lemon cha Browser Extension

简体中文

lemon cha is a lightweight Chrome and Firefox extension that opens a dedicated AI chat tab for two kinds of model providers:

• Custom LLM APIs that use an OpenAI Chat Completions-compatible format.
• Local Ollama models, with http://127.0.0.1:11434/api/chat as the default endpoint.
• Requests are sent from the browser extension background context.
• Local Ollama requests include a direct-connect fallback that normalizes the local origin path.
• The interface supports streaming output, stop generation, Markdown rendering, copyable code blocks, multi-session history, folders, rename/delete actions, Markdown export, and JSON import/export.

Install locally in Chrome

1. Open Chrome and visit chrome://extensions/.
2. Enable Developer mode.
3. Click "Load unpacked".
4. Select dist/lemon-cha after running npm run build:chrome.
5. Click the extension action to open lemon cha in a dedicated browser tab.

Install locally in Firefox

1. Run npm run build:firefox.
2. Open Firefox and visit about:debugging#/runtime/this-firefox.
3. Click "Load Temporary Add-on...".
4. Select dist/lemon-cha-firefox/manifest.json.
5. Click the extension action to open lemon cha in a dedicated browser tab.

Development checks

lemon cha is a lightweight Manifest V3 extension without a frontend bundler. The repository uses npm scripts for validation and packaging:

npm run lint
npm run typecheck
npm run build

• npm run lint: validates required extension files, Chrome and Firefox Manifest V3 setup, HTML asset references, and JavaScript syntax.
• npm run typecheck: runs a lightweight syntax-oriented validation pass for this vanilla JavaScript project.
• npm run build: copies runtime files into dist/lemon-cha for Chrome and dist/lemon-cha-firefox for Firefox.
• npm run build:chrome: builds only the Chrome package.
• npm run build:firefox: builds only the Firefox package.

Firefox uses the same runtime code with a generated Firefox manifest. The generated manifest swaps Chrome's background.service_worker for Firefox's background.scripts and adds Gecko extension settings required for Firefox distribution.

Custom API settings

Choose the custom API workflow and configure:

• Endpoint, for example https://api.openai.com/v1/chat/completions
• Model name, for example gpt-4o-mini
• API key, if the provider requires one
• API format: OpenAI Chat Completions or Ollama /api/chat
• Thinking mode behavior
• Interface language, answer language, and default translation target language

If the API format is OpenAI-compatible and the endpoint only contains the domain or /v1, lemon cha completes it to /v1/chat/completions. Provider-specific paths can still be entered manually.

Built-in provider presets include DeepSeek, OpenAI, Qwen DashScope, Kimi, SiliconFlow, OpenRouter, Groq, and local Ollama.

Conversation management

• Create new conversations.
• Switch between saved chats from the sidebar.
• Organize chats inside folders.
• Rename or delete conversations.
• Export the current chat as Markdown.
• Import or export JSON records containing conversations, settings, and custom prompt presets.
• Clear history or restore factory defaults from the data settings area.

Message actions

Each assistant or user message can expose compact actions such as:

• Copy
• Regenerate
• Continue from here
• Delete

Connection checks

The connection checker supports:

• Ollama /api/tags for model discovery and reachability checks
• OpenAI-compatible lightweight test requests for endpoint, model, and API key verification

Generation experience

• Streaming model responses
• Stop generation while preserving partial output
• Dynamic waiting state for slower local models
• Generation timing and approximate token metrics
• Automatic first-turn conversation title generation
• Markdown rendering for paragraphs, lists, emphasis, code blocks, and tables
• Copy button inside code blocks
• Persistent bottom composer for long conversations
• Collapsible sidebar for a cleaner reading surface

Prompt presets

The composer includes reusable prompt presets:

• Summarization
• Translation and polishing
• Code review
• Troubleshooting
• Writing revision
• Task planning

Built-in templates are localized for Simplified Chinese, Traditional Chinese, English, Japanese, Korean, French, Spanish, and German.

When a preset is applied:

• If the composer is empty, lemon cha replaces the input and selects the first {placeholder}.
• If the composer already contains text, lemon cha asks before replacing it.
• Translation presets can resolve the default translation target language selected in Settings.

Ollama setup

Start Ollama and install a local model:

ollama serve
ollama pull llama3.1

Default lemon cha settings:

• Endpoint: http://127.0.0.1:11434/api/chat
• Model: llama3.1

Connection strategy:

• localhost is normalized to 127.0.0.1.
• /api/tags is used for model listing and connectivity tests.
• /api/chat is used for multi-turn chat and streaming.
• Local Ollama requests are handled from the extension background context.

Ollama 403 / Origin rejection

If Ollama rejects the browser extension origin, restart Ollama with extension-origin allowance.

macOS Ollama app:

launchctl setenv OLLAMA_ORIGINS "chrome-extension://*,moz-extension://*"

Quit the Ollama menu bar app completely, then open it again.

macOS/Linux command line:

OLLAMA_ORIGINS="chrome-extension://*,moz-extension://*" ollama serve

Temporary broad debug option:

OLLAMA_ORIGINS="*" ollama serve

Windows PowerShell:

$env:OLLAMA_ORIGINS="chrome-extension://*,moz-extension://*,http://localhost:*,http://127.0.0.1:*"
ollama serve

Repository layout

• manifest.json: source Chrome Manifest V3 configuration
• scripts/build-extension.js: Chrome and Firefox extension package builder
• background.js: background networking and provider calls
• chat.html: dedicated chat page
• styles.css: visual system and responsive layout
• popup.js: state, settings, conversations, and UI flows
• chrome-store/: Chrome Web Store listing, privacy, review, and release materials