Give your AI assistant a long-term memory
Every new chat starts from zero: the assistant forgets who you are, what your company does, and how you like your answers. You want it to remember — across conversations.
You need
- · An AI app that supports skills: Claude Desktop, Claude Code, Cursor, VS Code — don't have one yet?
- · The Memory skill (no API key)
- · About 4 minutes
Step by step
- 1Install the Memory skill
Use the install widget below. No API key needed — memories are stored in a local file on your machine.
- 2Restart and teach it
Fully restart the app, then tell the assistant what to remember using the prompt below as a template.
- 3Test it in a NEW chat
Open a fresh conversation and ask: "What do you remember about me?" — the answer should include what you taught it.
Install the Memory skill
Open Settings → Developer → Edit Config in Claude Desktop (this opens claude_desktop_config.json) and merge this block, then fully restart the app:
{
"mcpServers": {
"memory": {
"command": "npx",
"args": [
"-y",
"@modelcontextprotocol/server-memory"
]
}
}
}The prompt
Remember the following about me permanently: I run a small consulting company called <YOUR COMPANY>, my main customers are <YOUR CUSTOMERS>, and I prefer short, direct answers. Store this in your memory so you can use it in future conversations.
What you should see
The assistant confirms it stored the facts. In any new chat, asking "what do you remember about me?" returns your details — and answers start matching your preferences without re-explaining.
Last reviewed 2026-06-10
If it doesn't work
Run a website? Make sure agents can read it.
Assistants with skills like these are already visiting websites on their owners' behalf. agentics checks free of charge whether yours is readable, findable, and trusted by AI agents.
Run free check →More recipes
You want your AI assistant to actually open websites — compare prices, check availability, read pages that need a real browser — instead of telling you it cannot browse.
Your documents, invoices, and contracts live in folders on your machine — and you want to ask questions about them ("find the contract that mentions X", "sum up these invoices") without uploading anything anywhere.
Your assistant answers from training data that is months old. You want it to search the web right now — news, prices, fresh documentation — and answer with sources.