Teaspoon IDE Settings Guide

1. Opening the settings dialog

• Click the ⚙️ button in the top-right corner of the AI chat pane.
• If no API key is configured, trying to send a chat message also opens the settings dialog.

2. UI language

Settings > Appearance > Language switches the UI language. It applies instantly — no save needed.

Languages are loaded from translation files named lang/<code>.json. The dev build reads lang/ in the project root; packaged builds read resources/lang/ next to the exe, so you can add or edit translations yourself. English and Japanese (日本語) ship by default.

This guide uses the English UI labels below (Japanese labels are noted where helpful).

3. Choosing an LLM provider

Settings > LLM Provider > Provider selects the AI backend. The switch applies immediately — no save button needed.

4. Setting up a Gemini API key

Steps

1. Get an API key from Google AI Studio (sign in with a Google account and use "Get API key". A free tier is available).
2. In Teaspoon IDE's settings, make sure the provider is Gemini API (cloud).
3. Paste the key into the Gemini API Key field. The 👁️ button toggles visibility of what you typed.
4. Pick a model under Model Selection. The default is Gemini 3.8 Flash, which is fine for most use (Gemini 3.8 Flash is offered at an introductory price of $0.75/1M input and $3.75/1M output through December 31, 2026 — exactly half of the standard $1.50/$7.50 rate taking effect January 1, 2027. For even cheaper usage, Gemini 3.5 Flash-Lite at $0.30/$2.50 is recommended). To use a model not in the list, choose Custom Model and type the model ID (e.g. gemini-2.0-flash-thinking).
5. Click "Save Settings". The API key, model, and proxy settings are only persisted when you click this button (unlike appearance and provider changes, which apply instantly).
6. Send a message in the chat — a response means you're done.

Notes

• "Save Settings" stays disabled until an API key is entered. When the provider is Ollama, the buttons are hidden entirely (everything Ollama-side applies on change).
• The API key is stored in plaintext in localStorage. Be careful on shared PCs. Use the "Clear API Key" button to erase it (this removes only the Gemini-related settings).
• If the selected model is unavailable, the app automatically tries to fall back to gemini-1.5-flash. If that also fails, pick a different model. (While signed in to an organization, there is no fallback — the organization controls which models you may use.)
• While signed in to an organization, the personal model dropdown is locked and your personal model setting is ignored.
• Requests time out after 120 seconds, and responses have a length cap. Break very large requests into smaller ones.

5. Hiding your API key behind a LiteLLM proxy (optional)

If you don't want the real Gemini API key on the client PC, you can route requests through a proxy such as LiteLLM.

Teaspoon IDE → dummy key → LiteLLM (e.g. on a VPS) → real API key → Google AI Studio

Steps

1. Set up LiteLLM on a VPS or similar:

   pip install litellm
   litellm --model gemini/gemini-3.8-flash --api_key YOUR_REAL_API_KEY

2. In Teaspoon IDE's settings, enable LLM Proxy (LiteLLM) > Use Proxy.
3. Enter the proxy URL (e.g. http://your-vps:4000, https://your-proxy.com).
4. Put a dummy value in the Gemini API Key field and click "Save Settings". The value you enter is sent to the proxy as an Authorization: Bearer <key> header, so if your LiteLLM setup issues virtual keys, enter one of those.

Benefits: the real key never touches the client, key rotation is easy, you can monitor and cap usage, and multiple AI providers can be unified behind one endpoint.

6. Ollama + Gemma (local LLM)

With Ollama, the AI chat works with no API key and fully offline.

Steps

1. Install Ollama.
2. Pull a model. Teaspoon IDE's default model is gemma4:e4b:

   ollama pull gemma4:e4b

   Other models (gemma3:4b, qwen3:8b, etc.) work too.
3. Keep Ollama running (ollama serve, or the resident tray app).
4. In Teaspoon IDE's settings, switch the provider to Ollama (local, offline).
5. Leave the Endpoint at the default http://localhost:11434. Only change it if Ollama runs on another machine or port.
6. The Model field auto-detects models installed in Ollama and shows them in a dropdown — just pick one. If detection fails, the field becomes a plain text input; type the model name directly.

"Connected - N model(s) installed." below the settings means the connection works. Provider settings apply immediately — no save needed.

With Ollama, replies stream token by token, so even slow local models feel alive while they work. Cancelling actually aborts the in-flight request. Each reply also shows the model used and its round-trip time (e.g. (6.1s)), which makes comparing model speeds easy.

7. Using unauthenticated OpenAI-compatible APIs

In Ollama mode, Teaspoon IDE POSTs OpenAI-compatible requests to {endpoint}/v1/chat/completions with no authorization header at all. That means any unauthenticated OpenAI-compatible server may work simply by pointing the endpoint at it.

Servers that may work

Tips

• Set the provider to Ollama and enter the server's base URL as the endpoint. Do not include a trailing /v1 — the app appends /v1/chat/completions itself.
• Model auto-detection uses Ollama's own /api/tags, so on other servers you'll see a "Cannot reach Ollama" warning. That only means detection failed — the model field becomes a text input, and chat itself often still works. Enter the exact model name the server expects and try it.

Limitations (important)

• Services that require API keys cannot be used. Ollama mode has no way to send an Authorization header, so endpoints like OpenAI proper or OpenRouter won't work. For authenticated services, consider running a local proxy such as LiteLLM.
• Destinations are restricted by CSP. Only http://localhost:<any port> and https:// are allowed. Plain-HTTP remote servers such as http://192.168.x.x:11434 on your LAN cannot be reached (serve them over HTTPS, or tunnel them to localhost, e.g. via SSH).

8. Organization mode (centrally managed, v0.5.0+)

Organization mode lets a company, school, or team centrally manage members' AI usage. When enabled, the whole app switches to a sign-in gate and stays locked until the member signs in with an account issued by the organization. See Teaspoon IDE Organization for the concept and deployment options.

Enabling it (admin / deployer)

1. In Settings > Organization, enter the organization's Server URL (e.g. https://llm.example.org).
2. Turn on "Require organization sign-in" (it stays disabled until a server URL is entered).
3. The app immediately switches to the sign-in screen and waits for credentials issued by the organization's server.

Signing in (member)

1. On the sign-in screen, enter the Username and Password issued by your organization.
2. On success, the server issues a per-user virtual key, and AI requests then go through the organization's LLM proxy (e.g. LiteLLM). The real API key never reaches your machine.

While signed in

• Remaining budget shown as a percentage: the chat header shows your allotted budget remaining as "Budget 99.9%" next to the model badge — never a currency amount. It warns below 20%, turns red at 0%, and its tooltip shows the reset date. It refreshes after each response and every 60 seconds.
• Models limited to the allowed list: usable models come from the server. Settings > Organization shows a model dropdown restricted to that list, and the personal model selector is locked (the automatic model fallback is also disabled in managed mode).
• Personal settings are preserved: managed credentials are stored separately from your personal Gemini API key. Signing out or disabling the mode never destroys your personal configuration.

Switching back to local use

• On the sign-in screen, choose "Continue without an organization" → confirm → "Disable" to turn off organization mode on this device and return to personal use. (The issued key is kept, so re-enabling later won't require a fresh sign-in.)
• While signed in, use "Sign Out" in Settings > Organization to discard the issued key.

9. Other settings

The remaining options in the settings dialog. Unless noted otherwise, changes apply immediately and are saved automatically.

The 🧠 button in the chat pane also toggles "automatic context ↔ manual file selection". In manual mode, the contents of files you pick in the Explorer are sent to the AI as-is (independent of the Context Mode setting).

10. Troubleshooting

Still stuck? Report it via GitHub Issues. If the packaged app fails to start at all, a log may have been written to %TEMP%/teaspoon-crash.log.

← Back to the Teaspoon IDE product page