Skip to content
Back to Work

Argus

Case study · Software · 2026

A desktop app for keeping track of opportunities, staying ahead of deadlines, and preparing applications without losing sight of the details.

My role
Product, design, architecture, engineering, testing and release
Platform
macOS and Windows
Status
Private beta, v0.12
Year
2026
The Argus Today screen: applications ready to send, opportunities closing within a week, and weekly progress

The idea

The job search was scattered across too many places. I wanted one place to manage it.

Jobs, scholarships and programmes each came with their own links, forms, deadlines and requirements. I wanted a way to capture an opportunity when I found it, keep the important details together, and make the next step obvious.

Argus runs on your computer. You can add an opportunity from a link, screenshot or photographed poster, track its deadline, review how well it matches your experience, tailor your CV, and prepare an email in your Drafts. It does not send an application for you; you review it and decide when to send.

This page covers how I built Argus. The product site has the full feature overview.

Argus Discover view with a list of found opportunities and their match scores
Opportunity discovery: Review what Argus finds and why an opportunity may be a match.
Argus CV tailoring screen showing suggested rewrites with reasons, and a flagged claim to check
CV tailoring: Suggestions include explanations, and claims that need checking are flagged rather than silently added.

Screenshots use synthetic test data.

How it’s built

Keep the architecture small, and make the important parts explicit.

Argus technology choices
LayerChoice
AppPython, Flask and SQLite in one processKeeps the application and its data straightforward to run and maintain.
Desktop windowpywebview and PyInstallerA desktop application without shipping a full Electron runtime.
InterfacePreact and htm as ES modulesAn interactive interface without a large build toolchain.
AIGemini, Claude or a local model, selected by taskLets the user choose a provider and decide which model handles each task.
SecretsOS keychain and encrypted CV storageKeeps sensitive values out of plain configuration files.
ReleasesGitHub Actions triggered by version tagsProduces Apple Silicon, Intel and Windows builds from the same release process.

Decisions that mattered

Applications stay under the user’s control

Argus prepares an application and leaves the email in Drafts. It doesn’t send messages under someone’s name without their approval. I wanted the automation to save time without taking away the final decision.

Prepare in batches, review one at a time

An early design had a separate multi-step wizard for every application. I changed it to a queue: prepare several in the background, then review each result in one place. That better fits the way a real job search works.

Local-first, with clear choices about AI

A job search contains personal information. The core records live on the user’s computer, and tasks can be directed to a local model instead of silently falling back to a cloud provider. When a cloud model is selected, the data sent to that provider depends on the task and the provider chosen.

Don’t make an older computer do unnecessary work

Long-running AI tasks run in the background, and the app doesn’t download large dependencies at runtime. The goal is to keep it practical on everyday hardware, including machines that aren’t new or particularly powerful.

Where it stands

Private beta · v0.12

Argus is currently invite-only. Releases include plain-language notes, the test suite has more than 650 cases, and beta telemetry is limited to anonymous usage counts rather than the contents of opportunities, CVs or API keys.

  1. v0.9.013 July 2026First private beta, with capture, reminders, CV matching and discovery.
  2. v0.10.020 July 2026Invite-only access, a full Settings page, one-click data export and a plain-language telemetry disclosure.
  3. v0.11.019 September 2026Redesigned interface, batch applications, career memory and CV tailoring that preserves document formatting.
  4. v0.12.021 September 2026Applications prepared as one PDF, capture from a phone through Telegram, and updates without a restart.

The current builds aren’t code-signed, so macOS and Windows may show a warning on first launch. I’d rather be clear about that than leave people wondering what it means.

Next project

Principia Works

The cybersecurity startup I founded to show people and organisations what attackers can learn about them from public information.

Explore Principia Works