AI Studio

AI Studio recreates the playground interface and interaction experience of Google AI Studio as a fully self-hosted web application. You can configure any Gemini or OpenAI-compatible endpoint (custom Base URL, API key, proxy), and all your data is stored in a local embedded SQLite (libsql) database — nothing ever leaves your machine except the requests you send to your own API endpoint.

This project is an unofficial, educational open-source recreation of the Google AI Studio interface, built with Next.js 14. It is for learning and research purposes only — it is not affiliated with, endorsed by, or sponsored by Google, and it involves no commercial interests whatsoever.

Features

Core Playground

• 💬 Streaming (SSE) and non-streaming chat completions
• 🔌 Custom Base URL / API key, switchable protocol (Gemini or OpenAI-compatible), optional HTTP proxy
• 🤖 Model selector with live model listing (append-only sync with the endpoint's catalog — new models are added, known ones never removed), plus custom model registration (custom context window, category, description)
• 💭 Thinking bubbles: readable thought summaries rendered as markdown
• 🔎 Search citations: Google-Search-grounded answers show footnote markers at the cited spans with a collapsible source list
• 📋 Paste-to-attach: paste images / PDF / audio / video / text files straight from the clipboard into the input box
• ✏️ Regenerate a response from any point without affecting later turns
• 🔀 Conversation branching / duplication, context JSON import & export
• 📜 Lazy-loaded history with automatic pagination on scroll-to-top

Long-term memory (RAG)

• 🧠 RAG (Retrieval-Augmented Generation): when the sliding window trims older messages out of the context window, the trimmed messages are embedded into vectors and stored in the local database, then semantically retrieved and re-injected into the request — so the model keeps access to early conversation history that no longer fits the window
• 🔌 Two embedding providers: API (your endpoint's OpenAI-compatible /v1/embeddings or Gemini embeddings) or Local (on-device ONNX model via @huggingface/transformers) — the Local provider works even when your relay/gateway offers no embedding channel at all
• 🛡️ RAG cooperates with the sliding window (never replaces it) and degrades gracefully: if embedding or retrieval fails, the chat simply falls back to plain sliding-window behavior
• ⚙️ Configurable in Run settings → Advanced settings: enable/disable, provider, embedding model, retrieval Top-K (1–20)

Tools

• Structured outputs (with a JSON Schema editor)
• Function calling (with a declarations editor)
• Code execution
• Grounding with Google Search / Google Maps
• URL context

Configuration & Personalization

• 🎛️ Run settings panel: system instructions, temperature, top-p / top-k, max output tokens, max context tokens cap, media resolution, thinking level, safety thresholds per harm category
• 🗂️ System instruction templates (save / apply / manage)
• 🌗 Theme: dark / light / system
• 🔒 Password protection with lock screen (optional, stored locally)
• 📊 Per-message and per-conversation token usage statistics

Data & Voice

• 🗄️ Local SQLite (libsql) database — conversations, messages, settings and usage are all persisted on disk; the Dashboard shows DB size and contents
• 🗣️ Edge-TTS speech synthesis: multiple voices, adjustable volume / rate / pitch, per-message playback and auto-read mode for AI responses
• 📱 Fully responsive — desktop sidebar and mobile drawer layouts

Demo Screenshots

Representative views of the application (all images live in Pictures/):

Playground — dark theme	Playground — light theme
Dashboard — database & usage	History — conversation list

🖼️ Full screenshot gallery (click to expand)

Interface & Themes

Home — mobile	Sidebar — mobile
Chat interface — dark	Chat interface — light

Settings & Tools

Settings window	API configuration
Tool selection	System instruction templates

Models & Structured Output

Model selection	Add custom model
Structured output Schema editor	Function declarations editor

Security

Password lock screen	Clear password

Installation

git clone https://github.com/SRInternet-Studio/AIStudio.git
cd AIStudio
npm install
npm run dev

Open http://localhost:3000, click the Settings button at the bottom-left of the input box, choose a protocol, enter your Base URL and API key, then select or add a custom model in the run settings panel (right side) to start chatting.

Configuration

Item	Where	Notes
Base URL / API key / protocol / proxy	Bottom input box → Settings button (bottom-left) → API configuration	Persisted locally
System instructions & templates	Run settings panel (right side)	Templates saved to DB
RAG long-term memory	Run settings panel → Advanced settings	On by default; use the Local provider when your endpoint has no embedding channel
Password protection	Settings window	Optional lock screen
Theme	Settings window	dark / light / system

Project Structure

├── src/
│   ├── app/            # Next.js App Router pages + API routes
│   │   └── api/        # force-dynamic backend endpoints
│   ├── components/     # chat, layout, settings, dashboard, documentation
│   ├── lib/            # db (libsql), api-client, context manager, RAG, models
│   ├── store/          # Zustand global state
│   └── types/          # Shared TypeScript types
├── public/             # Static assets (logo)
├── Pictures/           # Demo screenshots
├── Docs/               # Chinese documentation
└── data/               # Local SQLite database (auto-created)

Scripts

Command	Description
npm run dev	Start the development server
npm run build	Production build
npm run start	Serve the production build
npm run lint	Run ESLint

Documentation

In-app documentation is available under Manage → Documentation (English / 中文 switchable). Source files: WELCOME · DEVELOPMENT · SECURITY · CONTRIBUTING · CODE_OF_CONDUCT · DISCLAIMER — Chinese versions in Docs/.

Disclaimer

This project is an open-source educational recreation of Google AI Studio. It is not affiliated with Google and involves no commercial interests.

AI models may make mistakes, so double-check outputs. See DISCLAIMER for details.

License

Apache License v2.0.