Trust
What happens to your data, in plain words.
The privacy policy and the terms say all of this at length. This page says it so a person, or the assistant they asked, can read it in two minutes. Where the two differ, the long form governs.
The documents you check are never stored.
When you check a document against a condition, it is read once to answer that check. What stays on the record is the verdict and a receipt: the passage it rested on, a hash of the file, its name. The document itself is discarded. Our model providers keep their own short-lived logs on their published schedules; "never stored" describes our servers.
We do not train models on your content.
Midfire manages every model key. Your decisions and documents go to the provider to answer your request and to nothing else. The provider today is Anthropic, under its commercial terms.
Your record is yours and leaves as plain files.
Any member can export what they can open, from Settings, at any time, as plain files. Close the workspace and ask us at hello@midfire.ai: we delete the workspace, its members' accounts and its record.
Who sees what.
Editors run decisions. Readers read the decisions their folders open to them, and reading is free. Someone asked for input sees the decision they were asked about and nothing else. Naming a person in a role grants them nothing by itself: access is its own fact on the record. Every change carries a name and a time.
Where it runs.
In the United States, and our subprocessors are United States companies: Supabase, Vercel, Anthropic, Jina AI, Stripe, Google Workspace. What each one receives, and why it is in the path, is on the subprocessors page.
What we do not claim yet.
There is no SOC 2 report, no single sign-on, and no choice of data region. Each of those arrives when a customer needs it; ask, and we will tell you where it stands. The features are in beta and the terms say so.
The long form: the privacy policy, the terms, and the subprocessors. Questions to hello@midfire.ai.
See a decision record before you put yours in one.
Step into the demo, three fictional companies and their records, or bring a real decision to a session with us and ask your data questions there.