Ollama Native Coding Agent (OA) 🚀

This project is a lightweight, high-performance coding agent that completely separates the model server (e.g., Mac Studio) from the agent runner (Local Machine). It directly utilizes Ollama's native /api/chat (Tool Calling) API for seamless and efficient code interaction.

✨ Key Features

• Ollama Native Tool Calling: Directly uses Ollama 0.3+ tool calling capabilities without any additional compatibility layers.
• Stateful Multi-turn: Maintains conversation history to allow contextual code modifications and complex multi-step reasoning.
• Server-Client Separation: Decouples the heavy LLM inference from the agent logic, allowing for optimal resource utilization across your local network.
• Interactive REPL: Features a persistent command-line interface that allows you to chat and iterate on tasks continuously.

🛠 Installation (Global CLI Setup)

1. Clone the repository and enter the directory

   git clone <your-repo-url>
   cd LocalCode

2. Install the package in Editable mode This command registers the oa command globally on your system.

   pip install -e .

🚀 Usage

Basic Execution

Open your terminal in any directory and run:

oa --host "http://[MAC-STUDIO-IP]:11434" --model "qwen3-coder:30b"

Pro Tip: Setup an Alias

To avoid typing the host and model every time, add an alias to your .zshrc (or .bashrc):

# Add this to your ~/.zshrc file
alias codex='oa --host "http://192.168.1.100:11434" --model "llama3.1:8b"'

After saving, simply type codex in your terminal to summon your AI coding assistant.

🧰 Capabilities (Tools)

• search_code: Search for specific code patterns within the project.
• read_file: Read file contents with line numbers for context.
• propose_patch: Find a specific code block and apply precise modifications.
• run_test: Execute shell commands and run tests.
• final: Report task completion and provide a summary of changes.

⚠️ Important Notes

• Ollama Version: Recommended version 0.3.0 or higher.
• Model Support: Requires models that officially support Tool Calling (e.g., llama3.1, mistral-nemo, qwen2.5-coder).
• Server Configuration: Ensure the Ollama server on your Mac Studio allows external connections (OLLAMA_HOST=0.0.0.0).

---

Since this agent can directly modify your source code, it is highly recommended to have a clean git commit state before running tasks!