Data handling

Soffi is built so that your team keeps control of the things a security review cares about, your data, your repositories, and your release process, while Soffi takes on the work of running your app.

Stays in your control

Your data. By default, previews run on seed data, not your production data. A Data Twin stands in for your backend, so nobody needs a real customer account or a production database to work on a screen. Real data can still reach Soffi in ways your team chooses: a key an Admin adds for a third-party service lets previews call that service and show what it returns, and the content people bring in, such as tickets, Figma files, and uploaded documents, is stored and read like any other work in Soffi. So are the knowledge docs Soffi writes from your code.

Your CI. Your branches, reviewers, and checks don't change. Work from Soffi arrives as an ordinary pull request, your CI runs on it, and nothing merges without your review. Soffi never deploys to production.

Your repos. You choose which repositories the Soffi GitHub app can see when you install it, and you can change that selection or revoke the app anytime. See GitHub, which also covers what linking a personal GitHub account grants.

Soffi handles

Deployments. Building and running your app in Soffi, for every Project and every Workspace.

Updates. Rebuilding and upkeep as your codebase changes.

Infrastructure. A private, isolated environment for every Workspace. See Isolation.

What Soffi keeps

To do that work, Soffi stores:

• A copy of the repositories you grant, so it can run and edit your app.
• The work your team does in Soffi: Workspaces and their history, comments, and video comments.
• Your Project's configuration and keys. Key values are encrypted at rest, and only Admins and the Owner can enter or change them. See Environment variables and keys.
• Access to the tools you connect. The access Soffi holds for linked GitHub accounts, Linear, Jira, and Figma is encrypted at rest. Disconnecting an integration deletes it.

Removing a repository, deleting a Project, or uninstalling the GitHub app stops Soffi from working with that code, and none of them changes anything on GitHub. They aren't the same as deleting your data from Soffi. For customers with a Customer Agreement, deletion and return of data follow the timelines in that agreement and its Data Processing Addendum.

Soffi reads tickets and designs from your connected tools when someone attaches them or Soffi looks one up, and nothing in a Workspace changes them. The Jira and Figma connections are read-only, and Jira is limited further to the projects an Admin chooses. The Linear connection asks Linear for read and write access. See Linear, Jira, and Figma.

Third parties

Soffi uses third-party AI models and service providers to run the product, for things like hosting, error monitoring, product analytics, and email. When Soffi works in a Workspace, the AI models receive what the task needs, which can include your code, what's attached to the Workspace, and the conversation.

As Soffi's Terms of Service state, Soffi doesn't use your source code, documentation, knowledge content, prompts, or AI outputs to train, fine-tune, or improve any AI model, and doesn't permit its AI model providers to do so through Soffi-managed accounts.

The current list of subprocessors is in the Soffi trust center (opens in a new tab).

Keys you give a Project

A key someone adds to the Project for a service your app calls is the main way real data enters a preview. Use scoped, least-privilege credentials, such as test-mode or read-only keys, and keep real secrets out of keys your frontend compiles into the browser.

For the full controls and the report behind them, see Compliance.