Local Coding Agent

An autonomous AI coding agent integrated directly into VS Code as a lightweight, secure, and private extension. It operates entirely on your local machine, communicating directly with your local Ollama server. No cloud backends, no telemetry, and complete privacy for your codebase.

Key Capabilities

1. Local Agent Orchestration: The entire reasoning loop and tool execution engine run within the VS Code Extension Host (Node.js). There is no separate Python or Flask backend.
2. Transparent Reasoning & Thinking: Streams and displays reasoning/thinking tokens (such as from DeepSeek-R1 or reasoning-enabled Qwen models) in real-time inside a collapsible Thinking… details element that auto-collapses when the final response starts.
3. Interactive Tool Permissions (Copilot-Style): Safe execution of destructive and modifying operations. Before writing/editing files or running terminal scripts, the agent prompts you in the chat view with Allow and Deny buttons.
4. CORS Bypass: Direct host-level execution prevents browser CORS errors when connecting to the local Ollama API.

Settings & Setup

The extension is designed exclusively to work with a local instance of Ollama.

• Default Ollama Endpoint: http://localhost:11434
• Configuration Key: aibot.ollamaUrl
• Custom Endpoints: If your Ollama server runs on a different port, container, or local address, you can configure it via VS Code Settings:
  1. Open VS Code Settings (Ctrl+,).
  2. Search for Ollama Url.
  3. Enter your custom address (e.g., http://127.0.0.1:11434).

Available Agent Tools

The agent is equipped with native tools to inspect and modify your active workspace workspace:

📂 File Operations

• read_file: Inspect contents of any file in the workspace.
• write_file (Requires Permission): Create a new file or completely overwrite an existing one.
• edit_file (Requires Permission): Performs precise find-and-replace text swaps.
• delete_file (Requires Permission): Permanently deletes a file.
• get_file_info: Read metadata such as file size, path type, and modified dates.

📁 Directory Operations

• create_folder: Creates folders and nested parents.
• delete_folder (Requires Permission): Recursively removes a folder and all contents.
• list_directory: Lists files and folders inside any sub-folder.
• search_files: Recursively searches files matching glob patterns (e.g., *.js).

💻 Terminal execution

• run_terminal (Requires Permission): Executes standard shell commands (e.g., package installation, build compilation, test suits) and streams console lines directly into the webview.
• Persistent Interactive Sessions (Requires Permission): start_terminal_session, terminal_input, terminal_key, get_terminal_output, and stop_terminal_session allow the agent to run interactive terminal workflows.

⏱ Utilities

• get_current_datetime: Queries system date and time.

Setup & Running Instructions

1. Ensure Ollama is running locally on your machine (ollama serve).
2. Download your preferred model (e.g., ollama run qwen3.5:4b or ollama run deepseek-r1:8b).
3. Install the packaged extension ollama-coder-agent-3.2.1.vsix by opening VS Code, navigating to the extensions pane (Ctrl+Shift+X), clicking ... (More Actions), and selecting Install from VSIX...
4. Open your project workspace, click the AI Bot chat icon in the sidebar, select your model from the dropdown, and start coding!