the model that thinks · your memory stays the same
A brain is the model that thinks. Your memory stays the same on every brain, so you can switch whenever you like. Each guide below gets you a key in a few minutes, and the same guide is inside the app under /model → Add a brain…
Anthropic (Claude)
Works in the app today.
An email address and a payment card. About 5 minutes.
Keys it needs:Anthropic API keystarts with sk-ant-
Open the Claude Console and sign up (or sign in). Under Billing, add a payment card and buy a small amount of credit; $5 is plenty to start. A Claude.ai chat subscription is separate and does not include API use.
I use keys daily
platform.claude.com → sign in → Billing → add credit. API billing is prepaid and separate from a Claude.ai plan.
How and why it works
The app talks to Anthropic's API directly with your key, so Anthropic bills you per token. We never see the bill or the prompts.
In the Aedelgard app, type /model, choose Add a brain…, then Anthropic (or open this guide inside the app). Paste the key into the field on this step, press Save, then Test.
I use keys daily
/model → Add a brain… → Anthropic → paste → Save → Test.
How and why it works
The field sends the key to the app on your own machine (127.0.0.1), never into the chat: anything typed into a conversation reaches the model provider and may sit in their logs. The key is stored in the app's encrypted keyring.
4 The app checks the key
Your mind
New to this
Test makes the smallest call Anthropic allows and shows ✓ Working, ⚠ Failed (with the reason) or … Could not reach (you are offline, the key may be fine). A failed key stays flagged ⚠ in /model and /drone until you fix or remove it.
I use keys daily
One minimal request with your key, from your machine. Result is kept per key; saving a new key clears it.
How and why it works
The result is shown to you; the key never is. Only the provider sees it, exactly as when any other program uses your key.
5 Choose the brain
You
New to this
Type /model, pick Anthropic, then a model. The mind keeps all its memory; only the thinking moves.
I use keys daily
/model → Anthropic → model. Or let /drone pick a cheap one for coding work.
How and why it works
Your memory lives in the app on your machine and is the same on every brain. Switching brains changes who thinks, not who remembers.
OpenAI (GPT)
Works in the app today.
An email address and a payment card. About 5 minutes.
Open the OpenAI Platform and sign up (or sign in). Under Billing, add a payment card and some credit. A ChatGPT subscription is separate and does not include API use.
I use keys daily
platform.openai.com → Settings → Billing → add credit. Separate from ChatGPT Plus.
How and why it works
The app calls OpenAI's API directly with your key; OpenAI bills you per token.
In the Aedelgard app, type /model, choose Add a brain…, then OpenAI (or open this guide inside the app). Paste the key into the field on this step, press Save, then Test.
I use keys daily
/model → Add a brain… → OpenAI → paste → Save → Test.
How and why it works
The field sends the key to the app on your own machine (127.0.0.1), never into the chat: anything typed into a conversation reaches the model provider and may sit in their logs. The key is stored in the app's encrypted keyring.
4 The app checks the key
Your mind
New to this
Test makes the smallest call OpenAI allows and shows ✓ Working, ⚠ Failed (with the reason) or … Could not reach (you are offline, the key may be fine). A failed key stays flagged ⚠ in /model and /drone until you fix or remove it.
I use keys daily
One minimal request with your key, from your machine. Result is kept per key; saving a new key clears it.
How and why it works
The result is shown to you; the key never is. Only the provider sees it, exactly as when any other program uses your key.
5 Choose the brain
You
New to this
Type /model, pick OpenAI, then a model. The mind keeps all its memory; only the thinking moves.
I use keys daily
/model → OpenAI → model. Or let /drone pick a cheap one for coding work.
How and why it works
Your memory lives in the app on your machine and is the same on every brain. Switching brains changes who thinks, not who remembers.
Open Google AI Studio, sign in with your Google account, click Create API key, and copy it.
I use keys daily
aistudio.google.com → Get API key → Create API key → copy.
How and why it works
AI Studio has a free tier with limits. Read Google's terms for it: on unpaid use, Google may use your prompts to improve its products. Turning on billing for the project moves you to paid use.
In the Aedelgard app, type /model, choose Add a brain…, then Google (or open this guide inside the app). Paste the key into the field on this step, press Save, then Test.
I use keys daily
/model → Add a brain… → Google → paste → Save → Test.
How and why it works
The field sends the key to the app on your own machine (127.0.0.1), never into the chat: anything typed into a conversation reaches the model provider and may sit in their logs. The key is stored in the app's encrypted keyring.
3 The app checks the key
Your mind
New to this
Test makes the smallest call Google allows and shows ✓ Working, ⚠ Failed (with the reason) or … Could not reach (you are offline, the key may be fine). A failed key stays flagged ⚠ in /model and /drone until you fix or remove it.
I use keys daily
One minimal request with your key, from your machine. Result is kept per key; saving a new key clears it.
How and why it works
The result is shown to you; the key never is. Only the provider sees it, exactly as when any other program uses your key.
4 Choose the brain
You
New to this
Type /model, pick Google, then a model. The mind keeps all its memory; only the thinking moves.
I use keys daily
/model → Google → model. Or let /drone pick a cheap one for coding work.
How and why it works
Your memory lives in the app on your machine and is the same on every brain. Switching brains changes who thinks, not who remembers.
Mistral (EU)
Works in the app today. Hosted in the EU.
An email address; a payment card for paid use. About 5 minutes.
Keys it needs:Mistral API keya 32-character string
In the Aedelgard app, type /model, choose Add a brain…, then Mistral (or open this guide inside the app). Paste the key into the field on this step, press Save, then Test.
I use keys daily
/model → Add a brain… → Mistral → paste → Save → Test.
How and why it works
The field sends the key to the app on your own machine (127.0.0.1), never into the chat: anything typed into a conversation reaches the model provider and may sit in their logs. The key is stored in the app's encrypted keyring.
4 The app checks the key
Your mind
New to this
Test makes the smallest call Mistral allows and shows ✓ Working, ⚠ Failed (with the reason) or … Could not reach (you are offline, the key may be fine). A failed key stays flagged ⚠ in /model and /drone until you fix or remove it.
I use keys daily
One minimal request with your key, from your machine. Result is kept per key; saving a new key clears it.
How and why it works
The result is shown to you; the key never is. Only the provider sees it, exactly as when any other program uses your key.
5 Choose the brain
You
New to this
Type /model, pick Mistral, then a model. The mind keeps all its memory; only the thinking moves.
I use keys daily
/model → Mistral → model. Or let /drone pick a cheap one for coding work.
How and why it works
Your memory lives in the app on your machine and is the same on every brain. Switching brains changes who thinks, not who remembers.
Berget AI (Sweden)
Works in the app today. Hosted in Stockholm; your prompts stay in the EU.
Berget runs open models (Llama, Gemma, Qwen, Mistral and others) on its own servers in Stockholm and bills in euros. Its API speaks the same language as OpenAI's, which is how the app talks to it.
In the Aedelgard app, type /model, choose Add a brain…, then Berget (or open this guide inside the app). Paste the key into the field on this step, press Save, then Test.
I use keys daily
/model → Add a brain… → Berget → paste → Save → Test.
How and why it works
The field sends the key to the app on your own machine (127.0.0.1), never into the chat: anything typed into a conversation reaches the model provider and may sit in their logs. The key is stored in the app's encrypted keyring.
4 The app checks the key
Your mind
New to this
Test makes the smallest call Berget allows and shows ✓ Working, ⚠ Failed (with the reason) or … Could not reach (you are offline, the key may be fine). A failed key stays flagged ⚠ in /model and /drone until you fix or remove it.
I use keys daily
One minimal request with your key, from your machine. Result is kept per key; saving a new key clears it.
How and why it works
The result is shown to you; the key never is. Only the provider sees it, exactly as when any other program uses your key.
5 Choose the brain
You
New to this
Type /model, pick Berget, then a model. The mind keeps all its memory; only the thinking moves.
I use keys daily
/model → Berget → model. Or let /drone pick a cheap one for coding work.
How and why it works
Your memory lives in the app on your machine and is the same on every brain. Switching brains changes who thinks, not who remembers.
Logos are trademarks of their owners, shown so you can recognise the right page.
Brains are one kind of extension. See Extensions for what else your mind can learn to do.
Aedelgard
Anthropic (Claude)
OpenAI (GPT)
Google (Gemini)
Mistral (EU)
Berget AI (Sweden)