Ultimate Voice Bridge (UVB) — KnightBot AI Assistant

A modular, extensible AI-Human interface suite with a stunning galaxy-themed UI

Jack of all trades, master of... also, all trades.

Overview

The Ultimate Voice Bridge is a web-based AI assistant interface that bridges the gap between the immense power of local Large Language Models and the diverse ways humans communicate. Built around the KnightBot AI Assistant, it provides a unified, beautiful frontend for multi-modal AI interaction — text chat, voice analysis, image/video understanding, podcast creation, and persistent memory.

Designed to run on high-end consumer hardware with local LLMs (LM Studio, Ollama), the UVB keeps everything on your machine — no cloud dependencies required.

Current Local Cockpit Status

UVB is now wired for the local KnightBot stack:

Chat uses an OpenAI-compatible local model endpoint, defaulting to http://127.0.0.1:8003/v1.
Dashboard voice recording uses local Faster Whisper at http://127.0.0.1:8001/v1/audio/transcriptions.
Spoken replies use local Kokoro TTS at http://127.0.0.1:8880/v1/audio/speech.
The top-right health badge checks LLM, STT, TTS, Qdrant, and reranker services.
Settings can export/import the UVB model and voice profile for quick recovery by another agent.
Runtime model/voice defaults are saved under ignored .uvb/ files so local workers can share the dashboard configuration.
A local realtime voice-agent sidecar is staged at ws://127.0.0.1:8765/live for the Pipecat v1 / SmallWebRTC-first / LiveKit-later path.

Fast Local Launch

cd D:\UVB-KnightBot-Export
.\scripts\start-uvb.ps1

To recreate the desktop shortcut:

cd D:\UVB-KnightBot-Export
.\scripts\create-desktop-shortcut.ps1

The shortcut opens UVB at http://localhost:3010 and starts the realtime voice sidecar and Telegram worker unless -SkipVoiceAgent or -SkipTelegram is used.

Realtime Voice Sidecar

UVB keeps the cockpit frontend in Next.js and runs realtime voice behind it as a local sidecar. The first sidecar implementation uses the stable local providers already running on this machine:

Faster Whisper for STT
vLLM/OpenAI-compatible chat on 8003
Kokoro for TTS

The sidecar exposes a WebSocket bridge at ws://127.0.0.1:8765/live. This is the first working runway toward Pipecat, Parakeet Realtime EOU, Chatterbox Turbo, VibeVoice-Realtime, and later LiveKit transport.

Install sidecar dependencies once:

cd D:\UVB-KnightBot-Export
.\services\voice-agent\install.ps1

Run only the sidecar:

cd D:\UVB-KnightBot-Export
bun run voice-agent

Optional Pipecat runtime install:

.\services\voice-agent\install.ps1 -WithPipecat

The current sidecar mode is baseline-websocket-pipecat-v1-ready: it uses the working Faster Whisper, vLLM, and Kokoro services today while preserving a stable contract for Pipecat v1, SmallWebRTC, Parakeet Realtime EOU, MOSS-TTS-Nano, MOSS-TTSD, Chatterbox Turbo, VibeVoice-Realtime, and LiveKit transport upgrades.

Telegram Bridge

Telegram secrets stay in .env.local, which is intentionally ignored by git. Use .env.example as the safe template.

cd D:\UVB-KnightBot-Export
notepad .env.local

Required values:

TELEGRAM_BOT_TOKEN: token from BotFather.
TELEGRAM_ALLOWED_CHAT_ID: the personal chat ID allowed to control UVB.
UVB_PUBLIC_URL: usually http://127.0.0.1:3010 for local polling.
TELEGRAM_API_ORIGIN: Bot API origin, defaulting to https://api.telegram.org; set this to a local Bot API server for large file downloads.
TELEGRAM_FILE_ORIGIN: file download origin, defaulting to TELEGRAM_API_ORIGIN.
TELEGRAM_CLOUD_DOWNLOAD_MAX_MB: cloud Bot API download ceiling (20 MB by default).

Optional Telegram voice controls:

TELEGRAM_SEND_TEXT_REPLIES: keep text replies on while also sending audio (true by default).
TELEGRAM_TEXT_CHUNK_CHARS: approximate text size per Telegram message for long replies (3600 by default).
TELEGRAM_SEND_TTS_REPLIES: send Kokoro-generated audio replies back to Telegram (true by default).
TELEGRAM_TTS_VOICE: Kokoro voice used for Telegram replies, defaulting to UVB_TTS_VOICE or af_nova.
TELEGRAM_TTS_CHUNK_CHARS: approximate text size per audio part for long replies (4200 by default, roughly a three-minute target).
TELEGRAM_TTS_MAX_PARTS: maximum number of sequential audio parts per reply (6 by default).
TELEGRAM_DOCUMENT_MAX_CHARS: maximum text-document characters routed into UVB (120000 by default).
TELEGRAM_VIDEO_MAX_MB: maximum Telegram video size routed locally (500 MB by default).
TELEGRAM_VIDEO_FRAME_COUNT: number of sampled frames to send as a video storyboard (6 by default, capped at 12).
TELEGRAM_VIDEO_FRAME_MAX_WIDTH: maximum width for sampled video frames (960 by default).

The worker can route Telegram text, voice/audio, videos, photos, image documents, and text documents into UVB. Photos and image documents are forwarded to /api/chat as OpenAI-style image_url content, so the active local model still needs vision support for detailed image understanding. Video routing uses local ffmpeg/ffprobe to extract an audio transcript plus an ordered multi-frame storyboard sampled across the clip. Text-like documents such as .txt, .md, .json, .csv, logs, and common code/config files are downloaded and wrapped into the prompt with clear file delimiters.

Telegram's hosted Bot API only allows bot file downloads up to 20 MB. To process larger videos, run a local Telegram Bot API server and point TELEGRAM_API_ORIGIN/TELEGRAM_FILE_ORIGIN at it; TELEGRAM_VIDEO_MAX_MB is UVB's local processing guardrail after Telegram download access is available.

Run only the Telegram worker:

cd D:\UVB-KnightBot-Export
bun run telegram

Features

KnightBot Chat

Multi-modal input: text, local voice transcription, staged image/video/file attachments
Thread-based conversation management with auto-naming
OpenAI-compatible model bridge with configurable backend
Typing indicators, message actions (copy, regenerate, bookmark)
Kokoro spoken replies with configurable voice and volume
Live Voice mode through the local sidecar for turn-level STT → LLM → TTS with latency metrics

Voice Analysis

Real-time recording and file upload (WAV, MP3, FLAC, OGG, M4A)
Scientific metrics: fundamental frequency, spectral centroid, RMS energy, zero-crossing rate, spectral rolloff, MFCC coefficients
Waveform visualization with animated bars
Voice quality assessment: jitter, shimmer, HNR
Audio restoration and noise reduction tools

Media Studio

Image Captioning: detailed descriptions, object detection, scene classification, OCR, dominant color analysis
Video Understanding: scene segmentation with timestamps, key frame analysis, audio track analysis, transcription
Drag-and-drop upload with animated analysis states

Podcast Studio

Up to 6 individually configurable seats
Voice profile selection: default, presets, or custom zero-shot clones
Zero-shot voice cloning: 3-5 second sample requirement
Mix controls: master volume, output format, noise gate
Real-time voice visualization per seat

Memory Bank (RAG)

Persistent local memory with semantic search
1536-dimension vector embeddings for retrieval
Category filters: conversation, knowledge, context, preference
Search by title, content, and tags
RAG stats: dimensions, recall rate, retrieval latency

Settings

Profile: display name, email, password management
Voice & Audio: TTS/STT engine selection, speech rate, barge-in toggle
Live Voice: sidecar URL, Pipecat/SmallWebRTC transport staging, VAD, STT/TTS provider slots, MOSS candidate endpoints, voice profile, and voice identity prompt
Appearance: theme (Galaxy Dark, Deep Space, Neon Night), accent colors, particle effects
AI Settings: model backend (LM Studio/Ollama/API), context window, temperature, CoT, RAG
Security: local-only data, AES-256 encryption, auto-save, telemetry toggle
Notifications: configurable alerts for tasks, voice, system events

Design System

Visual Language

Galaxy particle background: canvas-based animation with mouse-reactive particles and glow connections
Glass panels: frosted blur with neon borders
Animated effects: glow orbs, scan lines, status pulses, laser-sweep loading
3D typography: layered text shadows for display headings

Color Palette

Token	Hex	Usage
Neon Green	`#39ff14`	Primary accent, active states, glow effects
Steel Blue	`#4a6fa5`	Secondary accent, gradients, icons
Deep Teal	`#0d4f4f`	Panels, hover states
Royal Purple	`#4a0e78`	Gradients, user avatar
Matte Black	`#0a0a0a`	Base background
Dark Gray	`#141418`	Cards, surfaces
Accent Yellow	`#f5a623`	Warnings, highlights
Accent Orange	`#ff6b35`	Contrast elements

Typography

Font	Role	Source
Orbitron	Display / Headings	Google Fonts
Inter	Body text	Google Fonts
JetBrains Mono	Code / Metrics	Google Fonts

Tech Stack

Technology	Version	Purpose
Next.js	16.x	React framework with App Router
React	19.x	UI library
TypeScript	5.9.x	Type-safe development
Tailwind CSS	4.x	Utility-first CSS (CSS-first config)
Zustand	5.x	Client state management
Framer Motion	12.x	Animations and transitions
Heroicons	2.x	Primary icon set
Lucide React	1.x	Secondary icon set
Headless UI	2.x	Accessible UI primitives
Bun	Latest	Package manager

Project Structure

src/
├── app/                           # Next.js App Router
│   ├── layout.tsx                 # Root layout (Orbitron, Inter, JetBrains Mono)
│   ├── page.tsx                   # Main dashboard shell (section-based SPA)
│   ├── globals.css                # Tailwind v4 @theme + UVB design tokens
│   ├── api/health/route.ts        # Health check endpoint
│   ├── chat/ChatInterface.tsx     # Chat UI with thread management
│   ├── voice-analysis/            # Voice analysis page
│   ├── media/                     # Media studio (image + video)
│   ├── podcast/                   # Podcast creation suite
│   ├── memory/                    # RAG memory bank
│   └── settings/                  # User settings (6 tabs)
├── components/
│   ├── animated/
│   │   ├── GalaxyBackground.tsx   # Canvas particle system
│   │   ├── UIEffects.tsx          # GlowOrb, ScanLine, FloatingDot
│   │   └── VoiceVisualizer.tsx    # Real-time audio bars
│   ├── layout/
│   │   ├── Sidebar.tsx            # Collapsible navigation
│   │   └── Header.tsx             # Top bar with status/search
│   └── ui/                        # Reusable UI components
├── stores/
│   └── appStore.ts                # Zustand global state
└── lib/                           # Utilities

Getting Started

Prerequisites

Bun installed
Node.js 20+

Installation

# Clone the repository
git clone https://github.com/YOUR_USERNAME/uvb-knightbot.git
cd uvb-knightbot

# Install dependencies
bun install

# Start development server
bun dev

Open http://localhost:3000 in your browser.

Available Scripts

bun dev          # Start development server
bun build        # Production build
bun start        # Start production server
bun lint         # Run ESLint
bun typecheck    # Run TypeScript type checking

Architecture Decisions

Single-page app shell: All sections render within one page.tsx using Zustand activeSection state — no page transitions, instant switching
Component-per-section: Each feature is a self-contained component with local state
CSS-first Tailwind v4: Design tokens defined in @theme block, not JS config
No backend in Phase 1: All data is mock/simulated — designed for easy integration with local LLM APIs

Roadmap

Phase 1 — Frontend Foundation (Complete)

All 6 main sections with full UI
Galaxy particle background with animations
Zustand state management
Context-aware chat responses
Design system with Tailwind v4 tokens

Phase 2 — Backend Integration

LM Studio API connection for live LLM chat
Web Audio API for real-time voice recording/analysis
Vision model integration for image/video processing
Drizzle + SQLite for persistent storage
WebSocket streaming for real-time AI responses

Phase 3 — Advanced Features

User authentication with password protection
Thread branching and conversation trees
Voice cloning pipeline integration
Podcast recording with multi-track export
RAG pipeline with local vector store
Browser automation integration

License

This project uses free/open source code, models, APIs, and solutions designed to bridge the gap between humanity and AI.

The UVB & KnightBot would be the link. The system that brings it all together.

Truly an extensive and elegant swiss army knife in AI-Human interface evolution.

Name		Name	Last commit message	Last commit date
Latest commit History 95 Commits
.kilocode/rules/memory-bank		.kilocode/rules/memory-bank
docs		docs
public/avatar		public/avatar
scripts		scripts
services		services
src		src
.env.example		.env.example
.gitignore		.gitignore
AGENTS.md		AGENTS.md
README.md		README.md
UVB-BACKEND-INTEGRATION-PLAN.md		UVB-BACKEND-INTEGRATION-PLAN.md
bun.lock		bun.lock
eslint.config.mjs		eslint.config.mjs
next.config.ts		next.config.ts
package.json		package.json
postcss.config.mjs		postcss.config.mjs
setup.ps1		setup.ps1
tsconfig.json		tsconfig.json

Folders and files

Latest commit

History

Repository files navigation

Ultimate Voice Bridge (UVB) — KnightBot AI Assistant

Overview

Current Local Cockpit Status

Fast Local Launch

Realtime Voice Sidecar

Telegram Bridge

Features

KnightBot Chat

Voice Analysis

Media Studio

Podcast Studio

Memory Bank (RAG)

Settings

Design System

Visual Language

Color Palette

Typography

Tech Stack

Project Structure

Getting Started

Prerequisites

Installation

Available Scripts

Architecture Decisions

Roadmap

Phase 1 — Frontend Foundation (Complete)

Phase 2 — Backend Integration

Phase 3 — Advanced Features

License

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages