Skip to main content
The desktop app manages multiple runtimes through one engine registry. The picker under the input decides which one this request uses. Open Settings → CLI configuration. The left list is grouped as Enabled / Disabled / Coming soon. Move an engine to Disabled and it no longer appears in the input.

​
Currently available

Detect, one-click install, and PATH troubleshooting: Detect and install.

​
Coming soon

Gemini CLI and entries such as Cursor / Copilot / Qwen appear in the Coming soon group. Do not write config flows as if they were shipped runtimes.

​
How to choose

  • Full permission dialogs, plan, MCP / Skills: prefer Claude Code
  • OpenAI / Codex ecosystem: use Codex
  • You already use OpenCode / Grok / Kimi / Pi in a terminal: switch to that engine
  • You already run DeepSeek Harness: use DSH
You can switch providers in the same workspace. History is isolated per runtime.

​
Config entry

Open Settings → CLI configuration:
  • Hosted providers for Claude / Codex / Kimi / Grok / OpenCode
  • Pi auth.json and subscription-auth status
  • DeepSeek Harness host / port / auto-start
  • Local binary detect, one-click install, and custom paths
  • Custom models and prices (for usage stats)
The provider catalog is ~/.ccgui/config.json. After you enable one: Claude does not change ~/.claude/settings.json; Codex writes ~/.codex/; Kimi / Grok write their own config.toml (backed up as .bak first); OpenCode uses env injection and does not change opencode.json.

Claude Code

API key, local config, and login.

Codex

~/.codex/ and sandbox.

Kimi

config.toml and third-party providers.

Grok

config.toml; use WSL on Windows.

OpenCode

opencode.json, env injection.

Pi

API key and /login.

DeepSeek Harness

host / port and Web UI.

Detect and install

Install plan, PATH, and troubleshooting.