BUD-E is a friendly AI assistant that runs on your phone, tablet, or computer. It can answer questions, search the web, create documents and presentations, generate images and music, and even search through education curricula — all through a simple chat interface with voice support.
Think of it as a personal assistant for students and teachers: you talk or type, and BUD-E helps you learn, create, and explore. It connects to AI services through a middleware server (Admin Bud-E) that your school or organization controls — so your data stays private and costs stay manageable.
-green?style=for-the-badge&logo=android){kind=icon}
- Chat with LLM — Streaming responses via middleware (Gemini 2.5 Flash/Pro)
- Voice Input (ASR) — Microphone button, Whisper transcription
- Text-to-Speech (TTS) — Read responses aloud, per-message TTS toggle
- Auto Language Detection — Responds in the user's language
- Multilingual UI — German, English, French, Spanish (welcome screen, status messages, agent feedback)
- Conversation Branching — Regenerate button, branch navigation (← 1/3 →)
- Clickable Links — URLs in responses open in browser (url_launcher)
- Code Blocks — Syntax label, dark theme, copy button, inline edit mode
- Word (.docx) — Calibri font, blue headings, embedded images, automatic HTML-to-Markdown conversion
- PDF (.pdf) — Helvetica font, WinAnsi encoding (correct umlauts ä/ö/ü/ß, em-dash, smart quotes, €)
- PowerPoint (.pptx) — 4 layouts (Title / Image+Bullets / Image-only / Text-only), adaptive image sizing based on aspect ratio, Calibri font, slide numbers, image-first enforcement
- HTML (.html) — Inline CSS, base64-embedded images
- Markdown (.md), RTF (.rtf)
- File Chips — Generated documents shown as clickable icons with file type symbols
- Models: Gemini 3 Pro Image, Imagen 4.0, FLUX.2 (Pro/Max/Klein), nano-banana
- Aspect Ratios: square (1:1), landscape (16:9), portrait (9:16), photo (4:3), wide (21:9)
- Image Editing: Reference images for style transfer
- Image Registry: Every image gets a unique IMG_xxx ID
- Full Songs with Vocals — Up to 184s / ~3 minutes, MP3 output
- Lyrics Support —
[Verse],[Chorus],[Bridge],[Outro]tags - Two-Step Workflow: Draft prompt + lyrics → user confirms → generate
- Content Policy Error Handling — Clear message + rephrasing tips
- Brave Search — Web search with top-5 parallel scraping
- Web Scrape — Extract page content (5s timeout, parallel-friendly)
- Wikipedia — Article lookup (de/en)
- Weather — Current conditions via wttr.in
- News — Latest from tagesschau.de
- 1,548 pages from 32 Hamburg education plans indexed
- 4 school types: Grundschule, Stadtteilschule, Gymnasium Sek I, Studienstufe
- 17 subjects: Deutsch, Englisch, Mathematik, Biologie, Chemie, Physik, Informatik, Geschichte, Geographie, Sport, Religion, Wirtschaft, Psychologie, and more
- Bundled Assets — Index auto-extracted on first launch (works offline)
- Clickable PDF Links with page numbers (
URL#page=27) - Snippet Extraction with query term highlighting
- Safety Net — Auto-triggers search even if LLM skips the tool call
- Episodic Memory — Conversation summaries, auto-saved per session
- Semantic Memory — Long-term knowledge with trigger words and related concepts
- Working Memory — Active session context
- BM25 Memory Search — Keyword search across all memory files
- Priority-Based Context Construction — ~20K token budget, 1st/2nd order activation, German stopwords
- Sub-Agent with tool calling (web_search, web_scrape, generate_image, write_file, run_python, etc.)
- Agent Task Widget — Gear animation, progress bar, expandable step history (tap to show all)
- Health Checks — Verification at 2s/5s/10s, auto-restart on failure
- Fallback Regex — Detects tool calls even with escaped quotes in LLM output
- Floating Agent Widgets — Running agents always visible even without message link
- Image Generation Limit — Max 10 per agent run to prevent infinite retries
- Quality Control — Automatic file verification and content check
- Python Execution — Server-side (via middleware) or local fallback
- Persona Export/Import — ZIP file with personality, memories, conversations, workspace files
- Skill Explorer — Visual skill browser with categories, enable/disable toggles
- System Prompt Editor — Customize personality and behavior
- Language Selection — Deutsch, English, Francais, Espanol
- Token Budget Sliders — Episodic and total context budgets
- TTS/ASR Toggles
- API Key with encoded middleware URL (works on any device)
- Android — APK (tested on emulator + real devices)
- Windows — Desktop (native)
- iOS — Build-ready (requires Mac with Xcode)
- macOS — Build-ready
┌─────────────────────────────────────────────────┐
│ BUD-E Flutter App (Android / iOS / Windows) │
│ │
│ ┌──────────┐ ┌──────────┐ ┌──────────────┐ │
│ │ Chat UI │ │ Voice IO │ │ File Viewer │ │
│ └────┬─────┘ └────┬─────┘ └──────────────┘ │
│ │ │ │
│ ┌────▼──────────────▼─────────────────────┐ │
│ │ Chat Provider (State) │ │
│ │ ┌─────────┐ ┌────────┐ ┌───────────┐ │ │
│ │ │ Memory │ │Context │ │ Skills │ │ │
│ │ │ Store │ │Builder │ │ (Tools) │ │ │
│ │ └─────────┘ └────────┘ └───────────┘ │ │
│ └─────────────────┬───────────────────────┘ │
│ │ │
└────────────────────┼────────────────────────────┘
│ HTTPS / API Key
▼
┌───────────────────────┐
│ Admin Bud-E Server │
│ (Middleware Proxy) │
└───────────┬───────────┘
│
┌──────────┼──────────┐
▼ ▼ ▼
Google Black Forest Other
Vertex AI Labs (FLUX) Providers
Detailed documentation:
- Context & Memory System — How BUD-E remembers and constructs context
- Skills & Tools — All available tools and how they work
- Document Generation — DOCX, PDF, HTML, PPTX creation
- Curriculum Search — BM25 search over education plans
- Flutter SDK (3.x+)
- An Admin Bud-E server running
- An API key from your Admin Bud-E server
git clone https://github.com/LAION-AI/bud-e.git
cd bud-e
flutter pub get
flutter run- Open BUD-E and go to Settings (gear icon)
- Enter your API Key in the format:
<key>#<encoded-server-address> - Start chatting!
# Android APK
flutter build apk --release
# iOS
flutter build ios --release
# Windows
flutter build windows --releaseDeveloped by LAION and collaborators. Contributions welcome!
Apache 2.0