AI聊天插件

Bulbasaur Chat

An AI chat plugin for Obsidian with multi-provider LLM support, MCP integration, and deep vault context.

Features

Multi-Provider LLM Support

Talk to any major LLM provider from inside your vault:

All providers support streaming responses and real-time model fetching.

MCP (Model Context Protocol) Integration

Connect to MCP servers to extend what Bulbasaur Chat can do:

• Browse and connect to MCP servers via /mcp
• Execute tools, load prompts, and browse resources
• Auto-discovery of available servers
• Server status indicators in the header

Vault Integration

Your notes are first-class citizens in Bulbasaur Chat conversations:

• @ mentions — Type @ in the input to search and attach notes as context
• [[Wiki links]] — Click to add as context, Ctrl+click to open the note
• Visual indicators show what context is currently attached

System Prompts

Define who the LLM is. Load persona and behavioral prompts from .md files in a configurable vault folder.

/sys         → pick and load a system prompt
/sysclear    → remove the active system prompt

Mental Models

Define how the LLM thinks. Mental models are reasoning frameworks injected with an explicit instruction for the LLM to actively apply them as an analytical lens — distinct from the system prompt's identity/persona.

/mental       → pick and load a mental model
/mentalclear  → remove the active mental model

Both can be active simultaneously. The system prompt shapes who the LLM is; the mental model shapes how it thinks.

Conversation Management

• Multiple conversation threads with persistent history
• Smart auto-save (only persists conversations with messages or custom names)
• Auto-cleanup of empty/unnamed conversations
• Rename, delete, and browse history

Chat Interface

• Clean sidebar UI with streaming markdown rendering
• Code block syntax highlighting, headers, lists, tables
• Copy button on assistant messages
• Customizable background image with opacity control
• Auto-hide header on scroll
• Optional token count display
• Loading animation with timer during API calls

Installation

Manual Install

1. Download the latest release
2. Extract main.js, manifest.json, and styles.css into your vault at:
   复制

   .obsidian/plugins/bulbasaur-chat/
   

3. Restart Obsidian
4. Enable Bulbasaur Chat in Settings → Community Plugins

From Source

cd /path/to/vault/.obsidian/plugins
git clone https://github.com/PoweredbyPugs/Bulbasaur-Chat.git
cd Bulbasaur-Chat

Getting Started

1. Open Settings → Bulbasaur Chat
2. Select your LLM provider
3. Enter your API key (not needed for local providers)
4. (Optional) Configure paths for system prompts and mental models folders
5. Open Bulbasaur Chat from the ribbon icon or the command palette: Open Bulbasaur Chat Chat

Commands

All commands are typed directly into the chat input. Type /help to see the full list.

Context

Conversations

Interface

Special Input

Configuration

API Settings

Model Settings

UI Settings

Paths

MCP

How It Works

Bulbasaur Chat runs as an Obsidian sidebar view. When you send a message, it:

1. Composes a system message from your active system prompt + mental model + attached note context
2. Sends the full conversation to your selected LLM provider
3. Streams the response back with real-time markdown rendering
4. If MCP tools are available and the LLM invokes one, Bulbasaur Chat executes the tool call and feeds the result back

All conversation data is stored locally in the plugin's data.json file within your vault. API keys are stored in Obsidian's plugin settings — nothing is sent anywhere except to the LLM provider you configure.

Development

Bulbasaur Chat is developed in a bulbasaur-chat/ source folder and deployed into the vault's plugin folder.

Building from Source

cd bulbasaur-chat
npm install
npm run build      # TypeScript check + esbuild bundle
npm run deploy     # Build + copy main.js to the vault plugin folder

Project Structure (Development)

bulbasaur-chat/
├── main.ts                  # Plugin entry point
├── src/
│   ├── types/               # TypeScript interfaces
│   ├── services/            # Cache, logger, MCP client
│   ├── providers/           # LLM provider implementations
│   └── views/modals/        # Modal UI components
├── styles.css               # Plugin styles
├── manifest.json            # Plugin manifest
├── esbuild.config.mjs       # Build configuration
└── package.json

Provider Architecture

Bulbasaur Chat uses a factory pattern for LLM providers. Each provider implements a common interface, and getProvider() returns the appropriate implementation based on settings. All providers receive a unified ProviderContext containing the system message, conversation history, and streaming callbacks.

License

MIT

Author

PoweredbyPugs