Model heating mode when rooms have net heat loss in cold weather (<10°C).
AC units with heat pump capability (canHeat) provide heating capacity,
with the same 20% headroom threshold used for cooling. Adds cold risk
detection, cold-weather actions, and full frontend support including
heating mode timeline colors, room budget heating display, and i18n.
Move dynamic HTML generation from JS template literals to native
<template> elements rendered server-side with i18n. JS now clones
templates, fills data-slot attributes via textContent (auto-escaped),
and uses event delegation instead of inline onclick/window globals.
- Use custom image registry (somegit.dev) with pull secrets and "latest" tag.
- Enable HTTPRoute with hostname "heatguard.dev" and TLS via cert-manager.
- Activate autoscaling with min/max replicas and resource metrics.
- Switch Dockerfile to Alpine runtime with nonroot user.
- Add helm directory to .dockerignore.
- Replace Ingress template with HTTPRoute (gateway.networking.k8s.io/v1)
- Add cert-manager.io/cluster-issuer annotation for auto TLS
- Add imagePullSecrets, env, envFrom support to deployment
- Update NOTES.txt with ITSH Cloud DNS records
Insert step 5 (Configure AI) into the getting-started guide with
mentions of Setup > AI Summary, --llm-provider/--llm-model CLI flags,
and Ollama local support. Renumber Fetch Forecast to step 6 and View
Dashboard to step 7.
Remove unused crypto module, DataDir/DefaultDBPath (SQLite remnant),
and ListenAndServe (replaced by direct http.Server in main). Strip 17
unreferenced i18n keys from en/de translations. Add --llm-provider,
--llm-model, and --llm-endpoint CLI flags for runtime LLM override
without a config file. Rewrite README with correct Go 1.25 version,
shields, LLM providers table, Docker/Helm deployment docs. Fix
.gitignore pattern to not match cmd/heatguard/ directory.
Add health endpoints (/healthz, /readyz), graceful shutdown with
SIGTERM/SIGINT handling, multi-stage Dockerfile with distroless
runtime, and a full Helm chart with security-hardened defaults.
Pass profile timezone through DashboardData so the frontend can
compute the current hour and highlight it on the heatmap (white ring +
orange triangle marker). Only activates when the dashboard date matches
today in the profile timezone.
In cold weather (peak < 22°C, risk low), the risk card now shows a
teal "Comfortable" presentation with a checkmark icon instead of the
generic green "Low" shield.
- Fix ComputeRoomBudget: no-AC rooms check if open-window ventilation
can offset gains instead of defaulting to Overloaded. Net-cooling
rooms are now Comfortable; ventilation-solvable rooms are Marginal.
- Add "comfort" cool mode for hours where outdoor is >5°C below indoor
and budget is not overloaded (winter/cold scenarios).
- Reorder determineCoolMode: sealed now before overloaded, fixing
humid+cold+no-AC giving "overloaded" instead of "sealed".
- Update LLM prompts: document comfort coolMode, add cold-weather
guidance for summary and actions generation.
- Add dashboard forecast refresh button: fetches fresh forecast +
warnings, then re-runs compute and LLM pipelines.
- Extract forecast fetch into shared fetchForecastForProfile() in db.js,
deduplicating logic between setup.js and dashboard.js.
- Add indoor humidity support, pressure display, and cool mode sealed
integration test.
Replace Unicode edit/delete characters with SVG pencil/trash icon buttons,
add edit mode visual feedback (orange ring + Save/Cancel buttons), group
action buttons consistently across all entity types, add hover effects to
list items, and constrain all pages to 1800px max-width via layout template.
Skip LLM actions call when peakTempC < 22°C and risk is low. Add
low-risk guidance to both summary and actions system prompts so the
LLM returns appropriate responses on mild days. Restructure dashboard
into a top-level two-column grid with sidebar beside all main content
and remove max-w-7xl cap for full-width layout.
Replace the 720x200 SVG line chart and separate cooling strip with a
compact two-row heatmap: 24 colored temp cells (48px tall) with a thin
cooling mode row below. Hour labels every 3h, tooltip on hover/click,
responsive mobile layout (temp values hidden <640px, color-only).
Replace CLI + SQLite architecture with a Go web server + vanilla JS
frontend using IndexedDB for all client-side data storage.
- Remove: cli, store, report, static packages
- Add: compute engine (BuildDashboard), server package, web UI
- Add: setup page with CRUD for profiles, rooms, devices, occupants, AC
- Add: dashboard with SVG temperature timeline, risk analysis, care checklist
- Add: i18n support (English/German) with server-side Go templates
- Add: LLM provider selection UI with client-side API key storage
- Add: per-room indoor temperature, edit buttons, language-aware AI summary
Add full LLM configuration section to the setup page with provider
dropdown, model/API key fields for cloud providers (Anthropic, OpenAI,
Gemini), and endpoint/model fields for Ollama. API keys are encrypted
with AES-256-GCM and stored in the database.
Add Gemini provider using the generativelanguage.googleapis.com API
with systemInstruction support. Activated via --llm gemini with
GEMINI_API_KEY env var. Default model: gemini-2.0-flash.
Add server-side rendered setup UI accessible via `heatwave web`.
The dashboard is now re-rendered per request and includes a nav bar
linking to the new /setup page. Setup provides full CRUD for profiles,
rooms, devices, occupants, AC units (with room assignment), scenario
toggles, and forecast fetching — all via POST/redirect/GET forms.
- Add ShowNav field to DashboardData for conditional nav bar
- Extract fetchForecastForProfile() for reuse by web handler
- Create setup.html.tmpl with Tailwind-styled entity sections
- Create web_handlers.go with 15 route handlers and flash cookies
- Switch web.go from pre-rendered to per-request dashboard rendering
- Graceful dashboard fallback when no forecast data exists
