How to use Magi-System & FAQ
Go to app.magi-system.app. No account registration required. A setup screen will appear on your first visit. Works on both PC and smartphone.
To use Magi-System, you need API keys from AI providers. Create an account and get your API key from any of the following: - Anthropic (Claude): console.anthropic.com → API Keys → Create Key - OpenAI (GPT): platform.openai.com → API keys → Create new secret key - Google (Gemini): aistudio.google.com → Create API key You can start with just one API key for normal mode. Setting up two or more enables Magi Mode (deliberation), and all three maximizes it. Note that usage fees may apply depending on your usage, though some providers offer free tiers.
Enter your API key(s) in the setup screen (or settings). Keys are stored only in your browser and never sent to external servers. Once set up, simply type your question in the chat. With Magi Mode enabled, three AIs independently respond → cross-review → generate a synthesized answer. If something doesn't work, check the Troubleshooting section below.
Leverage three-AI deliberation to produce a reliable research report
Go to app.magi-system.app and set up at least 2 API keys
Turn on the "Magi Mode" toggle at the top right of the chat screen
Enter a specific research topic (e.g., "Analyze the impact of remote work on employee productivity, covering both pros and cons")
Wait for the three AIs to respond → cross-review → generate a synthesized answer (1-2 minutes)
Review the synthesized answer and send follow-up instructions if needed (e.g., "Include more specific data" or "Strengthen the counterarguments")
Adding "Please include specific examples and statistics" to your prompt produces more actionable reports.
Launch Magi-System like a native app from your home screen
Open app.magi-system.app in your smartphone's browser
[iPhone] Tap Safari's share button (□↑) → "Add to Home Screen"
[Android] Tap Chrome's menu (⋮) → "Install app"
The Magi-System icon appears on your home screen
Tap the icon to launch as an app — no browser address bar
The PWA version supports offline caching, so previously loaded screens display even with an unstable connection.
Learn how to use Magi-System while keeping API expenses low
Start by getting a Gemini API key (Google AI Studio offers a relatively generous free tier)
Set up only the Gemini API key in Magi-System settings and use normal mode
Once comfortable, add a Claude or GPT API key to try Magi Mode (deliberation)
Set usage limits on each provider's dashboard to prevent unexpected charges
Use normal mode for everyday questions and Magi Mode for important research or decisions
Free tier availability varies: Google (Gemini) tends to be generous, while Anthropic (Claude) and OpenAI (GPT) may offer small free credits. Check each provider's official site for the latest information.
Magi-System itself is free to use. However, API usage fees from AI providers are your responsibility. Some providers offer free tiers, so we recommend starting within those limits.
Yes, securely. API keys are stored only in your browser's localStorage and are never sent to Magi-System servers. AI requests are sent directly from your browser to each provider. For details, please see our Privacy Policy.
The Magi-System interface supports Japanese and English. Questions and answers can be in any language supported by the AI providers, including Japanese, English, Chinese, Korean, French, and many more.
Magi Mode is Magi-System's unique feature where three AIs (Claude, GPT, Gemini) deliberate to generate an answer. Each AI responds independently, then cross-reviews, and finally generates a high-quality synthesized answer. It's especially effective for complex questions and situations where accuracy is critical.
No, you can use normal mode with just one API key. However, to use Magi Mode (deliberation mode), you need API keys from multiple providers. Setting up all three allows you to fully utilize the deliberation feature.
Yes, Magi-System features a responsive design that works comfortably on smartphones and tablets. Add it to your home screen as a PWA (Progressive Web App) for a native app-like experience. Use the "Add to Home Screen" feature in iOS Safari or Android Chrome.
Magi Hub is a server process that runs on Termux (Android) or a PC. Through Hub, the Magi-System browser app can execute Claude, Codex, and Gemini CLIs, helping you save on API costs. Hub includes a built-in Watchdog (auto-restart) feature that detects AI server failures and automatically recovers. On Android Termux, start it with "magi-hub start" and pair it via Settings > Termux Server Settings.
The Watchdog is a process monitor built into Magi Hub. It periodically checks if the AI server (port 3001) is responsive and automatically restarts it if it stops responding. Note: if Hub itself crashes, the Watchdog stops too. On Android, OOM Kill (forced termination due to low memory) can take down Hub entirely — disabling battery optimization for Termux is recommended.
Yes, by default CLI (command-line interface) takes priority. If the CLI server is unavailable or CLI is not installed, it automatically falls back to API. You can switch between API priority and CLI priority in the Settings screen under "Connection Mode" (visible in Electron and Termux environments only).
Open the Settings screen (gear icon in the top right) and look for the "Connection Mode" section. This setting is only visible in the Electron (desktop app) or Termux environment. In the browser version, this option is not shown — if no server connection is available, it automatically operates in API mode.
In Magi Mode, three AIs each independently generate responses, then cross-review each other's answers, and finally produce a synthesized response. This three-stage process takes longer than normal mode (typically 1-2 minutes). In return, you get more accurate and multi-faceted answers than a single AI can provide.
No. While Magi Mode's deliberation process improves accuracy, AI responses may still contain errors. For specialized fields like medicine, law, and finance, always treat AI responses as reference information and consult professionals. Magi-System is a tool for improving response reliability — final judgment responsibility lies with the user.
Conversation data is stored only in your browser's localStorage. It is not sent to Magi-System's servers. However, your questions are sent to AI providers (Anthropic, OpenAI, Google) as API requests. For each provider's data handling policies, please refer to their official websites.
No. API keys and conversation history are stored in your browser's localStorage, so they don't transfer between different browsers or devices. You'll need to set up API keys again on a new browser. Clearing browser cache/data will also reset everything.
Magi-System stores API keys in your browser and operates serverlessly, making data breach risks relatively low compared to typical web services. However, on shared PCs, API keys could be accessed from localStorage. On shared devices, we recommend clearing browser data after use.
AI Roles (Skills) let you preset a specific persona or expertise for the AI. For example, you can create roles like "English Teacher," "Code Reviewer," or "Cooking Advisor" and start a conversation with that role active. Create, edit, and switch roles in the "Skills" tab in Settings. Use the selector above the chat input to activate a role.
Custom CLI providers are an advanced feature that lets you add CLI tools beyond Claude, Codex, and Gemini to Magi-System. You can add them from the "API & CLI" tab in Settings. Configure the CLI path, arguments, and timeout to call any locally running AI CLI from Magi-System. Available in Termux or Electron (desktop) environments only.
Open Settings (gear icon at top right) and select the "Features, Appearance & Language" tab. Click the "Appearance" sub-tab to find theme options. You can choose from "Light", "Dark", or "Follow system setting". The "Follow system setting" option automatically syncs with your OS dark/light mode preference.
A download icon appears next to the provider badge ("Claude", "Gemini", etc.) on each AI response message — tap or hover to reveal it. Click it to save that individual message as a Markdown file. The filename format is "conversation-title_provider_YYYYMMDD-hhmm.md".
Hover (or long-press) a conversation in the left sidebar to reveal rename, delete, and download icons. Click the download icon to save the full conversation as a Markdown file. Great for archiving TRPG session logs or any conversation you want to keep.
LittleNotebook lets you attach text or PDF files as reference material for the AI. Tap the notebook icon on the left side of the input bar to open the panel and upload files. The file content is automatically included as context when you send a message. Useful for summarizing long documents or asking questions based on specific materials.
【Left sidebar】Shows your conversation history. You can start new conversations, rename, delete, or export them as Markdown. 【Gear icon (top right)】Opens Settings, where you manage API keys, CLI settings, theme, language, and Skills (AI roles). 【Selector above input】Choose an AI Skill (role) such as "English Teacher" or "Code Reviewer". 【Notebook icon (input left)】Opens LittleNotebook to attach reference files. 【Magi Mode toggle】Located near the right of the input bar — toggles three-AI deliberation mode on/off.
Verify your API key was copied correctly. Check for leading/trailing spaces and ensure it hasn't expired. Check the API key status on each provider's dashboard and generate a new key if necessary.
Check your network connection. The AI provider's servers may be temporarily congested. Wait a few minutes and try again. If the issue persists, clear your browser cache or try a different browser.
PWA installation requires an HTTPS connection and a compatible browser. On iOS Safari, tap the share button and select "Add to Home Screen." On Android Chrome, open the menu and select "Install app." PWA installation may not work in private browsing mode.
Check your usage on each AI provider's dashboard and set usage limits. Magi Mode uses multiple AIs simultaneously, which increases API costs accordingly. To reduce costs, consider using AI CLI free tiers through the Extension. CLI integration lets you maintain deliberation quality while significantly cutting API expenses.
When you start the server on Termux, the token is automatically saved to ~/.magi-bootstrap-token. Tap the "Auto-fill" button in the app's "Termux Server Settings" dialog to automatically retrieve it — this works when the browser is on the same Android device as Termux. If you're using a browser on a different device, run cat ~/.magi-bootstrap-token in Termux to view the token and enter it manually.
Android may forcibly terminate background apps when memory runs low — a mechanism called OOM Kill. Termux and Magi Hub can be affected. To prevent this: ① go to Android Settings and set battery optimization for Termux to "Unrestricted" (or "Don't optimize"), ② ensure background activity is not restricted, ③ keep Termux's wake lock notification visible in the notification bar. After a crash, restart Hub by running "magi-hub start" in Termux.