remote-jobs-digest

Stop refreshing ten job boards. Get one daily digest of the remote roles you can actually get.

"Remote" on a job board often means remote, if you live in California. rjs reads the whole posting and keeps the ones that fit where you live, the years you have, your stack and your salary floor. It pulls from public company career pages (Greenhouse, Ashby, Lever…) and remote job APIs, drops 99 % of the noise with plain rules at zero cost, and optionally asks an LLM about the few dozen survivors.

Quickstart

pipx install remote-jobs-digest          # or: uv tool install remote-jobs-digest
rjs init                                 # 13 questions, or: rjs init --from backend-ai-eu
rjs run --no-ai                          # prints today's digest, no API keys needed

A first run takes under a minute. It prints a summary and the digest, and writes a CSV, a JSON and digest_latest.md to the data dir (rjs paths shows where).

💼 Remote jobs digest — 9 shortlisted, 🆕 9 new (out of 4853 collected)
🚫 173 management/staff+ · 🏢 4 consultancy · ⚠️ 44 to review

1. 🆕 Frontend Web Application Developer — KoboToolbox
   [stack 3/10] · 💰 90,000–105,000 USD
   https://remotive.com/remote-jobs/... (via Remotive)

To cover more companies, build the list of company ATS boards once (about 10 minutes, all public APIs):

rjs boards build --fetch     # ~1,100 remote-friendly companies from public GitHub lists
rjs boards discover          # finds their Greenhouse/Ashby/Lever/... boards (~500)

Why this and not…

How it works

flowchart LR
    A[ATS boards<br/>Greenhouse · Ashby · Lever · …] --> C[collect + dedup]
    B[Remote OK · Remotive · Himalayas<br/>Working Nomads · NoDesk · HN] --> C
    C --> D[rules: location eligibility · level/years<br/>stack score · salary floor · consultancy]
    D -->|~1% survive| E{LLM pass?<br/>optional}
    E --> F[rank]
    F --> G[digest: stdout · Markdown · Telegram]
    F --> H[CSV + JSON]

1. Location eligibility uses roles, not a country list: home (where you live: onsite, hybrid and remote all fine), region (e.g. EMEA, LATAM), away (other countries in your region: remote only), blocked. "Remote" with no scope, or "Remote — LATAM" for a Europe-based user, goes to REVIEW, never straight to MATCH.
2. Level comes from what the posting asks, not the title: "Senior Engineer, 3+ years" is mid; "Engineer, 8+ years" is not.
3. Stack is a weighted table: strong words count anywhere, weak words ("python", "cloud") only in the title or tags, so a job does not score 10/10 because the company blog mentions AWS.
4. Salary floors per currency. Unknown currencies are flagged, not assumed to be USD.
5. The optional LLM reads the full description of the survivors and returns real level, fit 0–10, workload and a one-line reason. Its instructions are generated from your profile.

Configuration

rjs init writes ~/.config/rjs/config.yaml. Everything the filter uses is in that file; rjs config show prints the effective values and rjs config check validates edits. Bundled examples (rjs init --list-examples): backend-ai-eu, frontend-latam.

Keys you leave out take built-in defaults (the backend-ai-eu example), so check rjs config show after editing by hand.

Environment variables (~/.config/rjs/.env works too): see .env.example. The LLM pass takes any OpenAI-compatible endpoint: RJS_LLM_BASE_URL, RJS_LLM_API_KEY, RJS_LLM_MODEL (Groq by default; OpenAI, OpenRouter, Gemini, Ollama and LM Studio work). Telegram: TELEGRAM_BOT_TOKEN + TELEGRAM_CHAT_ID.

Running it daily

rjs run is idempotent and remembers what it already showed you (🆕 marks new jobs). Pick one: cron, a systemd user timer, or a GitHub Actions schedule in a private repo (no server).

Sources, ethics and terms of service

Please keep the defaults polite: Remotive asks for at most ~4 requests a day and Himalayas caches for 24 h, so one run a day is plenty. The digest is for you; do not republish it as a job board.

The apply extra (supervised pilot)

pip install 'remote-jobs-digest[apply]' adds tools the author uses after the digest: dossiers with a draft cover letter (rjs apply), LLM re-verification of full postings (rjs verify), screening-question drafts (rjs answer), ATS form pre-fill in a real browser that stops before submit (rjs fill), cold-email drafts (rjs outreach) and inbox triage (rjs gmail). Personal data lives in ~/.config/rjs/identity.yaml. Nothing is sent or submitted for you. These tools are less polished than the core, their prompts and messages are still partly in Spanish, and automating applications can break job sites' terms. Use them as drafts.

Limits and non-goals

• Not an auto-apply bot, and it will not become one.
• Without the LLM pass, stack scoring is keyword-based and coarse: a Python job that mentions Kubernetes can reach a Go profile. The LLM pass, or a finer stack.weighted table, fixes most of it.
• The rules are tuned for software roles; other fields need your own stack table and signals.
• Location matching is text matching. Unusual phrasings end up in REVIEW, which is the point.
• Some code comments are in Spanish; the CLI, logs, output and docs of the core are English. The CSV verdict column keeps its original values (APTA = match, REVISAR = review, DESCARTADA = rejected). PRs welcome.
• Linux and macOS. Windows is untested.

Roadmap

• --export career-ops to feed its pipeline.
• Labelled golden sets per example profile, with precision/recall in CI.
• More notification targets (Slack, email) via apprise.

License

MIT. Job data belongs to the sources and employers; respect their terms.