Your AI forgets. We show you how to work with it reliably anyway.
ChatGPT, Claude or Gemini are brilliant for a quick question. The moment it turns into a matter that runs for weeks, one doctor's report after another, building a house, an inheritance, a dispute with your insurer, the trouble starts. We lived through it ourselves for months and offer you the shortcut.
A question is not a case.
Questions need no memory. Cases need a file. Nobody tells you that when you take out the subscription.
Where is the nearest hospital?
The hospital is 4 km away, here are the directions.
“What do these test results mean?”
Second doctor's report, new chat
Second opinion, new medication
“What did the doctor say after the surgery?”
You explain everything from scratch
With your own file, everything stays put.
Every report goes into a folder on your computer, with a one-page overview on top. The AI reads it before every answer. Week 1 or week 20: it knows what the doctor said.
Sound familiar?
Real mistakes from more than six months of daily work with AI. None invented, none sugar-coated.
Forgets who you are
Your name, how you like to be addressed, your preferences: in the next chat it's all gone again.
Asks about settled things twice
What you sorted out last week comes back as a new question. For the third time.
Loses the thread
Doctor's letter, test results and follow-up question sit in five chats. No chat knows them all.
Doesn't learn from mistakes
It apologised and promised. Next time exactly the same thing happens.
Writes without the backstory
An email to someone you've been writing to for months sounds like it's for a stranger.
Reports “done” without proof
Done, it says. It never checked. You only notice when it's gone wrong.
Guesses instead of asking
If a date or a number is missing, it invents a believable one.
Quotes what was never said
“The doctor said …” – except it was in no report at all.
Claims “checked”
A glance at something similar already counts as proof.
Mixes up versions
It keeps working on the old document while the new one sits untouched.
Deletes large chunks
It was meant to change one sentence and removed half the page in the process.
Sends instead of drafting
The email was only meant to be prepared. It's already gone.
Builds what nobody ordered
New sections, new rules, new tools, unasked and for no reason.
Leaves things lying around
Tasks get collected and noted “for later” instead of being done straight away.
Doesn't think consequences through
“This week in Munich” is still online long after the week is over.
Long chats get expensive
The longer the conversation, the more every single answer costs.
Limit in the middle of work
Just before the finish line, your allowance runs out.
Too many rules
Every annoyance gets a new rule. After a hundred rules it no longer follows any of them.
It writes novels
You wanted a yes or no and get three paragraphs of explanation.
Daily frustration
You spend more time keeping watch than getting work done.
Three simple things solve almost everything.
No coding and no system so big that it confuses you yourself. You don't need technology, you need to manage your AI well, just like a new member of staff.
A job description on one page
Who you are, how you want to be addressed, what must never happen, when to ask. Saved once, it applies in every conversation.
A handover note for each session
At the end, the AI writes down in five lines what is done and what comes next. At the start, it reads the note. Nothing gets lost any more.
Your own file
A folder on your computer with an overview on top. The AI looks things up instead of guessing. Your documents belong to you, even if you switch provider.
How we did it at first
Every annoyance got a new rule, every problem a new program.
How it works
Someone we know has been working without trouble for months: one page about himself, a few rules, short clear instructions.
From chat to file to database.
The longer and bigger your work with AI gets, the more memory it needs. We have climbed all three steps ourselves and know when it's time for the next one.
The chat
For single questions
- Recipe, translation, letter
- No memory needed
- Done after one answer
The file
For cases that run for weeks
- Doctor's reports, building a house, inheritance, insurance
- One folder with an overview on top
- The AI reads the overview before every answer
The database
For everything that keeps growing
- Several projects, months, hundreds of documents
- The AI searches precisely instead of reading everything
- Every fact with a date and a source
- Mistakes are noted once and never repeated
- Lives on your computer, independent of the provider
Why we built a database ourselves
We run several websites with AI. After a few weeks, hundreds of documents, agreements and lessons were scattered across folders. With every task the AI read too much, missed what mattered, asked about things already settled and repeated old mistakes. Every new chat started almost from zero.
Only the database turned that around: at the start the AI reads one short page and looks up the rest precisely. Since then it knows what has been decided and where it has gone wrong before.
Our lesson: the database was right. The only mistake was patching every annoyance with a new rule. That's why we show you the lean version right from the start.
Do you need a database yet?
If you tick two or more boxes, it's time:
- You work on more than one case at a time
- A case has been running for more than a month
- You have more than 30 documents for it
- You regularly explain the same thing to the AI again
- You pay for the AI and keep hitting the limit
One glance and you see everything.
A memory is good. It gets really easy with a cockpit: one page on your computer where you see what your AI knows, what it has done and what comes next. You put it together exactly the way you need it.
Memory
Open tasks
- TodayPrepare questions for the follow-up appointment
- WeekCompare the roofer's quote with the first one
- WeekInsurance appeal, check the draft
- DoneDoctor's letter of 12 Sep filed
My cases
What the AI did today
- 09:12Read the test results of 3 Oct and added them to the overview
- 10:40Compared two quotes, noted a difference of €1,240
- 11:05Saved the email to the insurer as a draft, not sent
- 11:30Wrote the handover note for tomorrow
Repeated mistakes
“Date from memory instead of from the file” has happened for the 2nd time. The AI now always double-checks dates.
Reliability
Spot check: for 9 out of 10 questions the AI found the right place in your file, with a source.
You steer
In the cockpit you hand out tasks, set priorities and see what has been decided. The AI picks up its work there instead of asking all over again every time.
The database reports back
Everything the AI does goes into the memory with a time and proof. In the cockpit you see it at a glance, without reading through long chats.
Build and maintain a website with words alone.
That's how we run our own sites: we tell the AI in ordinary sentences what should change. It builds the page, checks it, makes a backup first and puts it online itself. No website builder, no coding, no manual uploading.
We help you set this up for your own site, including automatic uploading to your server.
Tell us your problem.
Write in your own words what annoys you about the AI. We'll show you straight away which solution fits.
Your text stays in your browser. It is not sent and not stored.
Shall we let you know as soon as it's ready? Optional, used only for that, and afterwards we delete your address.
What you get
Ready-made templates to copy and guides in plain language. Works with ChatGPT, Claude and Gemini.
One solution
For your most important problem, for example “The AI forgets everything” or “The AI makes things up”.
- The matching template to copy
- Where to save it in your AI
- Step-by-step guide with pictures
AI Memory
All six solutions plus building a memory of its own for your AI.
We never see your data. You set everything up yourself, and we show you exactly how.
Questions everyone asks sooner or later
Thousands of people type exactly these questions into search engines and chats every day. Here are the honest answers.
Why does ChatGPT forget what I told it?
Every new chat starts almost empty. The built-in memory only keeps the broad strokes, such as your name or a few preferences, and decides for itself what to keep. Details from earlier conversations, numbers and exact statements get lost. The fix is a fixed instruction, a handover note and a file of your own.
Why doesn't Claude remember earlier chats?
Claude can search old chats and has its own memory in projects, but both need to be switched on and well fed. Without a clear overview, Claude often can't find the right place and answers from general knowledge. A short overview page in the project makes a clear difference.
How do I teach my AI to remember me?
With one page it reads in every conversation: who you are, how you want to be addressed, what has already been decided and what must never happen. This page belongs in the permanent settings or the project instructions, not in the chat.
Can I manage my doctor's reports with ChatGPT, Claude or Gemini?
Yes, if the reports sit in a file of their own and the AI looks things up precisely instead of trying to remember. Scattered across many chats, it goes wrong. Important: the AI organises and explains, it never replaces your doctor.
Why does the AI make up answers?
It is built to always give a fluent answer. If a piece of information is missing, it fills the gap with something plausible. What helps is a file with the real facts and a firm rule to ask when unsure and to name sources.
Why doesn't the AI follow my instructions?
Usually there are too many, too long or contradictory instructions, or they only exist in an old chat. Five clear rules in the right place work better than a hundred scattered ones.
What can I do when I hit the limit or the chat gets too long?
Long chats cost more with every answer because the AI rereads the whole history. Better: have it write a handover note, open a new chat, paste the note in. That saves your allowance and the AI even works more precisely.
How do I stop the AI from deleting things or sending emails without asking?
Asking nicely is not reliable. Block it: emails as drafts only, no delete rights, backups on. We show you which settings do this with the common providers.
Doesn't my AI already have a memory?
Yes, and it's getting better. But it mostly keeps the broad strokes. What a doctor said exactly on 3 March, which decision has already been made and where the AI went wrong before are not reliably stored there. For that you need a file of your own.
Do I need to know how to code?
No. This offer is explicitly for people who simply work in the chat. If you build databases yourself, you don't need us.
Do you get access to my documents?
No. We sell guidance and templates. Everything you set up stays with you.
Is this medical advice?
No. We help you organise your documents and use AI more reliably. Please always discuss medical questions with your doctor.
Who is behind this?
Fiperly, an independent AI development company from Brandenburg, Germany. We work with AI every day, have fallen into every trap along the way and written down what really helps.
ChatGPT, Claude and Gemini are trademarks of their respective owners. Fiperly is independent and not affiliated with any of these providers.
Fiperly
Mutify
KIPODE
Job & Stay
Birthday Paper
Mia Amare
HealthMetricsLab
WoSiHo