LCA (Local Code Agent)

Local AI agent system for task automation using LM Studio or Ollama. Similar to how Claude Code operates but for local execution and written in Rust for cross-OS compatibility.

Quick Start

1. Setup LM Studio

1. Download and install LM Studio
2. Load a model (recommended: Llama 3, Mistral, or Code Llama)
3. Start the local server (default port: 1234)

2. Build

cargo build --release

3. Run

# Interactive mode
./target/release/lca --provider lmstudio interactive

# Execute a task
./target/release/lca --provider lmstudio execute "create a hello world script"

# Allow all operations without prompting (use with caution)
./target/release/lca --provider lmstudio --allow-all interactive

Commands

# Interactive mode - best for exploration
lca --provider lmstudio interactive

# Execute a specific task
lca --provider lmstudio execute "your task here"

# Use a specific agent
lca --provider lmstudio agent shell "list files"

# Enable verbose logging
lca --provider lmstudio --verbose interactive

Permission System

By default, the agent will prompt you before:

• Writing files (shows content preview)
• Executing shell commands

Options when prompted:

• y - Allow this operation
• n - Deny this operation
• a - Allow ALL operations for this session
• q - Quit/cancel task

Use --allow-all flag to skip all prompts (automated mode).

Interactive Mode Features

• Arrow keys to navigate command history
• Ctrl+C or Ctrl+D to exit
• Type exit or quit to quit
• History saved to ~/.lca/history.txt

Using Ollama Instead

# Install Ollama
curl -fsSL https://ollama.ai/install.sh | sh

# Pull a model
ollama pull llama2

# Run with Ollama (default)
./target/release/lca interactive

Troubleshooting

"Connection refused"

• Ensure LM Studio local server is running
• Check that a model is loaded (not just available)
• Default port is 1234