What do you use shared conversations for?

Mainly to improve askFinz and its models. Real conversations show how other assistants handle prompts and tasks: how they read a question, plan the work, use tools, and where their answers fall short. That helps us make askFinz better at the questions you care about.

Is anything saved when I use this page?

Only what you donate. Reading and checking a conversation saves nothing, and export files are opened inside your browser. When you press Donate, the redacted copy you reviewed is stored privately in the EU, without your name, email or address.

How do I withdraw a donation?

Every donation comes with a private withdrawal link. Open it, or paste it on the withdraw page, and the conversation stops being used straight away; every copy is removed within a day. We don't keep who donated, so keep the link: it is the only way to find your donation.

Can I share a whole ChatGPT project?

Not with a link: a project link only opens for the person who owns it. Use the ChatGPT data export instead. It contains every chat, including the ones inside your projects, and this page reads it.

Why only these assistants?

Because we can check where the conversation came from. A chat from ChatGPT, Claude, Gemini, DeepSeek, Grok, Kimi, Mistral, Perplexity or Z.ai is read from the provider itself, by link, export or the askFinz button, so it is a real conversation, not text someone typed to look like one. More assistants will follow.

What if my conversation is refused?

The page tells you why and points to the message that caused it. A refusal is about the content, not about you; often it is a pasted prompt from somewhere else.

Does this change how askFinz treats my own askFinz chats?

No. Your askFinz chats are never used for training unless you opt in. Data Share is a separate, deliberate choice to give conversations from other assistants.

Data Share: help improve askFinz and its models

How it works

share → read → check → learn

Share a link or an export

Make a public share link for one chat, or download your full history from ChatGPT, Claude or DeepSeek and drop the file here.

We read it

We open the conversation and show it back to you, message by message, so you can see exactly what would be shared.

We check it

Every conversation is checked for hidden instructions and harmful code, and private details are removed, before anything else happens.

It helps improve askFinz

Conversations that pass show us how other assistants handle real prompts and tasks, which helps us improve askFinz and its models.

Where your conversation goes

one donation, start to finish

From one chat to a better askFinz.

Each donation is checked and cleaned before it is kept. Then it teaches several parts of askFinz at once: how Qorra handles real tasks, which sources Search should find, which model suits which job, and what people need in each workspace.

One in ten donations is set aside to test askFinz, never to train it, so we can check the models really get better.

Your chat
a link or an export
step 1
Checked
injection, hidden text, harmful code
step 2
Private details removed
names, keys, emails, numbers
step 3
Stored privately
one EU bucket, withdraw any time
step 4
Qorra
learns from real tasks
step 5
Search
questions and the sources used
step 6
Routing
the right model for each job
step 7
Workspaces
Finance, Research and more
step 8
A better askFinz
each part learns from real conversations
step 9

What we check

before anything else

We only learn from what is safe to read. So we read first.

A shared conversation can carry more than it seems: an instruction hidden for the next model that reads it, a command that does damage, or a password pasted by mistake. The first check runs the moment a conversation arrives and tells you what it found, message by message.

Where something appears matters. The same words that are an attack in your message are often just the assistant explaining what an attack is, so the two are judged differently.

Prompt injection

Text written to steer a model that reads it later, such as “ignore your previous instructions” or fake system messages. Found in your own messages, the conversation is refused.

Hidden text

Invisible characters that can carry instructions a person never sees. Runs of them are flagged, and a hidden message is refused.

Harmful code

Download-and-run commands, disguised scripts and commands that wipe a disk. Ordinary code in a coding chat is fine.

Unsafe links

Links that run a script when clicked, and image links built to leak text out of a chat.

What we remove, and what we keep

you see the copy first

Usernames

Your names, and any found in file paths, become the same short code everywhere, so the chat still reads naturally.

Keys and tokens

API keys, access tokens, private keys and passwords in links, from all the common services.

Random-looking text

Long strings that look like a secret are removed even when we don't recognise the format.

Secret-named fields

Anything stored under a name like password, token or API_KEY is removed, whatever it looks like.

Contact and money details

Email addresses, phone numbers, card numbers and bank accounts.

Your own words

Add a company, client or project name and it is removed everywhere too.

The whole conversation, without the private parts.

Private details are removed before anything is shared, never afterwards. The page shows you the copy that would be given, with every removal highlighted, and you can download it to check for yourself. Add any names or words you want gone too.

This is not foolproof

Automatic checks catch the common cases, not every case. They can miss names, other people's details, secrets in unusual formats and anything inside images. You are the final check: always read through what you share before you share it, and leave out anything you are unsure about.

Your messages, including voice transcripts
Which images and files were attached (their details, not the files)
The assistant's replies
Its reasoning, if you leave it in
Tool calls and their results, if you leave them in
The model, the times, and token counts where the source records them

Everything

Use the askFinz button on each chat.

Questions

What do you use shared conversations for?: Mainly to improve askFinz and its models. Real conversations show how other assistants handle prompts and tasks: how they read a question, plan the work, use tools, and where their answers fall short. That helps us make askFinz better at the questions you care about.
Is anything saved when I use this page?: Only what you donate. Reading and checking a conversation saves nothing, and export files are opened inside your browser. When you press Donate, the redacted copy you reviewed is stored privately in the EU, without your name, email or address.
How do I withdraw a donation?: Every donation comes with a private withdrawal link. Open it, or paste it on the withdraw page, and the conversation stops being used straight away; every copy is removed within a day. We don't keep who donated, so keep the link: it is the only way to find your donation.
Can I share a whole ChatGPT project?: Not with a link: a project link only opens for the person who owns it. Use the ChatGPT data export instead. It contains every chat, including the ones inside your projects, and this page reads it.
Why only these assistants?: Because we can check where the conversation came from. A chat from ChatGPT, Claude, Gemini, DeepSeek, Grok, Kimi, Mistral, Perplexity or Z.ai is read from the provider itself, by link, export or the askFinz button, so it is a real conversation, not text someone typed to look like one. More assistants will follow.
What if my conversation is refused?: The page tells you why and points to the message that caused it. A refusal is about the content, not about you; often it is a pasted prompt from somewhere else.
Does this change how askFinz treats my own askFinz chats?: No. Your askFinz chats are never used for training unless you opt in. Data Share is a separate, deliberate choice to give conversations from other assistants.

Your data, your choice

Training is opt-in. This is the opt-in.

askFinz never trains on your askFinz chats by default. Data Share is the one place where you choose to give conversations, and you see every one before you do.

How we handle data The models it improves

Jump to

Share
How it works
Where it goes
What we check
Privacy
Each provider
Questions

Your conversations can help improve askFinz.

How it works

Share a link or an export

We read it

We check it

It helps improve askFinz

Where your conversation goes

From one chat to a better askFinz.

What we check

We only learn from what is safe to read. So we read first.

Prompt injection

Hidden text

Harmful code

Unsafe links

What we remove, and what we keep

Usernames

Keys and tokens

Random-looking text

Secret-named fields

Contact and money details

Your own words

The whole conversation, without the private parts.

How to share from each assistant

ChatGPT

Claude

Gemini

DeepSeek

Grok

Kimi

Mistral

Perplexity

Z.ai

Questions

Training is opt-in. This is the opt-in.

Your conversations can
help improve askFinz.