Stop rewriting the same prompt.
Configure the assistant once.
Studio gives every conversation its own instructions, its own goal and your own documents — across 300+ models from Anthropic, OpenAI, Google and DeepSeek. Bring your OpenRouter key or pay by the token. You see what each message costs while it is still typing.
$0.50 of credit on sign-up. Or connect your own key and Studio never touches your balance.
300+ models · one balance · switch mid-conversation
Anthropic · OpenAI · Google · DeepSeek · Meta · Mistral · xAI
You have had all three of these arguments.
“Every answer has the same shape.”
Three bullet points, a hedge, and a closing offer to help further. You wanted a straight answer.
“It folds the moment I push back.”
Agreement is not accuracy. Sometimes you need something that holds its position — and sometimes you need it to hold yours.
“I’m paying monthly for a model I barely use.”
Meanwhile the credit you already bought sits unspent, because nothing lets you bring it.
Studio is what happens when you stop working around all three.
Instructions
Set the mind, not just the message
Give a chat a persona — Straight answers, Devil’s advocate, Analyst, Writing partner, or one you write yourself — and a goal it keeps in view. It maintains a running plan and tells you the next step toward it.
Save any prompt as a persona and reach for it from the composer. Applying one copies its text into that chat, so editing a persona later never rewrites a conversation you already had.
Built in — and you can write your own
Straight answers
Conclusion first, no preamble, no hedging.
Devil's advocate
Argues the strongest counter-case, not a strawman.
Analyst
Quantifies claims and surfaces its own weak points.
Writing partner
Matches your voice instead of imposing its own.
Code reviewer
Correctness first, then design, then style.
Socratic tutor
Leads with questions instead of answers.
Knowledge
Teach it what you actually know
Upload the things you have worked out — your notes on a subject, your style guide, your architecture decisions, your research. Studio indexes them and every model in the catalogue argues from them.
Keep a document to one conversation, or make it apply everywhere. Text is extracted and indexed; the original file is not stored.
Personal
Your notes on how you think, so advice comes from your framework rather than a generic one.
Development
Your conventions and goals, so code arrives in your style on the first try.
Research
The papers and reports you are working from, cited rather than half-remembered.
Cost
The bill, in real time
Every message carries its exact cost. Every day, model and project rolls up on one page. No credit bundles, no token estimates, no surprise at the end of the month.
Studio reads the real figure back from the provider and writes it to an append-only ledger you can audit. Set a daily ceiling and it stops before you do.
This month
$2.47
- Claude Sonnet84 turns$1.94
- GPT-5 mini212 turns$0.38
- Gemini Flash146 turns$0.11
- DeepSeek31 turns$0.04
Your key
Your key. Your provider. Your invoice.
Connect an OpenRouter key and spend the credit you already bought. Connect a vendor key and Studio routes straight to the provider. Keys are encrypted at rest and never sent back to your browser.
- 1
Your vendor key
An openai/* model goes straight to OpenAI. No gateway in the middle, no margin.
- 2
Your OpenRouter key
Covers the whole catalogue, and spends credit you have already bought.
- 3
Studio credit
Only when you have connected nothing. Removing a key makes chats cheaper, never broken.
Studio Code
And when the work is code
Studio Code gives each project a real container. Plan mode literally cannot write a file — it is handed a different set of tools, not a different instruction. Switch to Build and choose how much it may do unattended: everything, ask first, or read-only.
Watch the terminal live, read the diff before it commits, preview the dev server in the page. Use one model to plan, another to write, a third to review the diff — and see what each of them cost.
How Studio Code worksEverything else that is already in there
Branch any message
Fork a conversation at any point and take it somewhere else. The original stays exactly where it was.
Runs Python
A sandbox with no network access, and a workspace that survives between turns.
Builds documents
Word, Excel, PowerPoint and PDF, produced in the sandbox and handed back as a download.
Generates images inline
Ask for an explanation and a diagram, and get both in one answer.
Notes that persist
Scratch space that outlives the chat it was written in.
Installs on your phone
A real PWA, built for one hand. Not a website in a shell.
Automations via n8n
Let the model send an email or write a sheet, with credentials that never enter Studio.
Daily spend caps
A ceiling per day, enforced before the request goes out rather than after the bill arrives.
Self-host it
One VPS, one compose file. The whole thing runs on your own box if you would rather.
Three ways to pay. No subscription in any of them.
Free
$0.50 credit
Standard models, 30 messages an hour, a $1 daily ceiling. No card. Enough to decide.
Pay as you go
Tokens + 25%
Top up any amount. Every model, images, code execution, Studio Code. That 25% is the whole business model, and it is on this page rather than in a footnote.
Your own key
$0 to Studio
Connect OpenRouter or a vendor key and the model bills you directly. Studio takes nothing on those turns.
What happens to your data
Keys are encrypted at rest
Decrypted only inside the request that uses them, and never returned to your browser.
Your conversations are not training data
Studio does not train on them and does not sell them.
Code runs in an isolated container
Unprivileged, capability-dropped, and firewalled off the internal network.
You can take it with you
Or run the whole thing yourself — it is one compose file on one box.
The detail is on the privacy page, including every third party a message passes through.