One CLI to juggle every AI account you're hoarding

What it does CCS wraps Claude Code and compatible CLIs in a single stable command surface. You keep multiple Claude subscriptions, OAuth providers (Codex, Kiro, Qwen, Kimi, and more), and API/local-model profiles (GLM, OpenRouter, Ollama, llama.cpp) as named profiles, then launch whichever fits the task. A local Anthropic-compatible proxy bridges Claude Code into OpenAI-compatible providers, so you’re not locked to a native Anthropic upstream. A React dashboard shows usage, costs, auth health, and quota state across profiles.

The interesting bit The proxy supports request-time profile:model selectors and scenario-based routing, and the SSE transformation work was openly informed by claude-code-router’s architecture — the README even tells you when to use CCR instead. It also provisions tools like WebSearch and image analysis for third-party launches rather than leaving you to wire them yourself.

Key highlights

• Switch between runtimes (Claude Code, Codex CLI, Factory Droid) and isolated account contexts in seconds
• Local proxy bridges OpenAI-compatible providers, with round-robin vs fill-first surfaced in both CLI and dashboard
• Live auth and health monitoring so you can see account state without digging into raw config
• Runs via Docker or as a global npm package; community-maintained CLIProxyAPIPlus fork is opt-in for plus-only providers
• OpenCode sync plugin keeps a second tool aligned with your live CCS setup

Caveats

• The old ghcr.io/kaitranntt/ccs-dashboard:latest Docker image is deprecated in favor of ghcr.io/kaitranntt/ccs:latest — worth checking before you deploy
• The README is deliberately thin; real detail lives on a separate docs site, so expect some hopping around

Verdict If you rotate between multiple AI subscriptions and providers daily, this solves a genuinely annoying problem. If you’re happily on a single Claude account, it’s a lot of machinery for nothing.