Sources - nigifabio/swiss-job-agent GitHub Wiki

Choose sources with PROVIDERS in .env or on the Settings page (;-separated).

Keyed sources and email stay silent until configured. The Careerjet and Jooble result formats were built from their documentation; report an issue if they change.

Job-alert emails: use a dedicated mailbox (for Gmail, a separate account with an app password), subscribe it to the alerts of LinkedIn, Indeed, jobs.ch and so on, and set the IMAP_* values. The mailbox is only read, never changed.

Never scraped: LinkedIn, Indeed and Glassdoor (terms of use, bot blocking, risk to your own account). Their alert emails cover them.

A title must contain a TITLE_KEYWORDS word (a trailing * matches a prefix, e.g. coordinat*) and none of TITLE_EXCLUDE; the location must match LOCATION_KEYWORDS, or be remote and open to CH / Europe / EMEA / worldwide (ALLOW_REMOTE, REMOTE_REGIONS); the ad must be in one of LANGUAGES.

1. Exact fingerprint: title + company + location (lowercased, spaces tidied).
2. Loose fingerprint: ignores gender markers (H/F, m/w/d), work rates (80-100%), accents, punctuation, company suffixes (AG, SA, Sàrl, Schweiz…) and everything after the first word of the location.

A posting matching either fingerprint of a stored one is skipped and the first source found is kept. Jobs you add by hand are never merged. Job-Room ads republished from jobs.ch/jobup.ch are skipped: the jobup source has them with the real employer.

app/watchlists/public.json lists large Swiss employers and administrations that are crawled for every person, on top of their own watchlist (the title, place and language filters still decide what is kept). Today:

PUBLIC_BOARDS=0 in .env turns them off. To add an employer that uses one of these systems to your own watchlist, give its address in a seed file (Company = https://...):

• SAP SuccessFactors: the feed address https://<career site>/services/rss/job/ (the site gives its 20 newest jobs plus 20 per search: one search per region and search term of yours);
• Prospective: an address containing /careercenter/<number> (in the page source of the career page);
• Oracle Recruiting: the career address https://<host>.oraclecloud.com/hcmUI/CandidateExperience/fr/sites/<site>.
• Hireserve: add the entry by hand to data/watchlist.json: {"company": "...", "ats": "hireserve", "slug": "<host>/<web_site_id>/<default town>"} (the number is the p_web_site_id in the portal's addresses).

docker compose run --rm scheduler python -m app.discover app/companies.romandie.txt

probes each company in the seed file for nine career-site systems and merges the hits into data/watchlist.json. Seed files in the repo: app/companies.seed.txt (Swiss tech), app/companies.romandie.txt (Geneva/Vaud employers with tech teams), app/companies.vaud-services.txt (hotels, retail, logistics, watchmaking, services). Write your own: one company per line, or Company = https://careers-url to add a board directly (Workday often needs this).

A board is kept only if its declared company name matches the company searched for, so a company with the same short name isn't picked up by mistake. Re-check an existing watchlist with python -m app.discover --verify. The setup wizard seeds a watchlist from app/watchlists/.

• Timeouts, connection errors, 429 and 5xx are retried with backoff (honouring Retry-After, at most 60 s); other 4xx fail at once; requests to the same host are paced.
• One failing board or source never stops the others. Details are fetched only for new postings that pass the filters.
• Scan health: per-source counts and errors are stored after each scan and shown on the dashboard, with a warning when a source errors or drops to zero. API keys, the IMAP password and URLs holding keys are never written into errors or logs.
• The scheduler does not rescan when a container restarts, and a lock keeps "Scan now" and the scheduler from overlapping.