The chat interface is the operator layer of the platform — two separate AI providers (Chat and Image), a TOOLCALL dispatch pipeline, persistent history, and a mode system that switches the persona on the fly. Lux runs inside it. So does everything else.
Before every message reaches the AI, the prompt builder assembles: personality, persistent memories, active notes, all tool catalogs with instructions, conversation history, and any file or voice context. The AI always has the full picture.
When the AI needs live information — hours, client records, task status, email content — it outputs a structured TOOLCALL. The response dispatcher intercepts it, executes the tool, injects the result, and the AI answers with real data.
The dispatcher handles parsing, routing, execution, and re-injection automatically. Tool results are returned in a clean format — the AI formats the response, not the raw JSON. You always see a human-readable answer.
Each mode has a defined provider fallback chain — if OpenAI is slow or down, Lux automatically routes to Grok, then Venice. Chat, image, and video each have separate fallback orders configured per mode.
Every message is stored per-user and per-mode in the WordPress database. Lux remembers the full conversation thread and uses it to maintain context across a long working session — not just the last few messages.
Upload a document and ask Lux to summarize it. Share an image and ask what to do with it. Speak instead of type and she switches to voice-aware response patterns. All of it flows through the same chat pipeline.
Chat providers
Image providers
Both dropdowns live in the chat topbar and are saved per user per mode. Switch image to Promptchan while keeping OpenAI for chat — settings persist across sessions, stored encrypted per user.
We built this from scratch. We can build the right version for your workflow.
Stark