You are viewing archived documentation for v0.4. Go to latest →

BirdNET-NG Feature Roadmap

Phase 1 — Core Loop (MVP) ✅

The minimum viable distributed bird identification system.

P1.1 — Satellite Audio Capture ✅

  • Audio capture from USB/I2S microphone via ALSA (arecord)
  • 3-second chunking with configurable overlap
  • WAV encoding at 48kHz mono (BirdNET native sample rate)
  • Local disk storage (write-before-send)
  • SQLite outbox queue (sql.js, persisted to disk)
  • Simulated capture mode (sine wave) for development
  • Replay capture mode (real bird audio from WAV files) for testing

P1.2 — Satellite → Hub Transport ✅

  • MQTT client with auto-reconnect
  • Base64 audio payload with JSON metadata envelope
  • Outbox drain loop with exponential backoff
  • Hub-side acknowledgment (chunk marked acked only after storage confirmed)
  • Configurable retention policy for local cleanup

P1.3 — Hub Audio Ingestion ✅

  • MQTT subscriber receiving audio chunks
  • Store audio blobs in MinIO (keyed by {tenant_id}/{satellite_id}/{chunk_id}.wav)
  • Insert audio_chunks record in PostgreSQL
  • Enqueue processing job in BullMQ

P1.4 — Inference Worker ✅

  • BullMQ consumer pulling audio processing jobs (async Python worker)
  • Download audio from MinIO
  • Run BirdNET TFLite model (via birdnetlib, tflite-runtime 2.14)
  • Write detection results to PostgreSQL
  • Basic species identification with confidence scores
  • Geo-aware species filtering (location + date)

P1.5 — Web UI ✅

  • Dashboard: satellite status cards, recent detections, species count
  • Detection cards: 2-column flex layout with SpectrogramPlayer
  • SpectrogramPlayer: unified spectrogram + audio player component on detection cards
  • SpectrogramPlayer: click-to-seek on spectrogram, drag to select region for loop playback
  • SpectrogramPlayer: playback cursor line synced with spectrogram
  • SpectrogramPlayer: signal boost (1x–20x via Web Audio GainNode), speed control (0.25x–2x)
  • SpectrogramPlayer: volume slider with mute toggle, download button
  • SpectrogramPlayer: times displayed with centisecond precision
  • Vote buttons (Yes/No/Unsure) under confidence bar
  • Satellite table layout with sortable columns (Name, Status, Profile, Version, Last seen)
  • Search by name, filter by status, satellite count (online/total) in header
  • Slide-in side panel on row click (General, Hardware, Schedule, Hub Configuration, Actions)
  • Smooth slide animation on panel open/close
  • Satellite version shown with amber warning when satellite version != hub version
  • Tenant selector with "All Tenants" option
  • Dark theme, responsive layout
  • Spectrogram visualization (Cooley-Tukey FFT, Viridis colormap, 0–12kHz)

P1.6 — Multi-Tenancy Foundation ✅

  • Tenant registration and management
  • API key generation scoped to tenant (viewer, member, admin, satellite roles)
  • Row-level tenant isolation on all queries
  • Tenant-scoped MinIO paths
  • API key authentication middleware with role-based guards
  • Tenant scoping enforcement (scoped keys cannot access other tenants)
  • Internal API key for web frontend (cross-tenant admin)

P1.7 — Deployment ✅

  • Docker Compose for full 8-container hub stack
  • Traefik integration (HTTPS + MQTTS TLS termination)
  • Four FQDNs (web, MQTT, MinIO console, documentation site)
  • All configuration via .env environment variables
  • Persistent storage via bind mounts
  • Satellite install script for Raspberry Pi

Phase 2 — Intelligence

Smart features that improve detection quality and user experience.

P2.1 — Geo-Aware Inference ✅

  • Satellite GPS coordinate reporting
  • Live GPS via gpsd for Pi satellites
  • Species range filtering based on location + month
  • BirdNET location filter integration in inference worker
  • Reduced candidate list → faster and more accurate inference

P2.2 — Rare Species Alerts ✅

  • Frequency-based rarity model (first sighting + low frequency detection)
  • Real-time alert generation on unusual detections
  • WebSocket push to connected UI clients (toast notifications)
  • Webhook support (Slack, custom URL)
  • Per-tenant custom watchlists (tenant settings, watchlist_species array)

P2.3 — Confidence-Based Re-Processing ✅

  • Configurable "uncertain" confidence band (default: 0.4–0.7)
  • Auto-requeue uncertain detections with lower threshold (0.05)
  • Second pass updates confidence with GREATEST(old, new)
  • Flag for human verification if still uncertain (pending_review)

P2.4 — Satellite Health & Telemetry ✅

  • Periodic telemetry reports from satellites (battery, storage, CPU, network)
  • Fleet dashboard in web UI (satellite cards with telemetry)
  • Satellite offline detection with configurable timeout
  • Alerts: offline, low storage, overheating, high packet loss

P2.5 — User Authentication & Authorization ✅

  • Email/password registration with bcrypt password hashing
  • JWT cookie-based sessions (httpOnly, secure, 7-day expiry)
  • Login and registration pages with form validation
  • Auth-gated routing (unauthenticated users see only login/register)
  • Auto-login after registration
  • Logout with cookie clearing
  • /api/auth/me endpoint returns current user + tenant memberships
  • Blocked user detection at login time

P2.6 — Multi-Tenant Membership ✅

  • Users are global (no tenant_id on users table)
  • Multi-tenant membership via tenant_members join table
  • Per-tenant roles: viewer, member, admin, owner
  • Tenant selector in sidebar for users with multiple memberships
  • Platform admin with cross-tenant access (via DB flag or env var)
  • Platform admin hidden from tenant member lists
  • Self-service tenant creation (gated by platform setting)

P2.7 — Invite System ✅

  • Admin-generated single-use invite links
  • Invite scoped to tenant + role (viewer, member, admin)
  • 72-hour default expiry (configurable per invite)
  • Registration via invite bypasses self-registration setting
  • Invite link displayed with copy-to-clipboard
  • Pending invite list with revoke capability
  • Invite validation before user creation (no orphan accounts)

P2.8 — Admin Page ✅

  • Platform tab: global user management table with search
  • Platform tab: block/unblock users
  • Platform tab: delete users with modal confirmation
  • Platform tab: reset password with modal dialog
  • Platform tab: platform admin accounts shown as read-only
  • Platform tab: platform admin self-service password change
  • Tenants tab: member list with role display
  • Tenants tab: role change via dropdown (viewer/member/admin)
  • Tenants tab: remove member with modal confirmation
  • Tenants tab: invite link generation with role selection
  • Tenants tab: pending invites list with revoke
  • Tenants tab: tenant selector for multi-tenant admins

P2.9 — Settings Page ✅

  • Tenant settings: confidence thresholds (high, low, minimum) with sliders
  • Tenant settings: offline timeout (minutes)
  • Tenant settings: verification consensus (votes needed)
  • Tenant settings: satellite retention (hours)
  • Platform settings: self-registration toggle
  • Platform settings: tenant creation toggle
  • Platform settings: invite links toggle
  • Settings API: /api/settings/tenant (GET/PUT)
  • Settings API: /api/settings/platform (GET/PUT)

P2.10 — UI Enhancements ✅

  • WebSocket connection status indicator (Live / Reconnecting...)
  • Toast notifications for action feedback (success and error)
  • Modal confirmations for all destructive actions
  • User menu in sidebar (name, email, platform admin badge, sign out)

P2.11 — On-Device Audio Pre-Filtering ✅

  • Adaptive noise floor + spectral peak SNR detection (replaced RMS + bird-band frequency ratio)
  • All filter settings configurable from hub tenant settings (filter_enabled, filter_min_rms, filter_peak_snr, filter_bird_band_check, filter_noise_floor_alpha)
  • Settings pushed to satellites via MQTT config channel when changed
  • Satellites report noise_floor_rms in heartbeat telemetry
  • Default thresholds: min RMS 0.003, peak SNR 1.5, noise floor alpha 0.02
  • Reduces unnecessary uploads and processing

P2.12 — Custom Watchlists for Species Alerts ✅

  • Per-tenant watchlist_species array in tenant settings
  • Alerts triggered when watchlisted species are detected

P2.13 — Web Container Auth Refactor ✅

  • nginx no longer injects HUB_INTERNAL_API_KEY on API requests
  • All user auth flows through JWT cookies (set by login, forwarded by browser)
  • Internal API key retained for backwards-compatible machine access only

Phase 3 — Community

Features that enable collaboration and data sharing.

P3.1 — Collaborative Verification ✅

  • Backend voting endpoint (confirm / reject / unsure)
  • Configurable consensus threshold (default: 3, adjustable in tenant settings)
  • Verification review queue in web UI (card-based with spectrogram + audio)
  • Keyboard shortcuts (1=Yes, 2=No, 3=Unsure, arrows=navigate)
  • Auto-advance to next unvoted detection
  • Progress bar and vote tracking
  • Verification statistics and user leaderboard

P3.2 — Scheduled Recording Profiles ✅

  • Profile definitions (dawn_chorus, night_migration, low_power, continuous)
  • Sunrise/sunset calculation from GPS coordinates (NOAA solar algorithm)
  • Recording scheduler pauses capture outside active windows
  • Hub → satellite profile + filter settings push via MQTT
  • Hub pushes config to all online satellites when tenant settings change
  • Satellites apply config on reconnect
  • Heartbeat interval configurable from tenant settings (default 30s)
  • Per-satellite profile assignment in web UI
  • Schedule preview on satellite cards (resolved sun times + windows)
  • Schedule display in mobile app (read-only)
  • Night migration uses sunset-relative times (adapts to season)

P3.3 — Data Federation & Export (partial)

  • CSV/JSON export of detections (filterable, downloadable from UI)
  • eBird checklist generation
  • xeno-canto recording submission
  • iNaturalist observation export
  • Scheduled automatic exports

P3.4 — Multi-Language Bird Names ✅

  • Species translation service (38 languages, loaded from BirdNET label files)
  • SpeciesName component with SpeciesI18nProvider context
  • User preferences for primary and secondary display languages

P3.5 — Multi-Hub Workers (partial)

  • Worker registration and heartbeat
  • [~] Contributed worker onboarding (partial: worker dashboard shows connected workers)
  • Worker performance metrics (avg processing time, queue stats)
  • Worker dashboard in web UI with IP-based worker dedup
  • GPU vs CPU worker differentiation

Phase 4 — Scale & Polish

P4.1 — Phone App ✅

  • Cross-platform mobile app (Capacitor)
  • Android phone satellite app with in-app registration, telemetry, GPS
  • Native AudioRecord plugin for Android
  • Background mode for continuous recording
  • In-app log viewer
  • Tap-to-record status bar (recording rectangle is the toggle)
  • Default state: paused (not recording)
  • GPS gate: requires location before recording (auto or manual)
  • Clear GPS/manual location toggle
  • GPS auto-update toggle with configurable interval (30s–10min)
  • Keep screen on toggle
  • Settings redesigned as full page (not popup) with verbose chunk logging toggle
  • All hub filter settings received and applied via MQTT
  • Heartbeat keepalive (configurable interval, default 30s, QoS 1)
  • Schedule display (read-only, from hub)
  • Storage telemetry via Storage API
  • [~] Push notifications for rare species (WebSocket toasts implemented, no native push yet)

P4.2 — Analytics Page ✅

  • Detection trends over time (daily trend chart)
  • Hourly species activity chart
  • Top species summary

P4.3 — User Preferences ✅

  • Per-user language settings (primary_language, secondary_languages on users table)
  • Date format preferences: primary display + secondary tooltip
  • Date format options: relative, short, medium, long, ISO 8601
  • User preferences API (PUT /api/auth/me/preferences)
  • UserPreferences page in web UI

P4.4 — Species Images ✅

  • Bird thumbnails from Wikipedia (Wikimedia Commons API)
  • Background download queue (max 2 concurrent, configurable delay)
  • Wikimedia rate limiting: sliding 1-hour window, auto-throttle at 80%, pause at 95%
  • Unauthenticated: 500 req/hr; with OAuth 2.0 token: 10,000 req/hr
  • Wikimedia OAuth token and contact email configurable in Platform Settings
  • Respects Retry-After header on 429 responses
  • Permanent storage in MinIO (no TTL expiry)
  • Proxy endpoint serves from MinIO (Wikipedia never hit at request time)
  • Placeholder display while downloading (pulsing indicator)
  • Lightbox with full-size image + Wikipedia link
  • Platform admin cache management (stats, retry failed, clear)

P4.5 — MQTT Heartbeat ✅

  • Lightweight keepalive every 30 seconds (QoS 1)
  • Decouples online status from audio chunk sending
  • Heartbeat state visible on satellite cards (Recording, Paused, Scheduled off)
  • Telemetry upgraded from QoS 0 to QoS 1 for reliability
  • Satellite version reported in heartbeat, stored on satellites table
  • Web UI shows amber warning when satellite version differs from hub version

P4.6 — Responsive Web UI ✅

  • Hamburger menu on mobile (< 768px)
  • Sidebar slides in as overlay with backdrop
  • Responsive sidebar (scrollable, hidden scrollbar)
  • Single-column card grids on mobile
  • Scrollable tables
  • Stacked filters

P4.7 — Versioning & Documentation ✅

  • Unified semver across all packages (version-bump script syncs all package.json + pyproject.toml)
  • Version displayed in web UI sidebar and mobile app header
  • Documentation site (Express + markdown-it) with version switcher
  • CHANGELOG.md with structured release notes
  • Release protocol documented
  • Design system documented (buttons, badges, color palette)

P4.8 — Species Catalog & Detection Map ✅

  • Species catalog page with sortable cards and rich stats
  • Per-species: detection count, confidence, first/last seen, satellite count
  • Sort: most detected, most recent, alphabetical, confidence, rarest, most satellites
  • Detection map with Leaflet.js + OpenStreetMap
  • Satellite markers colored by status, sized by detection count
  • Popup with detection stats and top 5 species

P4.9 — Account & User Management ✅

  • Account page: edit name, email, change password, language preferences
  • Account self-deletion with confirmation (blocks sole tenant owners)
  • Login rate limiting (20 attempts/IP/15min)
  • Account lockout after 5 failed attempts (15min)
  • Last login tracking (timestamp + IP)
  • Audit log: all admin and user actions recorded
  • Audit Log page for platform admins (paginated, color-coded)
  • Tenant ownership transfer
  • User search filters (status, tenant)
  • Platform admin auto-persisted from PLATFORM_ADMIN_EMAILS env var
  • Auth resolves live user data from DB (not stale JWT)

P4.10 — Design System ✅

  • Unified CSS classes for all buttons (btn-primary, btn-secondary, btn-danger, btn-ghost)
  • Button sizes: btn-sm (tables), default (modals), btn-lg (forms)
  • Hover, disabled, active states in CSS (no inline JS)
  • Unified badge classes (online, offline, degraded, muted, accent, rare, danger)
  • Responsive layout with mobile hamburger menu
  • Version display in sidebar

P4.12 — Advanced Analytics (remaining)

  • Migration pattern visualization across satellite network
  • Biodiversity indices per location

P4.13 — Per-Satellite Configuration Overrides

  • Per-satellite filter settings override (RMS, SNR, bird-band check, noise floor alpha)
  • Configurable from hub satellite detail panel or directly on the satellite device
  • Tenant defaults apply unless satellite has explicit overrides
  • Visual indicator on satellite card when using custom overrides

P4.14 — Model Management

  • Support multiple BirdNET model versions
  • A/B testing of model versions
  • Custom fine-tuned model deployment
  • Model performance comparison dashboards