Atelier

A unified creative console for KiraAI. One plugin page exposes the four generative capabilities already configured in your KiraAI instance — text, image, speech synthesis, and speech recognition — without any extra provider setup.

Features

• Chat — conversational text generation backed by the default LLM provider, with multi-turn context and a configurable system prompt.
• Image — text-to-image generation backed by the default image provider, with a hover-to-save gallery.
• Voice — text-to-speech backed by the default TTS provider, with an inline custom audio player (play/pause, seek, time).
• Listen — speech-to-text backed by the default STT provider, recording from the microphone with a live input-level meter.

Capabilities are switched through a segmented tab control at the top of the page; only one view is active at a time. The page follows a minimalist tech design with light and dark themes and syncs with the parent WebUI theme.

Requirements

Every capability relies on the default providers configured in KiraAI (Settings → Providers). A capability that has no corresponding default provider will simply return an error when used; everything else keeps working.

Capability	Provider
Chat	Default LLM
Image	Default image model
Voice	Default TTS
Listen	Default STT

Configuration

The plugin exposes a single option in the WebUI plugin settings:

• System prompt — optional system prompt prepended to every chat conversation. Leave empty to use the built-in default.

API

All endpoints are registered by the plugin and reachable at /api/plugin/kira-ai-plugin-atelier/…. Every endpoint returns { "ok": true, … } on success and { "ok": false, "error": "…" } on failure.

Method	Path	Body	Returns
GET	/status	—	{ available: { llm, image, tts, stt } }
POST	/chat	{ messages: [{ role, content }] }	{ text }
POST	/image	{ prompt }	{ image, mime }
POST	/tts	{ text }	{ audio, mime }
POST	/stt	{ audio } (data URL or base64)	{ text }

/chat accepts user and assistant messages; a system prompt is prepended server-side and unknown roles or empty content are dropped.

Project structure

kira-ai-plugin-atelier/
├── main.py           # Plugin entry: API endpoints + page registration
├── manifest.json     # Plugin metadata and locales
├── schema.json       # WebUI settings schema (system prompt)
├── icon.svg
└── web/              # Static frontend served at the plugin page
    ├── index.html
    ├── style.css
    ├── lucide.min.js # Vendored icon library (no CDN dependency)
    └── js/
        ├── app.js    # Entry: tabs, theme, module wiring
        ├── api.js    # Typed wrappers around the plugin API bridge
        ├── chat.js   # Chat capability
        ├── image.js  # Image capability
        ├── voice.js  # TTS capability
        ├── hearing.js# STT capability (MediaRecorder + level meter)
        ├── player.js # Custom audio player
        ├── theme.js  # Theme handling (follows WebUI, local override)
        └── ui.js     # Small shared DOM helpers

Development notes

• The frontend is plain HTML/CSS/ES modules with no build step — edit files under web/ and reload the plugin page.
• The page runs inside the WebUI iframe and communicates through window.PluginPageContext (/plugin-bridge.js), which provides theme context and same-origin API calls.
• Icons come from the vendored Lucide UMD build (web/lucide.min.js), so the page works fully offline; no external requests are made.
• Changes to main.py require restarting KiraAI (or reloading plugins) to take effect.