Skip to content

Memories: your AI keeps what it figured out

Save what your AI worked out once, and it follows the same steps next time instead of starting from zero.

What a memory is

A memory is something your AI worked out that you never want it to work out again. When it finally gets a job right, it saves how: the steps and commands, how to check the result, and what not to do. Next time you ask, it follows those steps instead of guessing.

Example, for illustration: one saved memory

Validate the invoice parser

billing-apiv1 in use
Steps
npm run fixtures -- --schema {schema}, then npm test parser
Check
Each invoice total equals the sum of its line items
Don't
Edit snapshots by hand to make a test pass

Run it as /mcp__fmp__validate-invoice-parser

How it works

You decide what is worth keeping. FixMyPrompt keeps it in a shape your AI can follow later.

Tell it to save once it works

When the AI finally gets something right, say "save how you fixed that to memory". The secret check runs first, then it saves the steps, the commands, how to check the result and what to avoid, not just the prompt you typed.

  • Filed for you. Memories go into a project named after the folder you are working in.Pro
  • Secret check first. Keys and passwords are caught before anything is saved.
  • Versions, with one in use. Save a better way and it becomes a new version. Switch back any time.

Ask for it in plain words

"Use fmp to find how we deploy the staging site." Ask in plain words in any connected tool and your AI reads back the memory that fits. You can also open the Memories page in the app.

  • Search by meaning. Finds the memory even when you word it differently.Pro. The free plan searches exact words.
  • Similar memories. Spot two memories that cover the same job.Pro
  • Projects kept apart. Each working folder has its own memories, so one repo's steps don't leak into another.Pro

Run it like a command

Your AI follows the saved steps. Give a memory a short name and it shows up as a slash command in Claude Code. Anything that changes between runs is a blank it asks you to fill each time.

  • Fill-in blanks. Branch, environment, file name: asked at run time.
  • Tighten. Trims a long saved prompt down to what the AI needs. Works on saved prompts, not saved steps.
  • Merge and undo. Fold two memories into one on the Memories page, and undo it if you change your mind.Pro

A week with and without it

Example, for illustration

Without memories

  • Each new session works out the same fix from scratch.
  • The mistake you already ruled out comes back.
  • A teammate spends an afternoon on work you finished last week.

With memories

  • The AI finds the saved steps and follows them.
  • It checks its work the way the memory says.
  • It avoids the mistake the memory warns about.
  • Your teammate asks for it and gets the same steps.

It gets better as you use it

Memories are not frozen. Each run tells you whether they still hold up.

  • It reports back. After following a memory, your AI says whether the steps worked.Pro
  • Failures are marked. A memory whose steps failed is marked, so you know to fix it.Pro
  • Versions, with one in use. A better way becomes a new version, marked as the one in use. Undo goes back to an earlier one.Pro has undo
  • Merge duplicates. Two memories that cover the same job become one.Pro
  • Tighten. Trims a long saved prompt down to what the AI needs, with the token count before and after.

Where it works

Connect your AI tool once, then sign in. Each connected tool finds the same memories.

Claude Code

Add FixMyPrompt once and sign in. Your memories become tools and slash commands.

$ claude mcp add --transport http fmp <your link>

Cursor

Add FixMyPrompt to the MCP servers in Cursor's settings and sign in. Your memories show up as tools.

.cursor/mcp.json: "fmp": { "url": "<your link>" }

Your link is in Settings in the app. It connects over MCP.

Limits by plan

How much each plan keeps.

Free
1 project, 500 memories. Kept 60 days after you last use them. Upload 1 document of up to 10 pages.
Pro
5 app projects plus folder projects, 5,000 memories. Kept 6 months after you last use them. 20,000 credits every month for prompt checks and Tighten. 7 days free, then $19 a month. Monthly credits start with your first paid month. They pause while your own key is on. Upload 10 documents of up to 200 pages each.
Biz
Per seat: 10 app projects plus folder projects, 10,000 memories, pooled across the company. Kept 12 months after you last use them. 20,000 credits a month per seat, shared by the company. 7 days free, then $25 a seat a month. Upload 50 documents a seat, pooled across the company, of up to 200 pages each.

Questions

Is a memory just a saved prompt?

No. A prompt is what you asked. A memory is what the AI learned while answering: the working steps, how to check it worked and what to avoid, so the next session can follow them instead of guessing again.

Does my AI look things up on its own?

Not yet. Today you ask for one ("use fmp to find how we deploy staging"), or run it as a slash command in Claude Code when it has a short name.

What if I hit the limit?

Saving stops at your plan's limit, and nothing is deleted. You can still open everything. You get a heads-up once you have used 80% of the limit. To save more, upgrade in Settings > Billing and plan.

Which AI tools work with it?

Claude Code and Cursor, connected over MCP. Connecting a tool is part of Pro. On the free plan you can save and read memories in the app.

What about passwords and keys?

Every memory is checked for keys, passwords and other secrets before it is saved. A save that holds one is stopped and tells you which field to fix.

Keep what your AI works out

Start free trialSee plans