Integrations

Local-first AI companion for daily individual use. Casual conversational tone. Voice, vision, Spotify, maps, app launcher, and desktop automation all enabled. No authentication required.

luna_variant=personal · Start with: luna dev

Professional team assistant. Multi-user JWT authentication, sliding-window rate limiting, Slack/Telegram/Discord messaging channels. No Spotify, no desktop launchers — focused on productivity and Q&A.

luna_variant=business · Start with: luna docker:business

Runs models locally on your machine. The default provider. No API key, no internet required. Supports GPU acceleration via CUDA or Metal.

llm_provider=ollama · ollama_base_url=http://localhost:11434 · ollama_model=qwen2.5:7b

Local model server with a GUI. Exposes an OpenAI-compatible endpoint. Good for trying new GGUF models without configuring Ollama.

llm_provider=openai-compatible · openai_base_url=http://localhost:1234/v1

Minimal OpenAI-compatible server. Maximum control over quantization and hardware utilisation.

llm_provider=openai-compatible · openai_base_url=http://localhost:8080/v1

Open-source local model manager with a desktop GUI. Exposes an OpenAI-compatible API on localhost.

llm_provider=openai-compatible · openai_base_url=http://localhost:1337/v1

Native Anthropic Messages API. Claude Sonnet and Opus models. Recommended for the Business variant — strong instruction following, long context, and function calling.

llm_provider=anthropic · anthropic_api_key=sk-ant-… · anthropic_model=claude-sonnet-4-5

Native Google AI REST API. Gemini 2.0 Flash by default. Exceptional long-context performance (1M tokens) and multimodal support.

llm_provider=google · google_api_key=AIza… · google_model=gemini-2.0-flash

Purpose-built LPU inference — LLaMA 3.3 70B at ~300 tok/s. Free tier available. Best for low-latency chat where speed matters more than context length.

llm_provider=groq · groq_api_key=gsk_… · groq_model=llama-3.3-70b-versatile

Command R+ with native RAG tooling. European data residency option. Strong at retrieval-augmented tasks and long structured outputs.

llm_provider=cohere · cohere_api_key=… · cohere_model=command-r-plus

Mistral Large, European-hosted and GDPR-friendly. Strong at coding and multilingual tasks.

llm_provider=mistral · mistral_api_key=… · mistral_model=mistral-large-latest

NVIDIA-hosted or self-hosted NIM endpoints using the OpenAI-compatible chat completions API.

llm_provider=nvidia-nim / nvidia_nim_base_url=https://integrate.api.nvidia.com/v1

GPT-4o, GPT-4o mini, o1, and future models via the official OpenAI API.

llm_provider=openai-compatible · openai_base_url=https://api.openai.com/v1 · openai_model=gpt-4o

One API key for every major model: Claude, Gemini, GPT-4o, Mistral, Llama, Command R, and hundreds more. Pay-per-token, no subscriptions.

llm_provider=openai-compatible · openai_base_url=https://openrouter.ai/api/v1

Set up a bot via @BotFather, point the webhook at your Luna instance. Supports markdown replies and per-user conversation history.

telegram_bot_token=… · Endpoint: POST /api/channels/telegram

Register a Discord application and set the Interactions Endpoint URL. Ed25519 signature verification built-in. Replies are capped at Discord's 2000-character limit.

discord_bot_token=… · discord_public_key=… · Endpoint: POST /api/channels/discord

Subscribe to the message.channels event in a Slack app. HMAC-SHA256 request signature verified on every inbound message.

slack_bot_token=xoxb-… · slack_signing_secret=… · Endpoint: POST /api/channels/slack

Post any JSON payload with user_id, user_name, and text fields — Luna replies with a plain-text reply field. Integrate with any platform.

POST /api/channels/webhook · {"user_id":"…","user_name":"…","text":"…"}

Receives push, pull request, issue, issue comment, and release events from any repo or organisation. HMAC-SHA256 signature verified on every request.

github_webhook_secret=… · Endpoint: POST /api/channels/github

OpenAI Whisper reimplemented in CTranslate2. Runs locally, significantly faster than the original. Recommended.

whisper_model=base (tiny / base / small / medium / large)

Offline speech recognition toolkit. Very fast, low memory footprint, good for wake-word detection on low-end hardware.

voice_runtime=vosk

Microsoft Edge neural TTS voices. High quality, 300+ voices and languages. Requires an internet connection — audio is synthesised by Microsoft's servers.

tts_provider=edge-tts

Fully offline TTS using the OS voice engine (SAPI5 on Windows, NSSpeechSynthesizer on macOS, eSpeak on Linux). No internet required.

tts_provider=pyttsx3 · tts_voice_index=0

Launch any app by name across Windows, macOS, and Linux. Say "open VS Code", "launch Figma", "open Discord". Discovery uses curated profiles, macOS Spotlight (mdfind), Linux .desktop files, Windows registry, and Start Menu — covering virtually any installed app.

luna dev → "open <app name>" to launch. Covers all categories below.

Get and set system volume (0–100), mute/unmute. Works on Windows (PowerShell/nircmd), macOS (osascript), and Linux (pactl/amixer). Say "set volume to 40", "mute", "unmute".

API: GET/POST /api/system/volume · POST /api/system/volume/mute|unmute

Get and set display brightness (0–100). Windows uses WMI, macOS uses the brightness CLI (brew install brightness), Linux uses brightnessctl or xbacklight. Say "set brightness to 60".

API: GET/POST /api/system/brightness

Instantly lock the workstation. Windows: rundll32 LockWorkStation. macOS: Command+Control+Q. Linux: loginctl lock-session or xdg-screensaver. Say "lock my screen".

API: POST /api/system/lock

Put the system to sleep / suspend. Windows: SetSuspendState. macOS: pmset sleepnow. Linux: systemctl suspend. Requires confirm permission by default.

API: POST /api/system/sleep · Permission: confirm

Read and write the system clipboard. Windows: PowerShell Get/Set-Clipboard. macOS: pbpaste/pbcopy. Linux: xclip, xsel, or wl-paste/wl-copy. Say "copy this to clipboard" or "what's in my clipboard?"

API: GET/POST /api/system/clipboard

Turn off the display without sleeping the system. Windows: SendMessage. macOS: pmset displaysleepnow. Linux: xset dpms.

API: POST /api/system/display/off

Report OS, RAM, battery percentage, and machine name. Cross-platform via WMI (Windows), sysctl (macOS), and /proc (Linux).

API: GET /api/system/info

Playback control, queue management, search, and now-playing display. Trigger with phrases like "play something chill" or "skip this".

Requires spotify_client_id and spotify_client_secret in .env.

Open URLs and read public page content. Playwright can be enabled for full browser automation including JavaScript-rendered pages.

Playwright: pip install playwright && playwright install chromium

Create, list, update, and complete tasks in Luna's local SQLite database. Proactive reminders fire when tasks are due.

DuckDuckGo HTML search — no API key required. Fires automatically when the LLM needs live information.

Switch the default Windows audio output device by name — headphones, speakers, virtual audio.

API: GET /api/system/audio-devices · POST /api/system/audio-device

Open interactive MapLibre map overlays with a location query.

Free, no-key weather API. Returns current conditions and forecasts for any coordinates. Default weather provider.

Set weather_lat, weather_lon, weather_city, weather_timezone in .env.

Live top headlines and category-filtered news articles for the dashboard. Falls back to RSS when no key is configured.

the_news_api=your-key

Stock quotes and market data. Used by default for equities in the dashboard — no API key required.

Cryptocurrency prices and market cap data. Used by default for crypto widgets — no API key required.

Extended stock market data and technical indicators. Used as an alternative data source when configured.

alpha_vantage=your-key

Full REST API coverage: steps, distance, calories, heart rate, HRV, all sleep stages, SpO2, breathing rate, skin temperature, and weight. OAuth2 setup takes about 2 minutes.

fitbit_client_id= · fitbit_client_secret= · GET /api/health/oauth/authorize/fitbit

Aggregates activity data from any Android wearable synced to Google Fit — Pixel Watch, Galaxy Watch, Wear OS devices, and more. Steps, calories, HR, weight, sleep via Google OAuth2.

google_fit_client_id= · google_fit_client_secret= · GET /api/health/oauth/authorize/google_fit

Best-in-class sleep staging, HRV, resting heart rate, readiness score, stress levels, and respiratory rate. Simple personal access token auth — no OAuth flow.

oura_api_key= — cloud.ouraring.com/user/api-tokens

Medical-grade smart scales (weight, BMI, body fat), blood pressure monitors, and sleep mats. The only platform that provides blood pressure readings.

withings_client_id= · withings_client_secret= · GET /api/health/oauth/authorize/withings

Detailed GPS workouts, VO2 Max, Body Battery (readiness), stress, sleep stages, and SpO2 via the garth library. Works with every Garmin wearable: Forerunner, Fenix, Venu, and more.

garmin_email= · garmin_password= · pip install garth

All HealthKit metrics from Apple Watch via webhook push. Apple has no public REST API, so Luna uses the free 'Health Auto Export' iOS app to push data on a schedule.

Endpoint: POST /api/health/webhook/apple · health_webhook_secret= in .env

Galaxy Watch and Galaxy Fit data via a compatible Android exporter app. Supports the same Health Auto Export JSON format as Apple Health.

Endpoint: POST /api/health/webhook/samsung · health_webhook_secret= in .env

HS256 JSON Web Tokens for multi-user access. Issue per-user tokens with expiry, rotate them without downtime, and list active users via the admin API.

jwt_secret=… · jwt_expiry_hours=720

Sliding-window in-memory rate limiter. Configurable per-minute limit and burst allowance. Returns 429 with Retry-After header when exceeded. No Redis required.

rate_limit_enabled=true · rate_limit_per_minute=60 · rate_limit_burst=20

Create, list, rotate tokens, and delete users via REST API. User store is a flat JSON file — no database migration needed.

GET/POST /api/admin/users · DELETE /api/admin/users/{id} · POST /api/admin/users/{id}/rotate-token

compose.business.yml includes a commented nginx block for TLS termination. Point a domain at the container and uncomment to enable HTTPS.

See the commented nginx service in compose.business.yml