# AtLine AI for Obsidian

Ask any LLM questions about your notes inline using custom triggers.

![AtLine AI demo](demo.gif)

> Set a hotkey in **Settings → Hotkeys → AtLine AI** to trigger a response (e.g. Cmd+Enter on Mac).

> **Desktop only.** This plugin uses Node.js APIs and is not available on Obsidian Mobile.

## Quickstart

1. Install the plugin (see [Installation](#installation))
2. Choose your connection method:
   - **API mode**: Go to Settings → AtLine AI → API Keys and add your API key (Anthropic, OpenAI, or Google)
   - **CLI mode**: Install the CLI tool (e.g., `npm install -g @anthropic-ai/claude-code`) and authenticate
3. Set a hotkey: **Settings → Hotkeys** → search "AtLine AI" → assign your preferred shortcut
4. In any note, type: `@claude What is this note about?`
5. Press your hotkey

The AI response appears inline below your question.

## Features

- **Multiple AI providers**: Claude, Google Gemini, Ollama (local models), OpenAI, and a dedicated chart/plot agent
- **Flexible connection modes**: Use CLI tools or direct API keys — no CLI required for API mode
- **Streaming responses**: See AI responses appear in real-time as they are generated
- **Linked notes context**: Automatically include `[[wikilinked]]` notes as context for the AI
- **Inline responses**: Answers appear directly in your note as blockquotes, callouts, plain text, or code blocks
- **Customizable agents**: Create multiple agents with different providers and system prompts
- **Configurable hotkey**: assign any shortcut via Settings → Hotkeys
- Works alongside other typing plugins without conflicts

## Installation

### Via BRAT (recommended for beta testing)

1. Install the [BRAT plugin](https://github.com/TfTHacker/obsidian42-brat) from the Obsidian community plugins
2. Open BRAT settings and click **Add Beta Plugin**
3. Enter: `https://github.com/ergesmema/obsidian-atline-ai`
4. Enable AtLine AI in Settings → Community Plugins

### Manual installation

1. Download the [latest release](https://github.com/ergesmema/obsidian-atline-ai/releases/latest) and extract it into your vault's `.obsidian/plugins/` folder
2. Enable the plugin in Settings → Community Plugins

## Requirements

- Obsidian v0.15.0 or higher
- Desktop only (Windows, macOS, Linux)

**Choose CLI or API mode per provider:**

| Provider | CLI mode | API mode |
|----------|----------|----------|
| **Claude** | Claude CLI (`npm install -g @anthropic-ai/claude-code`) | Anthropic API key |
| **Gemini** | Gemini CLI (requires Node.js 20+) | Google AI API key |
| **OpenAI** | Codex CLI | OpenAI API key |
| **Ollama** | Ollama running locally (`http://localhost:11434`) | — |
| **Plot** | Uses whichever provider the @plot agent is configured with | — |

API mode requires no CLI installation — just add your API key in settings.

## Usage

### Basic usage

1. Open any note in Obsidian
2. Type an agent trigger followed by your question:
   ```
   @claude What is the main topic of this note?
   @gemini Summarize this section
   @ollama Explain this concept
   @gpt Help me rewrite this paragraph
   @plot Draw a bar chart: Q1=120, Q2=95, Q3=140, Q4=200
   ```
3. Press your configured hotkey (set in Settings → Hotkeys → AtLine AI)
4. The AI reads your note and responds inline

The plugin searches up to 20 lines above your cursor to find the nearest `@agent` trigger, so you can place your cursor anywhere after the query.

### Command palette

1. Place your cursor on or below a line with an agent query
2. Open the command palette (Cmd/Ctrl+P)
3. Run **AtLine AI: Run AI agent on current line**

## Configuration

Go to **Settings → AtLine AI** to configure:

- **AI agents**: Create and customize multiple agents with different providers and system prompts
- **Connection mode**: Choose CLI or API mode per agent
- **API keys**: Add your Anthropic, OpenAI, or Google AI API keys
- **Include linked notes**: Automatically include `[[wikilinked]]` notes as context
- **Response style**: Choose how responses are displayed — blockquote, callout, plain, code, or custom
- **Hotkey**: Choose your preferred keyboard shortcut
- **Timeout**: Maximum wait time for AI responses (default: 2 minutes)
- **Ollama settings**: Configure Ollama base URL and model
- **CLI paths**: Configure paths to Node.js and CLI executables

### Configuring agents

Each agent has:
- **Alias**: The trigger word (e.g., `claude`, `gemini`, `gpt`, or any custom word)
- **Provider**: Which AI service to use
- **Connection mode**: CLI or API
- **Model**: Which model to use (e.g., `claude-sonnet-4-20250514`, `gpt-4o`)
- **System prompt**: Custom instructions for how the AI should behave
- **Include linked notes**: Whether to include `[[wikilinked]]` notes as context

## Troubleshooting

### Plugin not responding

- Check Obsidian's developer console (Cmd/Ctrl+Shift+I) for errors
- Verify the appropriate CLI is installed for your provider
- For Ollama: ensure Ollama is running at the configured base URL

### Claude / Gemini / OpenAI not working

**If using API mode:**
- Verify your API key is entered correctly in Settings → AtLine AI → API Keys
- Use the **Test** button next to your API key to verify it works
- Check that your API account has credits or quota available

**If using CLI mode:**
- Verify CLI is installed: `which claude`, `which gemini`, or `which codex`
- Check that you are authenticated with the CLI
- Test the CLI independently in your terminal
- **For Gemini CLI**: Requires Node.js version 20 or higher
- **Check CLI paths in settings**: Go to Settings → AtLine AI → CLI Paths and ensure all paths are correct
- Default paths use system PATH (e.g., `node`, `claude`). For custom installations, use absolute paths:
  - Homebrew: `/usr/local/bin/node`
  - NVM: `~/.nvm/versions/node/v20.x.x/bin/node`

### Ollama not working

- Verify Ollama is running: `curl http://localhost:11434/api/tags`
- Check that the model name in settings matches an installed model
- Verify the base URL in settings is correct

### Timeout errors

- Increase the timeout value in plugin settings
- Check your internet connection (for Claude/Gemini/OpenAI)
- For local models (Ollama): check system resources

## Network and privacy

This plugin can send note content to external AI services depending on which provider you use:

| Provider | Service | Privacy policy |
|----------|---------|----------------|
| Claude (API mode) | Anthropic API | [anthropic.com/privacy](https://www.anthropic.com/privacy) |
| Gemini (API mode) | Google AI API | [ai.google.dev/terms](https://ai.google.dev/terms) |
| OpenAI (API mode) | OpenAI API | [openai.com/policies/privacy-policy](https://openai.com/policies/privacy-policy) |
| Ollama | Localhost only | No data leaves your machine |
| CLI mode (any) | Routed through the respective CLI tool | See provider above |

**What is sent:** the content of your active note, any `[[wikilinked]]` notes you have enabled for context, and your query. No data is collected by this plugin itself.

To keep your notes private, use Ollama with a local model.

## Roadmap

### v1.2 - Context enhancements
- [ ] Selection-only mode (AI only sees highlighted text)
- [ ] Folder-wide context option
- [ ] Conversation history within a note

### v1.3 - Response improvements
- [ ] Edit and regenerate responses
- [ ] Save favourite prompts/templates

### v1.4 - More providers
- [ ] Mistral AI support
- [ ] Custom API endpoints

Have a feature request? [Open an issue](https://github.com/ergesmema/obsidian-atline-ai/issues)!

## Contributing

Contributions are welcome! You can:

- **Suggest features**: [Open an issue](https://github.com/ergesmema/obsidian-atline-ai/issues) or join [Discussions](https://github.com/ergesmema/obsidian-atline-ai/discussions)
- **Report bugs**: Include steps to reproduce and error messages from the developer console
- **Submit code**: See [CONTRIBUTING.md](CONTRIBUTING.md) for guidelines

## License

MIT