Personal profile site plus a public, source-checked threat service for defenders: an hourly watch of newly exploited vulnerabilities and vendor advisories, and a daily brief with a patch action and hunting focus per item. Published on GitHub Pages, delivered to WhatsApp and Microsoft Teams. Runs on your local model; keeps running on free cloud models when the Mac is off.
Architecture and full flow: docs/ARCHITECTURE.md.
| Situation | What happens |
|---|---|
| Mac on at 6:35 AM | Local llama-server (Qwen3.6-35B-A3B) writes the brief, pushes it, site deploys |
| Mac on but port busy (e.g. SEC545 batch) | Mac uses GitHub Models, then Gemini |
| Mac asleep/off | 7:20 AM ET: GitHub Actions writes it with GitHub Models, then Gemini |
| Every model fails or is rate-limited | Facts-only brief from KEV/NVD data, no commentary |
| Mac wakes later | launchd catch-up runs, sees today’s brief exists, does nothing |
| Any hour | Hourly watch runs on GitHub, no model; KEV hit on your stack alerts WhatsApp + Teams |
Cost: $0 (public repo Actions + Pages, free model tiers). Optional custom domain.
sources.yaml is the only list of inputs. Each entry records operator, tier and why it is trusted.
Enforced in code:
/method/.verified_on); new sources go through the admission criteria at the top of the file.<username>.github.io (gives you the clean root URL). Push this folder to main.CALLMEBOT_PHONE, CALLMEBOT_APIKEY | WhatsApp |
| TEAMS_WEBHOOK_URL | Teams (see below) |
| GEMINI_API_KEY | second cloud backup (free key at aistudio.google.com) |
| NVD_API_KEY | higher NVD rate limits (optional) |
| HEALTHCHECK_URL_WATCH, HEALTHCHECK_URL_BRIEF | missed-run alerts (optional, healthchecks.io) |
GitHub Models needs no secret: workflows request models: read for the built-in token.site.url in config.yaml; fill the [brackets] in site/profile.yaml.Office 365 connector webhooks are retired. In the target channel: ••• → Workflows →
“Post to a channel when a webhook request is received” → copy the URL into TEAMS_WEBHOOK_URL
(and the Mac Keychain). Posts appear as an Adaptive Card from the Workflows bot.
Use a personal or community team; check your organisation’s policy before posting into a corporate tenant.
~/watchfloor; copy mac/brief.env.example → mac/brief.env, set MODEL_PATH and GITHUB_REPO.zsh mac/setup.sh — venv, secrets into Keychain, launchd job (6:35 AM + on login), 6:30 AM wake.zsh mac/run_brief.sh && tail -50 ~/Library/Logs/watchfloor.logshow_employers: false hides names.uses: action to a commit SHA; enable Dependabot, branch protection, and a correction issue label.config.yaml against the GitHub Models catalog and AI Studio.python -m pytest -q tests # 14 offline tests
python -m digest.watch --no-alert # update hourly watch data
python -m digest.main --dry-run # full brief; print WhatsApp text and Teams card size
WF_LLM_CHAIN=github_models,gemini python -m digest.main --dry-run # simulate Mac-off path
python site/build.py && python -m http.server -d _site 8000
digest/ collectors, llm (provider chain), render (validate), notify, teams, watch, main
sources.yaml trusted source registry config.yaml chain, watchlist, delivery
site/ templates, CSS, logo, fonts, profile.yaml, build.py
mac/ launchd runner + setup .github/workflows/ watch, brief, pages, ci
data/ published JSON (git history = audit trail) docs/ architecture