Security & Data Handling

DiffGuardian ("we", "us") is a native macOS app for code review. This page explains how the app handles your code. The short version: your repositories, diffs and source code never pass through, and are never stored on, DiffGuardian's servers. You review either entirely on your own machine, or with your own AI provider keys under your own agreement with that provider.

Where your reviews run

Every AI feature flows through one router, and the backend you choose determines exactly where your code goes:

• Local models (Ollama / LM Studio) — requests go only to your own machine. Your code never leaves your Mac. This option needs no network access and is always available.
• Your own keys (Anthropic / OpenAI) — your code is sent from your machine directly to the provider, authenticated with your API key. It is never relayed through DiffGuardian. Because it is your key, that data is governed by your own agreement with Anthropic or OpenAI — for this flow, the provider is your subprocessor, not ours. Their commercial APIs (unlike the consumer chat apps) do not train on API inputs by default; review each provider's API data-usage terms for specifics.
• DiffGuardian's servers — used only for sign-in, billing, push notifications, and (if you opt in) anonymous usage counts. They never receive your code, diffs, prompts, or AI responses.

What leaves your machine

• Code and diffs for AI review — to the backend you choose: local (stays on device) or your own provider key (direct to the provider). Never to our servers.
• Knowledge-base embeddings — computed on-device by default (a built-in local model with an on-device sqlite-vec store). Sent to OpenAI only if you explicitly select OpenAI embeddings.
• API keys and your GitHub token — encrypted in the macOS Keychain and used only on your device to call the provider or GitHub directly. They are never sent to our servers.
• Account, billing and opt-in telemetry — to our servers: your sign-in and plan, and, if you opt in, aggregated usage (vendor, model, token counts, feature name) and anonymous product analytics. None of this includes code.

What our servers never receive

Your source code, diffs, AI prompts, AI responses, repository names, branch names, or pull request contents. None of these ever reach DiffGuardian's infrastructure.

Secrets

Every credential — your Anthropic, OpenAI, and GitHub keys — is encrypted at rest in the macOS Keychain and used only on your device. The app is notarized by Apple.

Subprocessors

We rely on a small set of established, industry-standard providers, each processing only what its function requires — and none of them ever receive your code:

• an identity provider, for sign-in;
• a PCI-compliant payments provider, for billing (we never see your card details);
• a managed database, for your account, plan and policy data;
• a realtime messaging provider, for notifications;
• a product-analytics provider, for anonymous usage events; and
• an error-monitoring provider, for crash diagnostics.

A current list of subprocessors is available on request. Separately, you connect your own integrations — your code host (e.g. GitHub, with your token) and your chosen AI provider, whether a local model or your own Anthropic or OpenAI key — which operate under your own agreements, not ours.

Compliance & certifications

DiffGuardian is an early-stage product and is not yet SOC 2 certified. We've built the product so that the highest-risk data — your source code — is architecturally kept off our servers entirely, which removes the largest class of exposure rather than simply controlling it. A data-processing agreement (DPA) and security questionnaires are available on request.

Reporting a vulnerability & contact

Evaluating DiffGuardian, need a security review or DPA, or want to report a vulnerability? Email security@diffguardian.ai.