Research, analysis, documents. Safata is a desktop and command-line application for macOS and Windows. There is no account to create and no server behind it: it runs as a program on your computer, the way a text editor does.
You give it work — read these documents and draft the summary, reconcile this spreadsheet against that report, go through this folder and tell me what changed — and it produces the artifact. The deliverable is the point; the conversation is how you get there.
It works with the models you already pay for. You bring an API key from the vendor you prefer, and Safata talks to that vendor directly from your machine. It never resells you a model and never puts a service of mine between you and one.
Most people's documents are not on their disk any more. Safata can connect to Google Drive so the agent can read the ones you are actually working on — a proposal in Docs, a model in Sheets, a deck in Slides.
It asks Google for one permission, drive.file, and
that permission is narrow on purpose:
Everything Safata knows sits in a folder in your home directory as plain files you can open, search, edit and delete. Sessions, summaries, the local search index, a ledger of what was spent and a log of every request that left the machine.
API keys and service tokens are not in those files. They live in the operating system's keychain — Keychain on macOS, Credential Manager on Windows — and are attached to a request only at the moment it leaves, and only for the one host that key belongs to.
There is no telemetry. No usage analytics, no crash reporter, no update check, no remote configuration. The trade is deliberate: bugs reach me as reports from people rather than as dashboards.
Safata is in build, from a one-person lab in Barcelona. There is no public download yet. If you would like to hear when there is, or you have a question about how it handles your data, write to me: victor.zhang.eu@gmail.com.