CEDAR turns a Dell Latitude 5430 Rugged into a fully local, privacy-first AI workstation. The app ships with a neon-styled PySide6/QML interface, orchestrates Ollama-hosted models, keeps long-term memory on-device, automates Windows tasks with guard rails, and bridges IoT environments — all without mandatory cloud calls.

• Local LLM Hub – REST/CLI Ollama client, dropdown model switcher, and raw ollama run "model" command entry. Missing models trigger pulls and progress feedback.
• Rich Tool Belt – DuckDuckGo search + trafilatura extraction, summarized via the LLM; Windows automation for apps/input/files/system info routed through confirmation policies; IoT control through MQTT + Home Assistant adapter with registry cache; camera preview, OCR, optional ONNX detection.
• Voice-First Interaction – “Cedar” wake word (openwakeword), VAD (webrtcvad), faster-whisper ASR, Piper TTS with optional Coqui XTTSv2 voices, live RMS pulse driving animations.
• Responsive GUI – PySide6 + QML neon theme, Lottie/GIF animations for idle/listening/thinking/speaking/alert states, tool cards, logs, permissions editor, configuration wizard, and voice pack rights acknowledgement.
• Memory + Safety – SQLite conversational memory, Ollama embeddings + Chroma vector store with fallback to sentence-transformers, JSON audit logs via loguru, policy-driven confirmations with two-step approvals for risky actions, sandboxed file access.
• Packaging Pipeline – PowerShell bootstrap installer (scripts/setup.ps1), launcher (scripts/start.ps1), PyInstaller bundler (scripts/package.ps1), and clear requirements pinned to Windows-friendly wheels.
• CI-Friendly – --headless mode stubs mic/camera, tests cover Ollama client, tools, and an integration path from wake word trigger to assistant reply.

• Dell Latitude 5430 Rugged
• Windows 11 Pro (64-bit), 64 GB RAM, 1 TB SSD
• Local administrator rights for first-time setup (winget installs)
• Working microphone/speakers/camera (or headless mode for testing)

1. Open PowerShell as Administrator and allow local scripts if needed:

   Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser

2. Clone or copy this repository to C:\Users\<you>\cedar.
3. Bootstrap dependencies and environment:

   .\scripts\setup.ps1

4. Launch CEDAR (starts Ollama serve if required, then the GUI):

   .\scripts\start.ps1

The first launch presents the setup wizard to confirm mic/camera, model selection/pulls, voice engine preferences, MQTT/Home Assistant configuration, and permission policies.

• Chat Panel – Type or use the wake word “Cedar.” Transcripts appear in real time; tool cards summarize searches, automations, or IoT actions.
• Model Panel – Dropdown lists ollama list. Enter raw commands (e.g., ollama run "llama3.1:8b-instruct") to pull/switch models; selections persist to config.json.
• Voice Panel – Choose input/output devices, toggle wake word and VAD thresholds, manage voice packs (rights checkbox required).
• IoT Panel – Configure MQTT broker and Home Assistant token, inspect device tree, and toggle entities with policy enforcement.
• Permissions Panel – View/edit policy.yaml via GUI editor; two-step confirmations apply for high-impact tasks.
• Logs Panel – Live JSON logs rendered in a readable format with download link for diagnostics.

• SQLite database at data/memory.db stores turns, tool calls, and running summaries.
• Chroma vector store in data/vector_store, using Ollama embeddings or sentence-transformers fallback.
• Policy configuration in policy.yaml; runtime preferences in config.json.
• Logs written to logs/cedar.jsonl (JSON-serialized, secrets redacted).
• Sandbox roots configured in policy.yaml plus config.json overrides.

• Camera snapshots saved to data/captures; cleanup handled by sandbox constraints.
• OCR via pytesseract (ensure Tesseract installed — handled by setup script).
• Optional ONNX YOLO detection requires placing a .onnx model in assets/models (documented in GUI).
• Default Piper voice en_US-lessac-medium.onnx downloaded at setup; additional voice packs go into voice_packs/ and require rights confirmation.

Generate a distributable one-folder build:

.\scripts\package.ps1

Artifacts land in dist\CEDAR\ with assets and QML bundled.

• Activate headless mode via config.json ("headless": true) or environment variable:

  $env:CEDAR_HEADLESS = "1"
  .\scripts\start.ps1

• Run automated tests:

  venv\Scripts\Activate.ps1
  pytest

Headless mode queues mock wake word events, reads audio/text fixtures, and disables camera preview for CI pipelines.

• Default code is distributed under the MIT license.
• Imported voice packs must respect rights; enabling them requires user confirmation.
• External services (Home Assistant, MQTT brokers) should be configured responsibly.

Enjoy operating CEDAR completely on-device! If extending the project, consult agents.md for architectural guidance.