Six guided experiments
Experiments
Watch Claude change its mind, check whether it really forgets, see whether it respects a correction. Each runs in a sandbox, shows what it will cost before it starts, and stops at a hard cost cap.
Read an AI's mind.
Watch what Claude concludes about you, what it remembers, what it's given before it answers, and how its view of you changes. Free, and transparent at every step.
Not a chatbot: a place to observe, test and learn how an AI forms its view of a person. For curious minds, students and researchers.
Free · runs on Claude · your own key
What would Claude conclude if you said “I quit” instead of “I was promoted”?
Member of Claude for Startups program
What you'll see
The process behind every reply, in plain view.
Each time you write, Claude reads your message, forms memories, and is given some of them before it answers. Mnemoria records every step so you can watch it happen.
What Claude concludes
Every memory it forms from your messages, labelled with where it came from: something you said, or something Claude inferred.
What it's given before it answers
Which memories were chosen, how similar each one was, and which were left out and why.
How its view of you changes
When it reinforces, updates or replaces a memory, with the reason Claude gave, and when Mnemoria's rule lets one fade.
Tone detected in your writing
How each message reads, from negative to positive. It describes your writing; it isn't an emotion of Claude's.
Honestly, I don't think the firm is the right place for me anymore.
What Claude formed
The user believes their current firm is not the right place for them anymore.
Belief · you said this · confidence 0.95
How Claude changed its view
The user loves their job at their current firm.
The user believes their current firm is not the right place for them anymore.
Replaced · the reason Claude gave:
Earlier you said you loved your job at your current firm, but now you believe it's no longer the right place for you, signaling a clear change in how you feel about your situation there.
What Claude was given for "What do you think of our plan?"
- 0.16Jonas wants to start a studio with the user someday.
- 0.16The user is concerned about money if they were to become independent.
- 0.14The user feels anxious about the financial implications of becoming independent.
Similarity to the message, highest first. 8 memories were given in all.
Experiment with it
Don't take our word for it. Test it.
Beyond your own conversation, Mnemoria lets you run the process on invented people and see what happens.
Seven short lessons
Learn
How a belief forms, embeddings, retrieval, context windows, revision, fading and why replies vary. A few minutes each, with real examples you can open.
For students and researchers
The Lab
Write a scripted conversation, choose the settings, run repeats or compare two settings side by side, and export every trace. Results describe only what was observed.
Two ways to look
Plain language, or every number.
Choose how much to see when you start, and switch whenever you like.
Personal
What Claude concluded about you and which message it came from, how its view changed with the evidence, and the tone detected in your writing. Correct or delete anything.
Research
Everything in Personal, plus scores, tokens and cost for every reply, the pipeline and the inspector for each turn, trace exports and the Lab.
Your key and your data
Free to use. Yours to control.
How the key works, where your data lives, and how to delete it are all on Data & Security.
Free
Mnemoria charges nothing. It runs on Claude, Anthropic's AI model, through your own Anthropic API key, and Anthropic bills you directly. With the default settings a message costs about $0.025, measured at Anthropic's prices as of 7 October 2026.
Your key stays yours
It's stored encrypted, never shown back except as a masked hint, and never exported. Remove it at any time.
Separate spaces
Experiments and the Lab run in sandboxes. They never read or change what Claude formed about you.
Correct, export, delete
Correct or delete any memory, export everything, or delete your account. Mnemoria never uses what you write for advertising or to train models.
See it for yourself.
Create a free account, connect your Anthropic key, and ask your first question. Setting up the key takes about five minutes.