jobboard

Public site: https://ceeyang.com/jobboard/ (same as ceeyang.github.io/jobboard/, which redirects there) - rebuilt automatically by GitHub Actions.

Update frequency: automatic, but not punctual. The workflow's cron says "every hour", but GitHub decides when a scheduled workflow really starts (best effort). Measured on this repository: 2.4-8 hours apart, about 6 hours on average lately. Don't treat it as an hourly promise. "Run workflow" on the Actions tab refreshes at once; the page shows "N hours ago" in its top bar. Want a true hourly run? Call the workflow_dispatch API from an external timer (optional, needs a token).

Make your own board (any kind of job)

What a site looks for is one config file, config/profile.json, not code. Pick a preset, edit the keywords, let GitHub Actions build the site:

python3 -m jobboard.cli presets                       # frontend, mobile, backend, data-ai, devops, qa, security, web3, game, product, design, ...
python3 -m jobboard.cli init --preset backend         # writes config/profile.json
python3 -m jobboard.cli profile --try "Senior Go Engineer" "Sales Manager"     # check your keywords on real titles

The site opens on an overview (new today, where the jobs are, 14-day trend, who sponsors visas) with the full list one tab away, and every site publishes an RSS / JSON feed of its newest recommended jobs. Three ways to deploy (fork and scrape yourself / a light repo that reads an aggregate site's feed / your own organisation with an aggregate site plus one site per role): docs/FORK.md. The local edition (this README) additionally reads your private Telegram groups, LinkedIn and key-based APIs.

Reads your Telegram job groups/channels (via cli-anything-telegram's Telethon session), finds postings that match Flutter / iOS / Android / web, dedupes them across channels, and shows them in a local web board.

uv sync
uv run jobboard fetch            # Telegram channels + job sites (add --no-web / --no-telegram / --only RemoteOK / --force)
uv run jobboard serve            # board at http://127.0.0.1:8765
uv run jobboard digest --dry-run # preview the daily summary (drop --dry-run to send to Saved Messages)
uv run jobboard reanalyze        # re-score stored jobs after editing jobboard/profile.py
scripts/install_launchd.sh       # hourly fetch + 09:00 digest + board always on

Edit jobboard/profile.py to change roles, remote keywords, risk rules. Data: ~/.local/share/jobboard/jobs.db. Only digest writes to Telegram (one message to your own Saved Messages); nothing is ever sent to other people.

Job sites (overseas, remote-friendly)

RemoteOK, Remotive, We Work Remotely, Himalayas, Jobicy, Working Nomads and Hacker News "Who is hiring?" are read through their public APIs/RSS (jobboard/sources/adapters.py; add a site by writing one function and listing it in SOURCES).

• Per-site minimum interval (Remotive and HN: 6h - Remotive's terms allow ~4 requests/day); one failing site never stops the rest.
• Every job keeps its source name and original link, as the sites' terms require. Don't republish the data.
• Site tags are often keyword-stuffed, so only the job title counts fully (profile.py: SPECIFIC_TAGS / NEG_TITLE / STACK_MISMATCH).
• geo marks postings as 🌍 worldwide / 🚧 region-restricted (e.g. "USA only") / unknown; the board can hide restricted ones. Not included on purpose: Indeed, BOSS/拉勾 and other login-walled or anti-bot sites.

More sources (batch 1-3 + LinkedIn)

uv run jobboard discover                  # find company boards: links in stored jobs + built-in seed + Common Crawl
uv run jobboard discover --file my.txt    # your own list: ATS links, `lever:slug` lines, or career-page URLs
uv run jobboard fetch --only Greenhouse,LinkedIn --force

Source	How	Needs
Greenhouse, Lever, Ashby, Recruitee, Personio, Workable, Teamtailor	each employer's public board API, only titles that match your roles are kept	companies in the companies table (discover fills it; a board is stored only after it answered)
Career pages	schema.org JobPosting JSON-LD (single job pages, or a listing page whose job links are followed)	page URLs added with discover --file
Adzuna	search API, queries = your roles' English keywords	ADZUNA_APP_ID, ADZUNA_APP_KEY (free)
USAJOBS	federal jobs	USAJOBS_KEY, USAJOBS_EMAIL (free)
The Muse, Arbeitnow	public APIs	nothing
NoDesk, Remote First Jobs, Real Work From Anywhere	remote-only boards' RSS feeds (public site too)	nothing
Eleduck 电鸭	Chinese remote / freelance community, its JSON API (local board only: no published terms)	nothing
LinkedIn	logged-out jobs-guest pages, no account, no cookies	nothing; locations = setting linkedin_locations (default ["Remote"], edit via PUT /api/settings)

• A source without its key shows as skipped in fetch output; it never counts as an error.
• fetch runs a cheap discovery first (links found in stored jobs; the seed list when the registry is empty). A board that returns 404 five fetches in a row is switched off; being offline never does that.
• LinkedIn is for personal, low-volume use only. It is an undocumented interface that can change or throttle at any time, and automated access is against LinkedIn's terms. It is kept gentle (2 pages per query, several seconds between requests, at most 40 description downloads per run, stops at the first HTTP 429, 6 h minimum interval). Don't publish or share what it collects. Using your logged-in account for this is deliberately not supported: the realistic risk is getting that account restricted.
• Public does not mean redistributable: every job keeps its source name and original link; the data stays on your machine.
• Re-scoring after editing roles (reanalyze) works for all new sources the same way.

Government job services and visa sponsorship

Source	Coverage	Notes
Germany (BA)	Bundesagentur für Arbeit "Jobsuche"	the API behind arbeitsagentur.de (public key header), last 28 days
Sweden (AF)	Arbetsförmedlingen / Platsbanken (JobTech open data)	no key
EURES (EU)	EU Commission job vacancy service, ~30 countries	the API behind the EURES portal; title search, last month
Canada Job Bank	jobbank.gc.ca search pages	obeys robots.txt (5 s delay); jobs whose employer requested an LMIA (= may hire a foreign worker) are flagged

Australia is not covered: Workforce Australia and APS Jobs publish no API or open feed (only third-party scrapers exist). France Travail needs a free OAuth client and is not wired up yet; UK "Find a job" has no public API.

Visa sponsorship is a flag on every job (visa: yes / maybe / empty) and a switch in the sidebar ("只看雇主担保签证"): yes when the posting says the employer sponsors or arranges a visa (English, German, Chinese phrases) or the source says so officially (Canada LMIA); maybe = only relocation support. An explicit refusal ("no visa sponsorship", "must have the right to work") always wins. It is a text heuristic - always confirm with the employer.

GitHub group (jobs kept in public repositories)

Sidebar group 🐙 GitHub, public edition included (excerpt + link to the issue, no contacts).

Source	Shape	Notes
Rebase who-is-hiring (rebase-network/who-is-hiring)	every job is an issue; read from the community's normalised JSON hire.rebase.network/jobs.normalized.json (~900 open, Chinese, blockchain and others, many remote)	the JSON also lists contact channels: they are never read
frontendbr/vagas, react-brasil/vagas, backend-br/vagas, androiddevbr/vagas, flutterbr/vagas	one open issue = one job (Brazil, Portuguese; many remote roles paid in USD)	list in config/github_job_repos.txt (add owner/repo, max 9)
SimplifyJobs New-Grad / Summer2027 Internships	structured listings.json (13 MB, fetched gzipped)	early-career roles; its sponsorship field feeds the visa filter ("Offers Sponsorship" -> yes)
Issues are read through the REST API; GITHUB_TOKEN (set automatically in the workflow) lifts the limit from 60 to 5000 requests/hour.
Repositories that are only directories (remoteintech/remote-jobs, tech-jobs-with-relocation, awesome-job-boards, open-source-jobs)
list companies or boards, not postings, so they are not sources; their company links are good input for jobboard discover --file.

Global remote filter

The sidebar switch 🌍 只看全球远程 keeps only jobs that are remote, not restricted to a region and not tied to a city. "Restricted" is read from the location field and from the text ("US only", "must be located in the UK", "authorized to work in the US", "仅限美国" ...). Jobs with no location at all (most Telegram/HN posts) count as global unless their text says otherwise. Run jobboard reanalyze once to apply the text rule to jobs stored earlier (the board and the public export also apply it on the fly).

Public site (GitHub Pages)

.github/workflows/update.yml is scheduled hourly in the file, but GitHub starts scheduled runs when it likes (about every 4 hours in practice, see the note at the top), and on demand: fetch --mode public -> export -> Pages. State lives on the orphan data branch (one commit, replaced each run).

• Reduced edition on purpose. Only whitelisted sources (jobboard/mode.py): remote job boards, HN, employers' own ATS boards, Sweden, EURES, and the public Telegram channels listed in config/telegram_channels.txt. Never LinkedIn, Adzuna/USAJOBS, Germany, Canada, The Muse, Arbeitnow, and never anything read with your personal Telegram login.
• Telegram on the public site is read through each channel's public web preview (https://t.me/s/<name>, no login), so only channels with a preview work; private groups and invite-only groups are skipped automatically and reported in the run summary. Edit the list in config/telegram_channels.txt (it is public, together with the repository).
• Excerpt + link only. Per job: title, company, location, salary, a <=600 character excerpt and the link to the original. No contact handles/e-mails (a final check fails the build if one slips through), no full texts, no notes.
• Read-only. Statuses (interested/applied/...) and saved searches live in your browser only. Roles, refresh and digest are edited in the local edition.
• Change the whitelist or window in jobboard/mode.py / publish.py (WINDOW_DAYS = 90), then push.
• Pages must stay public on a free account; what is published is public to everybody, so keep the whitelist conservative.
• If GitHub pauses the schedule after 60 days without repository activity, re-enable it on the Actions page.

How the posted time is decided

• Every date is stored and exported in UTC (+00:00); sorting and the 1d/3d/1w/1m filter compare real instants, not text.
• A job's date is the earliest date any of its sources reported (first appearance). If a source showed the same job again a day or more later (a Telegram channel re-posting an ad), the card adds "重发 X 前" and the detail panel shows both times.
• Sources that only give a day (SimplifyJobs, LinkedIn, Canada, Germany) are shown as "10/23 发布" instead of an invented hour, and count as "some time that day" for the filter and the ordering.

Time range

The board has a 发布时间 filter: any / 1 day / 3 days / 1 week / 1 month. The API takes ?since=1d|3d|7d|30d. A job's date is the earliest date any of its sources reported, so a posting re-listed elsewhere does not look new. Dates come from the source (Greenhouse first_published, Lever createdAt, Ashby publishedAt, ...); LinkedIn only gives a day, so its jobs count from 00:00 UTC of that day.

Settings (⚙ in the board)

• Auto refresh: off / 15 min … 24 h / custom (5–1440 min). Runs inside jobboard serve, so the board has to be running. Remotive and HN keep their own 6 h minimum.
• Daily digest (opt-in): sends the top matches to your Telegram "Saved Messages" at the time you choose, once per day.
• Common roles: name + comma-separated keywords + weight (1–10). Add from templates (React Native, 鸿蒙, 小程序, …), edit, disable or delete; existing jobs are re-scored immediately.
• Search syntax: space = AND, a|b = OR, -term = exclude, "exact phrase". Save a search as a chip in the sidebar.
• Layout: sidebar ≈14vw (collapse with b), list : detail = 5 : 6 of the remaining width.