Local Terminal Agent
If you prefer working continuously in the terminal or want to use Locus without an MCP-enabled editor, you can run the Locus Agent directly from your command line.
Overview
The Local Terminal Agent runs as a standalone process. It:
- Connects to your Locus Workspace.
- Pulls the active Sprint and assigned Tasks.
- Uses an authorized LLM to plan and execute code changes autonomously.
- Reports progress back to the dashboard.
Supported Models
Locus supports multiple LLM providers to drive the terminal agent. Select your provider for specific setup instructions:
- Claude CLI (Default)
- OpenAI Codex
Workflow
- Planning: The agent reads the task description and creates an implementation plan.
- Execution: It iteratively writes code, runs commands, and checks files.
- Review: Once complete, it marks the task as "In Review" or "Done" on the dashboard.
- Next Task: The agent automatically picks up the next task in the queue unless stopped.
Comparison: MCP vs Terminal Agent
| Feature | MCP Integration | Terminal Agent |
|---|---|---|
| Interface | Inside Editor (Cursor, VSCode) | System Terminal |
| Context | Shared with Editor (Open Files) | File System Only |
| Interaction | Chat-based guidance | Autonomous Loop |
| Best For | Pair Programming | Background Execution |
Future Updates
- Local Model Support: Running Locus with local models (e.g., Llama 3 via Ollama).
- Interactive Mode: A purely text-based interactive session within the terminal.