Ollama Sidebar

A local AI coding assistant in your VS Code sidebar. Chat with Ollama models, attach code, and ask the assistant to inspect or modify your project. Ollama runs inference locally; Codex app-server provides the coding tools.

This project was developed with OpenAI Codex as a reference and uses its app-server protocol and runtime for agent execution. Ollama Sidebar is an independent project and is not an official OpenAI extension.

한국어 사용법

Features

• Stream responses and view command output and file changes.
• Attach the current file or selected code (up to 32 KiB combined).
• Approve or decline action requests and interrupt tasks.
• Save and restore conversations for each workspace.
• Choose installed Ollama models that support tool calling.

Requirements

• Windows x64 and VS Code 1.106.0 or newer.
• Ollama with a tool-capable model; the default is gemma4:e4b.
• Codex 0.150.1, installed using the setup script below.
• Node.js 22 or newer for setup and source builds.
• A trusted local workspace. WSL, SSH, and virtual workspaces are not supported.

Installing the extension does not install Ollama, Codex, or model weights. Complete runtime setup before starting a conversation. The interface currently uses Korean labels; their meanings are explained below.

Runtime setup

Install Node.js and Ollama, then run these commands in PowerShell:

git clone https://github.com/startour812-sudo/ollama-sidebar.git
cd ollama-sidebar
powershell -NoProfile -ExecutionPolicy Bypass -File .\scripts\install-runtime.ps1
ollama pull gemma4:e4b

The script downloads the official Codex 0.150.1 Windows npm package to %LOCALAPPDATA%\SelfAgent\.runtime\vendor. It reuses a matching installation and stops without overwriting incompatible or incomplete installations. Runtime and model downloads require internet access.

The extension starts a separate Ollama server at http://127.0.0.1:11435 when needed. To reuse an existing server, set selfAgent.ollamaUrl to its loopback address, such as http://127.0.0.1:11434.

Getting started

Install Ollama Sidebar from VS Code's Extensions view, or install a VSIX built from source. Open a trusted project folder and run Ollama Sidebar: 채팅 열기 (Open Chat) from the Command Palette. Select a workspace folder when prompted.

• Click the model name to select a model. Switching models saves the conversation and starts a new one.
• Use 파일 첨부 (Attach File) or 선택 코드 (Selected Code) to provide context.
• Press Enter to send or Shift+Enter for a newline. Enter during Korean text composition does not send.
• Review approval requests before accepting or declining them.
• Use 중단 (Stop) to interrupt a task. Completed file changes remain in place.
• Use 대화 목록 (History) or ＋ to restore a conversation or start a new one.
• Use 로그 (Logs) or the Ollama Sidebar output channel to inspect logs.

Settings and storage

Search for selfAgent in VS Code Settings. Existing setting and command IDs are retained for compatibility.

Conversations, attached code, logs, and runtime configuration are stored locally in the extension's VS Code globalStorage, separated by workspace. Existing ~/.codex settings are not used. History from the former Self Agent extension is not automatically migrated.

The runtime uses on-request approvals, workspace-write file access, and the Windows unelevated sandbox. Automatically started servers remain available for reuse; idle models unload after five minutes. Image input, web search, plugins, and multiple agents are not supported.

Build and development

From the repository root:

npm ci
npm run check
npm test
npm run build
npm run test:integration
npm run test:live
npm run package
code --install-extension .\artifacts\ollama-sidebar-0.1.5.vsix

check validates TypeScript; test runs Node unit tests; build bundles the extension and webview. package creates a VSIX under artifacts/.

Integration tests use an isolated VS Code host and download VS Code on the first run. Set SELF_AGENT_TEST_CODE to a VS Code executable to reuse an installation. Live tests require the runtime and default model and validate file copying, reading, and thread reconnection. They may take several minutes; run them from a normal user terminal to avoid nested Windows sandbox errors.

Open this repository in VS Code and press F5 to launch the development host. Sources are in src/, tests in test/, UI assets in media/, and model information in resources/. src/protocol/ contains generated Codex types; runtime upgrades require regeneration and integration testing.

Troubleshooting

• Runtime missing or wrong version: run the setup script. Only Codex 0.150.1 is supported; other versions are not replaced automatically.
• Model missing or unsupported tools: check ollama list, download the model, and select one with the tools capability.
• Server startup failed: check the executable, configured port, and output logs.
• Slow first response: CPU inference and model loading may take several minutes. Keep prompts short and attach only relevant code.
• Command not visible: run Developer: Reload Window.
• Nested sandbox failure: run tests from a normal user terminal.

License

The extension code is MIT licensed. See third-party notices for generated Codex types and dependencies. Codex, Ollama, and model weights retain their respective licenses and are not included in this repository.