Vscode Coder Ollama

📖 Read in other languages / 阅读其他语言: 中文 | English

author：刘晨辉

Vscode Codex Ollama is a VS Code extension designed to provide a local agent experience close to the Codex extension workflow, while all model calls are routed entirely through your local Ollama service.

It currently supports:

• Reading the list of locally installed Ollama models
• Starting tasks from a Codex-style threaded interface in the sidebar
• Reading, creating, modifying, and deleting files in the current project through tool calls
• Strictly limiting file operations to the currently opened VS Code workspace
• Adding files and selections as context to the current thread

Feature Overview

• Codex Ollama Sidebar The sidebar includes a thread list, message stream, model selector, context attachments area, and Composer input area.
• Ollama Model Sync The extension retrieves installed models through the local Ollama API and lets you switch between them in the UI.
• Workspace Sandbox The agent's file tools only support relative paths within the current workspace and cannot escape the project directory.
• Inject Files/Selections into Thread You can add the current file or the editor selection to the thread so the model can read the context before modifying code.

Requirements

1. Ollama is installed and running
2. At least one model has been pulled locally, for example:

ollama pull qwen2.5-coder:7b

3. VS Code version 1.120.0 or later

Default Ollama Endpoint

The extension connects to the following endpoint by default:

http://127.0.0.1:11434/api

The following API routes are used:

• GET /api/tags
• POST /api/chat

If your Ollama address is different, you can change it in the settings.

Installation and Debugging

npm install
npm run compile

Press F5 in VS Code to launch the Extension Development Host.

After opening the project:

1. Open Codex Ollama from the activity bar on the left
2. Click "Refresh Models"
3. Select a local model
4. Describe your task in the input box and send it

Commands

• Open Codex Ollama Sidebar
• New Codex Ollama Thread
• Refresh Ollama Models
• Add File to Codex Ollama Thread
• Add Selection to Codex Ollama Thread

Settings

• vscodeCoderOllama.baseUrl Ollama API address, default: http://127.0.0.1:11434/api
• vscodeCoderOllama.defaultModel Default model name. If left empty, the first installed model is selected automatically.
• vscodeCoderOllama.requestTimeoutMs Ollama request timeout
• vscodeCoderOllama.maxToolIterations Maximum number of tool iterations for a single task
• vscodeCoderOllama.showThinking Whether to display the thinking text returned by the model

Permission Boundaries

This extension enforces workspace-level permission limits as requested:

• It can read files inside the current workspace
• It can create, modify, and delete files and directories inside the current workspace
• It cannot access paths outside the workspace
• Any ../, absolute path, or other path that attempts to escape the current project will be rejected

Notes on the Current Implementation

This version is already a runnable MVP, but it still differs from the official Codex extension in several ways:

• The UI style is close to Codex, but it is not a 1:1 replica
• The current thread UI is built with Webview rather than a deep integration with the official chatSessions
• The current toolset focuses on the file system and does not yet include advanced capabilities such as terminal execution, patch previews, or approval flows
• Whether tool calls work reliably still depends on whether the Ollama model you choose supports tool calling well

Known Limitations

• It works best with Ollama models that support tool calling and coding tasks
• Binary files are currently not recommended for agent modification
• write_file currently writes by fully overwriting the file instead of applying chunked patches

Development

npm run lint
npm run compile
npm run compile-tests
npm test