Most terminal coding agents are quietly bound to a single model vendor; this one started by deleting that assumption. It is a fork of Google's Gemini CLI (v0.8.2) that the Qwen team rebuilt around their own Qwen3-Coder models, then unhooked from upstream at v0.1 to develop independently — so the framework and the models can evolve together without locking you into either.
What Sets It Apart
- Provider-agnostic at runtime: the same agent drives OpenAI, Anthropic, Gemini, and Qwen APIs, plus any third-party or local model (Ollama, vLLM). You switch backends without switching tools.
- Parser and tool-call protocol re-tuned for Qwen-Coder rather than inherited unchanged — the real payoff of forking instead of just wrapping Gemini CLI.
- An agentic layer most CLI tools lack: subagents, agent teams, auto-memory, auto-skills, and MCP integration, so multi-step work keeps context instead of restarting each prompt.
- One core, many front ends — TUI, headless mode, IDE plugins (VS Code, JetBrains, Zed), a desktop app, daemon mode, SDKs, and IM bots — so it fits scripts and pipelines, not just an interactive shell.
Who It's For
A strong fit if you already run Qwen3-Coder, or want one open-source (Apache 2.0) agent you can point at whatever model is cheapest or best on a given day. The multi-interface surface suits teams wiring agents into CI or chat, not just typing in a terminal. Look elsewhere if you want a polished, opinionated single-vendor experience: the breadth of providers and modes means more configuration, and as a young independent fork its surface shifts fast across frequent releases.
