LMMs-Lab WriterWriter
Loading
Back to docs

OpenCode

LMMs-Lab Writer includes a built-in AI panel powered by OpenCode, giving you AI assistance directly inside the editor.

Getting Started

  1. Install OpenCode on your machine (via npm, Homebrew, or from source)
  2. Click the AI panel icon in the sidebar
  3. The app auto-detects your OpenCode installation and starts a local daemon
  4. Start chatting — the AI can read and edit your project files directly

The daemon runs locally in your project directory, so the AI has full context of your LaTeX files.

Chat Interface

Sending Messages

Type your prompt in the input box and press Enter or click Send. Examples:

  • "Write an abstract for this paper summarizing our three key contributions"
  • "Add a table comparing our method's performance against baselines in Table 1"
  • "Fix the bibliography entries that are missing DOIs"
  • "Rewrite Section 3 to be more concise"

Attaching Files

Drag and drop files into the input area, or click the attachment button. This is useful for providing additional context — images, data files, or reference papers — that the AI should consider.

Sessions

You can manage multiple chat sessions:

  • Create a new session for each task or section of your paper
  • Switch between sessions to maintain separate conversation threads
  • Each session keeps its own history

Model Selection

Choose from available AI models and providers for each conversation. The model selector appears in the chat interface.

How It Works

OpenCode runs as a daemon process on your local machine (default port: 4096). The app communicates with the daemon via a local API. When you send a message:

  1. Your prompt is sent to the OpenCode daemon
  2. The AI reads your project files for context
  3. The AI writes edits directly to your .tex files
  4. The editor detects the file changes and updates in real-time
  5. You see the result instantly

Status Indicators

The AI panel shows its current state:

  • Ready — Waiting for your next prompt
  • Writing — Actively generating and editing files
  • Busy — Processing a request

A pulsing indicator in the header also reflects the current AI status.

Task Tracking

When the AI works on multi-step tasks, a task bar appears showing progress:

  • Pending tasks
  • In-progress tasks
  • Completed tasks

This helps you track what the AI is doing, especially for larger edits that involve multiple files.

Troubleshooting

OpenCode not detected

The app searches for OpenCode in your system PATH and common installation directories (npm global, Homebrew, etc.). If it's not found:

  1. Verify OpenCode is installed: opencode --version
  2. Ensure it's in your PATH
  3. Restart the app after installing

Connection issues

If the daemon disconnects, the app automatically attempts to reconnect. If reconnection fails:

  1. Check if another process is using port 4096
  2. Try restarting the AI panel
  3. Check the terminal output for error messages

Port conflicts

If port 4096 is occupied, the app automatically finds the next available port. You can also configure a custom port in the settings.