Installation - nigifabio/swiss-job-agent GitHub Wiki

Requirements

  • Linux or macOS with Docker and the compose plugin (docker compose version), git, curl.
  • About 1 GB of disk for the image, 512 MB of RAM.
  • On Linux the app runs as uid 1000 inside the container: data/ must be writable by it (the installer fixes it when run as root, otherwise it prints the chown to run).

A small VM or a Linux container (Proxmox LXC with nesting) works well.

Install

git clone https://github.com/nigifabio/swiss-job-agent.git
cd swiss-job-agent
./install.sh            # or ./install.sh --yes to accept every default

The installer:

  1. checks Docker, compose, git and curl;
  2. creates .env from .env.example (mode 600) and asks:
    • setup wizard (default): import your CV / LinkedIn in the browser and let it build the search, or answer the search questions here instead (titles, region, languages, skills, sources);
    • port (8080) and listen address (127.0.0.1 = this machine only);
  3. builds the image and starts two containers: web (the interface) and scheduler (scans every 12 hours);
  4. waits for /healthz and prints the address.

An existing .env is never overwritten; re-run ./install.sh any time to rebuild and restart. Day-to-day commands (scan, logs, backup, discover…) are in the jobagent wrapper.

Prefer not to clone and build? Use the prebuilt Docker image.

First steps

  1. Open http://127.0.0.1:8080 and follow the wizard (or fill data/profile.json, see profiles, CVs and letters; an example is in examples/profile.example.json).
  2. Press Scan now on the homepage. The first scan takes a few minutes.
  3. Open a job, mark skills you have, generate a CV and letter, apply, track.

Upgrade

./install.sh upgrade

Takes a backup, git pull --ff-only, rebuilds, restarts and checks health. If the new version isn't healthy it checks out the previous commit and starts it again. Database changes are applied automatically at start-up. It refuses to run if you changed tracked files (commit or stash first).

Backup and restore

./install.sh backup      # -> backups/YYYYMMDD-HHMMSS.tar.gz (mode 600)

The archive holds all of data/: the database (a consistent copy taken from the running app), profile*.json, settings.json, the watchlist and generated PDFs. Your .env is not included: keep a copy of it somewhere safe, it may hold API keys.

Restore: stop (docker compose down), unpack the archive in the repo folder (it recreates data/), rename data/jobs.backup.db to data/jobs.db if present, then ./install.sh.

Reaching it from other devices

The app has no login of its own. By default it only listens on 127.0.0.1. To use it from your phone or another computer, choose one of:

  • VPN (Tailscale, WireGuard): set BIND_ADDR to the machine's VPN address, so only devices on your VPN can reach it. Or keep 127.0.0.1 and use an SSH tunnel (ssh -L 8080:127.0.0.1:8080 host).
  • Cloudflare Tunnel + Cloudflare Access (recommended for the internet): create an Access application for your hostname, then set in .env: 
    CF_TEAM_DOMAIN=your-team.cloudflareaccess.com
CF_ACCESS_AUD=<the application's Audience (AUD) tag>
[email protected]
    The app then verifies the Access token on every request itself, so even someone on your LAN gets 403 without a valid login.
  • Reverse proxy on another machine (nginx, Caddy, NPM): set BIND_ADDR=0.0.0.0 and make sure the proxy authenticates (Access, OAuth proxy, basic auth). Never expose port 8080 directly.

After changing .env: docker compose up -d.

Optional: more sources

Aggregator keys (free) and a job-alert mailbox are described in sources. Put the values in .env and restart.

Uninstall

docker compose down --rmi local

then delete the folder (after a backup if you want to keep your data).