Lumoswitch Docs
Agent setup

Agent setup

Each Agent page explains what the Agent is for and maintains its complete connection lifecycle: one-time launch, persistent use, everyday launch, and removal. When a mode is unavailable, the page explains why; you should not need to assemble one setup from several documents.

Before starting, have the API configuration Base URL, Lumoswitch Access Key, and client-facing model name ready. The Agent protocol must match the configuration output protocol. Start with General Agent setup if you are unsure.

Codex

How you use CodexSetup page
Only the Codex workspace in ChatGPT desktopChatGPT Desktop (Codex)
Only Codex CLI, isolated from desktop settingsCodex CLI
ChatGPT desktop, CLI, and IDE with shared settingsChatGPT Desktop + Codex CLI

If you do not know which path fits, start with Choose your Codex setup.

Command-line Agents

AgentTypical downstream protocolSetup page
Claude Code CLIAnthropic MessagesOpen setup
OpenCode CLIOpenAI-compatibleOpen setup
Mistral Vibe CLIOpenAI-compatibleOpen setup
DeepSeek HarnessOpenAI-compatibleOpen setup
GitHub Copilot CLIOpenAI-compatibleOpen setup
Gemini CLIGeminiOpen setup
Qwen Code CLIOpenAI-compatibleOpen setup
Aider CLIOpenAI-compatibleOpen setup
OpenClaw CLIOpenAI-compatibleOpen setup

Official interfaces available alongside these CLIs

AgentOfficial desktop or editor surfaceRelationship to Lumoswitch setup
Claude CodeClaude Desktop, VS Code, JetBrainsDesktop uses separate persistent setup; IDEs can connect to CLI or shared settings
OpenCodeOpenCode Desktop and IDE extensionGraphical surfaces read global settings and save the Key persistently
Mistral VibeVS Code, Web, and ACPThe local VS Code extension shares CLI settings; Web sessions do not inherit local credentials
DeepSeek HarnessLocal Web UINative DSH_HOME settings point the original command to a dedicated Lumoswitch configuration
GitHub Copilot CLICopilot app, VS Code, GitHub Desktop/ide connects the current CLI; other graphical surfaces configure BYOK separately
Gemini CLIIDE Companion and ACP editorsNo standalone coding desktop app; Companion reuses the current CLI process
Qwen CodeVS Code extension and IDE CompanionNo standalone desktop app; Companion reuses CLI, while the native sidebar uses user settings
AiderLocal browser UI and IDE terminalNo standalone desktop app or official IDE plugin; both reuse the Aider process
GooseGoose Desktop and ACP editorsDesktop reuses the custom provider but stores its Key separately in system credentials
OpenClawmacOS companion app and Control UIBoth reuse persistent settings from the same Gateway

Each linked page explains whether the graphical surface supports one-time setup, how persistent setup works, and whether removal belongs in the CLI, application settings, or a shared Gateway. Pages also state when no official desktop app exists, so an ordinary chat product is not mistaken for a coding Agent.

Editor and other Agents

AgentSurfaceSetup page
Zed AgentNative editorOpen setup
Roo CodeVS Code and JetBrainsOpen setup
ClineVS Code and JetBrainsOpen setup
ContinueCLI and editorOpen setup
Kilo CodeVS CodeOpen setup
CrushTerminalOpen setup
GooseDesktop and terminalOpen setup

For an Agent without a dedicated page, follow General Agent setup when it supports a custom Base URL, Access Key, model name, and matching protocol. Non-Agent integrations such as the OpenAI SDK and OpenWebUI now live under API and compatibility in the sidebar.

On this page