Add NexusAI: universal MCP service (TTS/STT/image/video/music/web-search)

Monorepo-lite with shared @nexusai/core and @nexusai/mcp-server (stdio).
Provider-agnostic tools with explicit provider selection; NordRouter adapter
(media generate/poll/download, estimate, upload, models) and search via
perplexity/sonar. Local STT via faster-whisper uv sidecar. SQLite journal of
every generation for usage stats and future admin. 10 MCP tools.

NexusAI/.env.example

@@ -0,0 +1,35 @@
+# --- NordRouter ---
+# API key from your NordRouter account (starts with sk-)
+NORDROUTER_API_KEY=sk-your-key-here
+NORDROUTER_BASE_URL=https://nordrouter.com
+
+# --- NexusAI general ---
+NEXUS_DEFAULT_PROVIDER=nordrouter
+# Where generated media files and the SQLite journal are stored
+NEXUS_MEDIA_DIR=./media-output
+# Optional: custom SQLite path (defaults to <media dir>/nexus.sqlite)
+# NEXUS_DB_PATH=./media-output/nexus.sqlite
+# Max bytes for embedding image/audio directly in MCP results (default 10MB)
+NEXUS_EMBED_MAX_BYTES=10485760
+# Log level: debug | info | warn | error (logs go to stderr)
+NEXUS_LOG_LEVEL=info
+
+# --- Local STT (faster-whisper via uv sidecar) ---
+# Launcher: "uv" (recommended, auto-installs deps) or a python interpreter path
+NEXUS_STT_PYTHON=uv
+# Whisper size (tiny|base|small|medium|large-v3) OR an absolute path to a
+# pre-downloaded CTranslate2 model directory
+NEXUS_STT_MODEL=small
+NEXUS_STT_DEVICE=auto        # auto | cpu | cuda
+NEXUS_STT_COMPUTE=auto       # auto | int8 | float16
+# Offline mode: require a locally present model (no downloads)
+# NEXUS_STT_LOCAL_ONLY=1
+# Cache dir for downloaded models
+# NEXUS_STT_DOWNLOAD_ROOT=./models
+# Optional custom sidecar path
+# NEXUS_STT_SIDECAR=/abs/path/to/sidecar/stt_server.py
+# HuggingFace mirror (for restricted networks)
+# HF_ENDPOINT=https://hf-mirror.com
+
+# --- Proxy (optional) ---
+# NEXUS_HTTP_PROXY=http://proxy:8080