This page shows what a normal day with Memo looks like once it is installed: bringing in what changed, finding something, preparing for a conversation and saving what you learned. Each step links to the page with the details. Most of the time you ask your agent and it runs the commands; the commands are here so you can check what it does.

Memo reads your folders and mailboxes when you ask, not all the time. One command brings in everything that changed since the last run:

memo sync

Only changed files and new mail are read, so a second run is quick. On Linux, memo auto start runs it for you every few minutes (keeping Memo up to date). Telegram and MAX messages come in through tg and max, not through Memo.

Use memo CLI to find what I wrote or received about the venue for the Harbour project.
Show the matching text and the source of each result.

Your agent searches your notes, mail and messages together and shows where each result comes from. When you know where something is, search only there:

memo search all 'venue'               # messages, mail and notes together
memo search notes 'venue'             # notes only
memo search mail 'from:kai invoice'   # mail only

Search explains the search words and how to give your agent the sources.

Use memo CLI to show what I know about Kai Sample on Telegram: recent messages, mail and my notes.
Summarise what is still open between us.

memo context telegram:"Kai Sample" gathers the person's recent messages, mail and the notes linked to them, each with its source. See People and projects.

memo notes add 'Prefers calls in the morning' --subject telegram:'Kai Sample'
memo tags add follow-up --note note:NOTE_ID
memo tasks add --project HRB --title 'Book the venue'

A note keeps your own words about a person, a message or a project (notes). A tag is a label that helps you find things again (tags). A task is something to do (tasks).

An agent can propose an action, such as a reply. An app connected through MCP can also save memories and facts. These wait for you:

Use memo CLI to show the memories, facts and actions waiting for my review, with their evidence.
Do not confirm or approve anything; I will decide.

Memories and facts explains how to confirm or reject them. Approving an action makes it a task for you; Memo never sends anything. A memory or fact your agent adds with the memo command waits for you too: memo records the agent as its author when AI_AGENT or CLAUDECODE is set. One you add yourself is confirmed at once.

Add --json to most commands to get an answer a program can read. The text answer is for people; agents and scripts should ask for JSON.

Every result names its source with a reference, such as note:12, document:3 or msg:telegram/1/77/42. You can pass a reference to other commands: to tag it, write a note about it or use it as evidence. How Memo works explains each kind.

Every JSON answer has the same shape. One thing is under item; a list is under items, with page saying whether more follow (page.has_more). meta names the run and the accounts read, and missing, when present, says what the answer could not cover and why. --json=a,b keeps only those fields of the item or of each item in the list. Dotted paths traverse nested objects and arrays. Use --json=ref,excerpt for compact memo search all hits. Paths are relative to the answer payload; pagination, coverage, next steps, missing data and execution metadata remain. Field selection reduces printed output, not the work scanned.

Where a search answer keeps its references:

content: narrows a search to the text of files, transcript: to voice transcripts and filename: to file names. Compare like with like: memo search messages 'content:invoice' with tg search messages 'content:invoice', a plain word with a plain word.

When you have several Telegram, MAX or mail accounts, Memo does not choose one for you. Commands that work in one account ask for it with --provider and --account, and tell you which accounts exist when it is missing. A search with no account searches all of them and says so.