Admin Bud-E — AI Middleware & Admin Dashboard

A unified middleware proxy that sits between AI clients (like School Bud-E) and multiple AI providers. Manages users, billing, routing, and provides a web admin dashboard.

Features

Chat & Language Models

• Google Vertex AI: Gemini 2.5 Flash, Gemini 2.5 Pro, and other Gemini models
• OpenAI-compatible providers: Any provider with OpenAI-style API
• Streaming & non-streaming support
• Auto-routing with priority-based failover between providers
• VLM support: Vision-language models with image/PDF input

Image Generation

• Google Vertex AI:
  • gemini-3-pro-image-preview — multimodal text+image generation
  • imagen-4.0-generate-001 — high-quality text-to-image
  • imagen-4.0-fast-generate-001 — faster generation
  • imagen-4.0-ultra-generate-001 — maximum quality
• Black Forest Labs FLUX.2:
  • flux-2-klein-4b — fastest, sub-second (~$0.014/image)
  • flux-2-klein-9b — better prompt understanding (~$0.014/image)
  • flux-2-pro — production-grade (~$0.03-0.05/image)
  • flux-2-max — maximum quality (~$0.07/image)
  • Supports reference images for style transfer (up to 4 refs for klein, 8 for pro/max)
• HyperLab / OpenAI-compatible:
  • nano-banana, nano-banana-pro, nano-banana-2
  • flux-2-dev, seedream-4.5
• Features: Image editing with reference images, negative prompts, seed for reproducibility, aspect ratios

Music Generation

• Google Vertex AI Lyria 3 Pro (lyria-3-pro-preview):
  • Full songs with vocals and lyrics (up to 184 seconds / ~3 minutes)
  • MP3 output at 44.1kHz / 192kbps
  • Supports lyrics with [Verse], [Chorus], [Bridge] tags
  • Caption/style guidance
  • Multi-language: English, German, Spanish, French, Hindi, Japanese, Korean, Portuguese
• Google Vertex AI Lyria 3 Clip (lyria-3-clip-preview):
  • 30-second instrumental clips
• Google Vertex AI Lyria 2 (lyria-002):
  • Instrumental music, 32.8s WAV at 48kHz (legacy)

Speech

• TTS (Text-to-Speech): Google Vertex AI, OpenAI-compatible
• ASR (Speech-to-Text): Whisper via OpenAI-compatible providers

Admin Dashboard

• Users: Create, manage API keys, set credit budgets
• Projects: Group users, set allowances, Common Pool for shared credits
• Providers: Configure upstream AI services (Vertex, BFL, HyperLab, etc.)
• Routes: Priority-based routing with automatic failover
• Pricing: Per-model pricing (tokens, images, audio seconds)
• Usage & Billing: Detailed usage logs, credit ledger
• Maintenance: Database backup/restore, batch user creation

---

Quick Start

Prerequisites

• Python 3.11+
• A Google Cloud project with Vertex AI enabled (for Gemini/Imagen/Lyria)
• Optional: BFL API key (for FLUX.2), HyperLab API key

1. Install

git clone https://github.com/LAION-AI/Admin_Bud-E.git
cd Admin_Bud-E
pip install -r requirements.txt

2. Configure Environment

Create a .env file:

# Google Vertex AI
VERTEX_REGION=europe-west4
VERTEX_PROJECT_ID=your-project-id
VERTEX_SA_JSON=/path/to/service-account.json

# Optional: Gemini API key (for AI Studio)
GEMINI_API_KEY=your-api-key

# Server
HOST=0.0.0.0
PORT=8787
ADMIN_PASSWORD=your-admin-password

3. Run

python serve.py

The server starts on http://localhost:8787. The admin dashboard is at http://localhost:8787/admin.

The Vertex AI proxy runs internally on port 8001 (auto-started by serve.py).

4. Configure Providers (Admin UI)

Open the admin dashboard and add providers:

Provider	Base URL	Notes
vertex	http://127.0.0.1:8001	Local Vertex proxy (auto-started)
bfl	https://api.bfl.ai	Black Forest Labs FLUX.2
hyprlab	https://api.hyprlab.io/v1	HyperLab (nano-banana, etc.)

5. Configure Routes (Admin UI)

Add routes for each service type:

Kind	Provider	Model	Priority
LLM	vertex	gemini-2.5-flash	1
IMAGE	vertex	gemini-3-pro-image-preview	1
IMAGE	bfl	flux-2-pro	2
MUSIC	vertex	lyria-3-pro-preview	1
TTS	vertex	(default)	1
ASR	vertex	(default)	1

---

API Endpoints

Endpoint	Method	Purpose
/v1/chat/completions	POST	Chat/VLM (streaming & non-streaming)
/v1/images/generations	POST	Image generation
/v1/audio/generations	POST	Music generation
/v1/audio/speech	POST	Text-to-Speech
/v1/audio/transcriptions	POST	Speech-to-Text

Image Generation

curl -X POST http://localhost:8787/v1/images/generations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "A beautiful sunset over the ocean",
    "model": "gemini-3-pro-image-preview",
    "size": "1024x1024",
    "n": 1,
    "response_format": "b64_json"
  }'

With reference image (editing):

{
  "prompt": "Make the sky more dramatic",
  "model": "gemini-3-pro-image-preview",
  "input_images": ["data:image/jpeg;base64,..."],
  "response_format": "b64_json"
}

Music Generation (Lyria 3 Pro)

curl -X POST http://localhost:8787/v1/audio/generations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "lyria-3-pro-preview",
    "prompt": "Upbeat pop song about summer with catchy chorus",
    "lyrics": "[Verse]\nSunshine on my face...\n[Chorus]\nSummer days are here..."
  }'

Response includes generated audio (MP3), lyrics, and description:

{
  "data": [{"b64_json": "...", "mime_type": "audio/mpeg", "duration_seconds": 120.5}],
  "lyrics": "Generated or echoed lyrics...",
  "description": "Song structure description..."
}

---

Architecture

Client (BUD-E App)
    │
    ▼
Admin Bud-E Middleware (FastAPI, port 8787)
    │  ├── User auth & billing
    │  ├── Route selection & failover
    │  └── Request transformation
    │
    ├──► Vertex AI Proxy (port 8001) ──► Google Cloud
    ├──► Black Forest Labs API ──► BFL Cloud
    └──► HyperLab / Other OpenAI-compatible APIs

Key Files

• serve.py — Entry point, starts both middleware and Vertex proxy
• main.py — FastAPI app with all routes (chat, image, music, TTS, ASR)
• vertex_openai_proxy.py — Translates OpenAI-format to Vertex AI native APIs
• providers.py — Provider communication, image/music forwarding
• models.py — Database schema (SQLAlchemy)
• billing.py — Credit charging functions
• admin.py — Admin API endpoints
• security.py — Authentication
• db.py — Database setup and migrations

---

Deployment

Docker

docker-compose up -d

Systemd Service

# Copy to server
scp -r . user@server:/opt/admin-bud-e/

# Create service file
sudo cat > /etc/systemd/system/admin-bud-e.service << EOF
[Unit]
Description=Admin Bud-E Middleware
After=network.target

[Service]
WorkingDirectory=/opt/admin-bud-e
ExecStart=/usr/bin/python3 serve.py
Restart=always
User=www-data
Environment=PYTHONUNBUFFERED=1

[Install]
WantedBy=multi-user.target
EOF

sudo systemctl enable admin-bud-e
sudo systemctl start admin-bud-e

Running Forever (Simple)

nohup python serve.py > buddy.out 2>&1 &
# or use the included script:
bash run_forever.sh

---

License

Apache 2.0 — Made by LAION and collaborators.