Teach your bot to remember what works, so it stops repeating mistakes
A bot that forgets everything between chats makes the same mistake twice. Give it a notes file it reads at the start and updates at the end, and it gets better every week. Anthropic published how it does this for Claude.
Seen on Hacker News ×2 · Official docs
What the bot does
- Starts each job by reading a short notes file of rules and facts it has already checked.
- Does the job, and when you correct it, writes the fix down as a one-line rule.
- Now and then tidies the notes: merges duplicates and marks what is verified and what is still a guess.
- Shows you the notes so you can edit or delete anything.
What you end up with
A bot (or Claude project) that carries lessons from one session to the next. You can open the notes any time and see what it believes.
Copy-paste prompt
Fill in the [brackets]. Then paste it into any tool below.
At the start of every job, read my notes file called [bot-notes] if it exists. When I correct you or a result turns out wrong, add one short line to the notes: what went wrong and the rule to follow next time. Mark each line 'checked' or 'guess'. At the end of each job, show me any new lines you added and ask before saving anything that includes personal or private details.
New to this? Start with what an AI agent is, in simple terms, then how to build an AI agent without coding, or set up a bot in 5 plain steps →
Want one new job like this each week, with a prompt to copy? Get the free BotsDoThat Weekly email and the free Starter Pack: 7 ready-to-paste AI agent setups, plus a power-user bonus.
Setup
- Pick where the notes live: a Project in Claude with a notes document, or a note/file your bot can read and edit.
- In Claude, open Settings and check that memory and 'Search and reference chats' are on (or off, if you prefer privacy). Use an incognito chat for anything you do not want remembered.
- Paste the prompt above as the bot's standing instructions.
- After a week, open the notes and delete anything wrong or too personal.
Tools that can do it today
- chatClaudeMemory and past-chat search are in Settings; paid plans can search past chatsFree plan; paid plans add more
- agentGrok Bot (personal agents)Included with Premium, Super and Ultra on The X Subscription; or link SuperGrok, SuperGrok Plus, SuperGrok Heavy or X Premium+; a Grok Business seat; or a legacy paid Cursor plan
Plans and limits change; check each tool's page.
Watch out
- A bad note is remembered as confidently as a good one. Anthropic's post says memory is a prediction about what will matter again, so review the notes now and then.
- If you let the bot tidy or rewrite its own notes (Anthropic calls this 'dreaming'), keep a dated copy of the notes file first, and ask it to list what it changed. A bad cleanup is then easy to spot and undo.
- Do not let the bot save passwords, account numbers or other people's private details into notes.
- Memory features differ by plan and app; check the Claude Help Center article for what your plan includes.
What this costs to run
In a chat app it runs inside your plan. If you run it as a bot on an AI API, the bill depends on how often it runs and how much it reads.
Estimate it in the cost calculator →
Tip: Use a small AI model for simple jobs, a big one for hard ones. More in Run your bots for less and how much an AI agent costs.
Questions people ask
Which AI tools can do this?
Claude or Grok Bot (personal agents) can do this today. Paste the copy-paste prompt on this page into any of them and fill in the [brackets].
How hard is it to set up?
It's rated Easy. Start with the steps under “What the bot does”.
What should I watch out for?
A bad note is remembered as confidently as a good one. Anthropic's post says memory is a prediction about what will matter again, so review the notes now and then.
Sources
- officialAnthropic Engineering (2026-11-06): The components of agent memory, how Claude keeps notes between sessions and 'dreams' to tidy them
- Hacker NewsHacker News: The Components of Agent Memory (189 points / 105 comments, measured 2026-11-19) 189 points · 105 comments · measured Nov 19, 2026
- Hacker NewsHacker News: Agent memory needs its OpenTelemetry moment (191 points / 111 comments, measured 2026-11-19) 191 points · 111 comments · measured Nov 19, 2026
- articleLoreKit blog (2026-11-07): every agent tool keeps memory in its own format
- officialClaude Help Center: Use Claude's chat search and memory to build on previous context
Last checked Nov 17, 2026. Self-reported results from posts are the poster's own claims.