Questions
The short answers to what people ask first. If yours isn’t here, write to hello@safata.ai.
What do I need?
A Mac with Apple silicon on macOS 13 or later, and an API key from one model vendor. No account with me, and nothing else to install.
Can I use my ChatGPT, Claude or Gemini subscription?
No. Those are consumer subscriptions, and no vendor lets another app sign in with one. Safata uses an API key: a pay-as-you-go account at the same company, billed for what you use. Same models, a different door.
Where do I get a key?
Pick one vendor; you can add others later. Sign up, add a payment method, create a key, and paste it into Safata when it asks. It takes about two minutes, and the app walks you through the same steps.
- OpenAIplatform.openai.com/api-keys
- Anthropicplatform.claude.com/settings/keys
- Googleaistudio.google.com/apikey
- Mistralconsole.mistral.ai/api-keys
- xAIconsole.x.ai
- DeepSeekplatform.deepseek.com/api_keys
- Moonshotplatform.kimi.ai/console/api-keys
- Z.aiz.ai/manage-apikey/apikey-list
- OpenRouteropenrouter.ai/keys one key, hundreds of models
What does a job cost?
What your vendor charges for the work, usually cents and sometimes a few dollars. The model matters most: the most expensive cost a hundred times the cheapest. Every session has a spend limit, and stops when it reaches it.
What leaves my Mac?
What the model needs for the job: your messages and what it reads from your files, sent to the vendor you chose and no one else. Your files, sessions and memory stay on your disk, and Safata lists every request it sent. The one other request is the update check: at launch and every four hours Safata asks GitHub whether a newer build exists, sending nothing about you.
Is there telemetry?
None: no analytics, no crash reports, no account. The price is that I only learn what breaks when you tell me.
Will my work be used to train models?
Not by me; I never see it. What a vendor does with API traffic is set by its terms, and the major ones do not train on it by default.
Does it always get it right?
No. In the cash-flow example on the home page, a smaller model got the deposit question wrong; the larger one got it right. That is why the workbooks keep their formulas live and the memos keep their sources, and why a second opinion is one more session on another model.
Which model should I start with?
The vendor you already have an account with. If you have none, the frontier models from Anthropic, OpenAI and Google are the safest choice for hard work, and DeepSeek is the one to know when the volume is high and the budget is not. Switching later is a menu, not a migration.
Do I need to know how to code?
No. When code is the right tool, for parsing a file or cleaning a dataset, Safata writes it, runs it and hands you the result. It is not a coding agent.
Can it read my Google Drive or Dropbox?
Yes. Google Drive, Dropbox and Box connect read-only, and the finished file still lands on your disk. Anything with an MCP server plugs in too.
Is it free?
Yes, for individual personal use. Company use is paid. To deploy Safata at your company, write to hello@safata.ai. You pay your model provider separately for API usage, with no markup from Safata.
Windows? Intel Macs?
Macs with Apple silicon today, and Windows later. There is no build for Intel Macs.
How does it update?
On its own. At launch and every four hours Safata asks GitHub whether a newer build exists, downloads it in the background and installs it when you quit. Settings › About turns that off.
Can another agent use it?
Yes. Claude Code, Claude Desktop, Codex and the ChatGPT desktop app can hand Safata a job over MCP and collect the files it makes. How to set it up.
Can I read the source code?
No. Safata ships as a signed app and the source is not published. What you can check from outside: there is no account, no server and no telemetry, and every request is listed.