# Orchestrator

The Orchestrator is your central command center for configuring the brain, personality, and capabilities of your Lyntaris agents. Because Lyntaris is a full-stack system linking Cloud LLMs to physical Unity hardware (Kiosks, Robots, VR), every setting on this page creates a cascading effect down to the physical client.

Prompts Tab

This tab configures the core language processing unit of the agent.

• LLM Provider & Model: Select the cognitive engine (e.g., Azure GPT-4o, Anthropic Claude 3.5). The Flowise backend uses this selection to instantiate the correct LangChain ChatModel class under the hood.
• AI Gender: Sets the demographic baseline for the avatar. This is critical as it dynamically filters the available voice models in the Speech tab (ensuring you cannot mistakenly assign a male voice to a female avatar) and heavily influences the baseline System Prompt.
• Main Instructions: The absolute core System Prompt. If you want the physical Kiosk to act as a stoic security guard, define it here. These instructions are concatenated with realtime data (like current time and user biometrics) before being sent to the LLM.
• Greet Prompts: When the local Unity client detects a user via its FaceTracker system, it sends an AssetsNotifyClient event to Flowise. Flowise intercepts this, bypassing the standard microphone STT wait, and immediately forces the LLM to generate a greeting based on either the "New visitors" or "Returning visitors" prompt defined here.

Tool Hints Tab

Lyntaris agents are distinct from standard chatbots because they can actively control their physical environment using Unity ICommand scripts. The Tool Hints tab governs these permissions.

• Command Toggles: By toggling tools (e.g., SwitchCamera, LoadFlowiseVideo), you are instructing the Flowise backend to append these specific JSON schemas to the LangChain bind_tools() array for the next conversation turn.
• Descriptions: LLMs decide when to execute a tool based solely on its description. If your Kiosk is not showing the map when asked, you must edit the "Action description" here to provide stronger semantic clues to the LLM.
• Architectural Link: When the LLM fires a tool, it outputs a strict JSON payload. Flowise intercepts this payload via the realtime WebSocket protocol, blocking the text response, and forwards the JSON directly to the Unity CommandExecuter. Only after Unity replies with a "Success" or "Failure" state does Flowise resume generating the spoken response.

Look Customization Tab

This tab bridges cloud content management with physical hardware deployment.

• Idle Media Slider: You can upload images and videos (10:16 or 9:16 aspect ratio) that will automatically rotate on the Kiosk screen when no user is present.
• Synchronization: Uploading media here stores it securely on the Lyntaris cloud. The moment you click Save, Flowise sends an assetsChanged WebSocket event. The remote UnityAppMonitor daemon receives this, silently downloads the new media files into the local C:\Deployments directory, and instructs the Unity IdleState to begin playing them—all without requiring a manual reboot of the physical unit.

Image Edit Instructions Tab

Lyntaris supports dynamic image generation and manipulation (e.g., "AI Photo Booths" on a Kiosk).

• Presets: Define explicit prompt-only or with-reference presets. This creates standardized buttons on the Unity UI.
• Manifest Sync: These presets are serialized into a manifest.json file. When the Unity client boots (or receives a sync event), it downloads this manifest to dynamically construct its user-facing UI for generating images.

Web Speech Pipeline

Lyntaris provides a "Wonder Widget" for bringing your Avatar to standard 2D websites (e.g., your corporate homepage).

• Embed Code Generation: This tab generates the raw <script type="module"> HTML embed code required to inject the Lyntaris webspeech-embed into a React, WordPress, or Vanilla JS site.
• Styling Overrides: Adjusting the background, colors, and data-rs-ar (aspect ratio) settings here automatically updates the generated embed code parameters, ensuring the Avatar iframe blends seamlessly with your corporate website CSS.