Settings

Settings is where you change how Diffgazer behaves before and during a review: which AI provider it uses, where it keeps your API keys, which agents run, and what the app is allowed to read. It opens from the main menu and lands on a hub that lists every settings area with its current value, so you can see your provider, theme, and storage choice at a glance before drilling in. This page assumes onboarding is done and the repository is trusted. If it is not, the provider and trust areas are where you finish that setup, and the rest of the hub will show "Not configured" or "Not trusted" until you do.

The hub is a single menu. Each row names a settings area and shows its current state on the right, so the screen doubles as a status summary. The areas are:

Pick a row to open that area. Provider data notice is not an editor: it opens the notice as a read-only dialog once accepted (or offers the acceptance while none is on record), and Esc returns to the hub. Configuration editors have Cancel and Save actions: they do not write changes before Save, and Cancel or Esc returns to the hub without applying them. The hub footer shows the project path this session is reviewing, plus a local settings caption. Your choices are persisted to ~/.diffgazer/config.json, which the footer does not print. For the full shape of that file and every field it holds, see configuration.

Diagnostics is not a configuration editor. It offers Refresh Diagnostics plus Generate Context or Regenerate Context, and each action runs immediately without a separate Save confirmation. Generating or regenerating context writes the local project snapshot under the project's .diffgazer directory. Press Esc to return to the settings hub.

Theme controls the look of the web UI. You choose between Auto, Dark, and Light, with a live preview that follows whichever option is focused. Auto follows your operating system's color scheme and tracks it live, so the app flips with the OS without a reload; the preview shows whichever of dark or light Auto currently resolves to.

Secrets Storage decides where your API keys live. Two options exist:

• File storage keeps secrets in a local file with restrictive OS file permissions (mode 0600). It is simple and portable.
• System keyring uses your operating system's keychain, which is more secure and integrated with the OS.

Changes to storage take effect immediately after you save. For more on how keys are handled and what stays local, see privacy.

Agent Execution controls scheduling, not which agents run. Sequential runs each agent one after another and works with every provider and tier. Parallel runs all agents at once, which is faster but can hit rate limits on free tiers. It defaults to sequential. Findings from successful agents use the same merge, filter, and sort rules in either mode. Reports can still differ because a rate limit or transient failure can stop one agent while the others produce a partial report.

The Provider area is where you pick the AI provider for reviews, enter or remove its API key, and choose a model. You can search and filter the provider list, and each provider shows its capabilities. API keys can be entered directly or pointed at an environment variable instead, so a key never has to be typed into the app at all. The active provider and model are global for this Diffgazer installation, so changing them affects every repository. The repository making an activation request must still be trusted, and its trust grant separately controls the repository reads used to build context and run reviews; trust does not create a project-specific provider selection. For background on providers and models, see providers.

The details pane shows at most three controls for the highlighted provider: one primary action for its state, at most one secondary, and a More menu.

The More menu always lists the same entries; one the row already shows is left out, and one the state cannot run stays listed, disabled, with its reason. Delete configuration is always last, behind a divider. Keyboard: Enter runs the primary of the highlighted row, m changes the model, e updates (or creates) the configuration, v runs Verify, d asks to delete the configuration, and c reopens the Provider data notice while it is still to be accepted. Every way to Delete configuration — the menu entry, its key, a click — opens the same Delete configuration? confirmation, which opens on Cancel; Esc keeps the configuration. While the More menu is open, ↑/↓ move through its entries and ←/→ stay inside it rather than stepping the action row (a toolbar's menu button keeps the arrows in the open menu); Esc closes it and returns focus to More. The footer lists only the keys the highlighted provider can run right now; the Help screen (?) lists them all.

The Analysis area chooses which agents run during a review. You toggle agents on and off, and at least one has to stay selected. If you have not picked any, every agent runs by default. The set you save here becomes the default agent list for reviews started from the web app.

Trust controls what Diffgazer is allowed to do in the current repository. It is scoped per repository, so trusting one repository does not trust another. The screen shows the target repository and the permissions you can grant, with Save and Revoke actions. Saving writes the trust grant for this repository; Revoke removes it.

One permission is available today: repository access, which lets Diffgazer read files and git metadata so reviews have something to analyze. The other listed permission, run commands (running tests or lint), is unavailable. The server forces it off on every save regardless of what the client sends, the option is disabled in the UI, and a security warning explains that, when it is eventually enabled, it would let the AI execute shell scripts. Treat run commands as off and not selectable for now. On the wire these permissions are the trust.capabilities fields (readFiles, runCommands).

• Configuration documents the config file these settings write to.
• Providers covers provider selection, keys, and models in depth.
• Privacy explains where keys and data live and what leaves your machine.
• Diagnostics runs health checks against the current directory.