Settings

Appearance, the crew character, your account, Gmail, auto-start, how boldly the app answers, and the model that writes. Settings is one long page with labelled groups.

Appearance and crew

Appearance

A segmented control: System, Light, or Dark. It changes the app immediately and is remembered.

Crew

The optional animal character that reacts while the app works. You can:

Turn Show a crew character off entirely.
Turn Animate the character off to keep it still.
Choose a character, and preview its reaction.

The crew is decoration, not a status indicator. The queue's real state is always in words.

Account

Shows whether you are signed in, the email on the account, and a Sign out button. Signing out keeps your workspace on this Mac.

Cloud workspace shows the sync status. If the cloud workspace is empty, Upload this Mac's workspace copies the local workspace up. Cloud features depend on the build's service configuration.

Mail and Gmail

Gmail is optional but it removes two kinds of interruption: emailed security codes during submission, and manual status updates after replies arrive.

The Gmail buttons and what each grant allows.
Button	Permission
Connect for security codes	Read-only. The app can look for a code from Greenhouse or Ashby arriving after Submit, and can read replies for status tracking.
Connect with outreach sending	Read and send. Adds the ability to send an approved recruiter email after a verified submission.
Verify connection	Checks the saved grant against Gmail without reading message bodies.
Check replies	Reads linked replies now and records them with applications.

Track application status from email is the switch for reading company replies and moving applications through History. It uses your model for each likely update; turn it off to save model usage. It still needs Gmail read access.

Read-only never replaces send access If Gmail can already send, the read-only button is disabled so a narrower grant cannot silently remove a capability you chose. Use Connect with outreach sending again if you need to renew both.

Without Gmail, a visible browser waits 90 seconds for you to type a security code. If nobody does, the job is saved as not submitted and the queue moves on. Continue in Needs you runs it again for the new code.

Automation

Setting	Default
Start the queue by itself when I'm not using the app	On
After … minutes without a click or key press in Job Applicator	5 minutes, adjustable from 1 to 120

“Not using the app” means exactly that: using other apps counts as being away. The queue starts once per quiet stretch, and never after you paused or stopped it, and never while setup, the terms, or an update is blocking work. See Queue → Starting by itself.

When the app isn't sure

This is the answering mode. It controls how much the app may infer before it stops and asks.

Mode	Behavior
Ask me	Asks about anything it is not sure of.
Balanced	Guesses low-risk questions such as how you heard about the job, referrals, relocation and commuting from your preferences, and total years of experience from your work history. Asks about the rest.
Fill it for me	Guesses nearly every question, choosing the option that best helps the application. Legal and eligibility questions still are not guessed.

Never guessed, in any mode Work authorization, visa sponsorship, citizenship, criminal history, security clearance, age, degrees, employers and dates, certifications, salary, demographic and veteran questions, and legal consents. These come from your profile or your own answers, or the app asks you.

How answers are written

Two related sections live here: the on-device model and the model connection.

The Default

Everyone starts on Default (recommended). The app picks the model and holds the credential; you do not enter a key. The hosted model's name is hidden while the Default is selected.

Other providers

Choice	Needs
Built into Job Applicator (this Mac)	Apple Silicon and at least 8 GB of memory. Downloads a model and runtime the first time; it runs locally.
Ollama app on this Mac	Ollama installed and running with a model downloaded.
LM Studio app on this Mac	LM Studio open, a model loaded, and its server started.
OpenAI	An API key. Billed by OpenAI.
OpenRouter	An API key. Billed by OpenRouter.
OpenCode Zen	An API key. Billed by Zen.
Another provider	An HTTPS address and an API key, for any OpenAI-style chat API. Your résumé and answers are sent there.

Keys and checks

A saved key is kept in the macOS Keychain and is never shown or copyable in the app. You can replace or remove it.
Check connection makes one short real request. A model that is not loaded yet can take a little while, and a hosted check can cost a small amount.
Save records the provider and model. Changing providers clears the model field and lists the new provider's models.
Thinking before answering can be Off (faster) or A little. In testing, thinking made answering two to five times slower without answering more questions, so it is off unless you choose it.

If the model stops answering The queue pauses under a gentle banner rather than failing every job. Open Settings in that banner brings you here. Local providers are the usual cause: the app is closed, the server stopped, or the model was unloaded.

About this Mac

Terms and policies lists the documents you accepted and links to the public copies. Check this Mac reports the architecture, macOS version, memory, and free disk space. This is a quick way to confirm the Mac meets the requirements.

Updates, version details, and sign-out are in the account menu at the foot of the sidebar: Check for Updates…, Copy Version Details, and Sign Out. The menu bar icon also has Check for Updates….

The app keeps account data under ~/Library/Application Support/Job Applicator. Uninstalling the app is not the same as deleting server-side account data; see the privacy notice.