaning. It is downloaded once and runs on your computer |
| MCP | Model Context Protocol: the standard way Claude plugs into tools like memo-mcp |
| Memory, or knowledge base | Everything memo-mcp holds for one project, kept in one file |
| Page | A summary Claude wrote about one topic, built from saved passages and pointing back to them |
| Passage | A short piece of a document, a few paragraphs long. Searches find passages, then show the document they came from |
| Promote | Raise an item’s trust level. Only a person can do it |
| Shelf | A label that groups documents within one memory, such as billing or infrastructure. Technically called a namespace |
| Stale | A summary page whose sources changed after it was written; it needs a refresh |
| Trust | Who has vouched for an item: Claude (agent), a person (user), or a checked reference (curated) |
| User | The middle trust level: a person saved or confirmed it |