Vendor agent CLIs: tracking first-party AI coding tools from model providers

The AI agent CLI space has been moving at a pace that makes it genuinely difficult to keep track of who's shipping what, and I realised I needed a reference document just to maintain my own mental model of the landscape. Specifically, I wanted to track which model vendors are maintaining their own first-party command-line agent tools — not third-party wrappers, not community projects, not IDE plugins, but the tools that the vendors themselves build, ship, and support. The distinction matters because first-party tools have a uniquely close integration with the underlying model that third-party wrappers typically can't replicate — things like optimised context management, early access to new model capabilities, and alignment between the tool's assumptions and the model's actual behaviour. The Vendor Agent CLIs repository is the result.

The current landscape (and it's smaller than you'd think)

As of the latest snapshot, the vendor-maintained CLI list includes eight tools: Claude Code from Anthropic (which I use daily and is my primary development environment), Codex from OpenAI, Gemini CLI from Google, Amazon Q from AWS, Groq Code CLI from Groq, Kimi CLI from Moonshot AI, Mistral CLI from Mistral AI, and Qwen Code from Alibaba. That's it — eight vendors shipping their own agentic CLI tools out of the dozens of companies building foundation models. The number is growing but it's more contained than you might expect given the hype around agent tooling. There's a common pattern I've observed: a vendor creates a CLI, it stays elementary and undocumented until it develops a cult following among power users, an ecosystem of plugins and configurations grows around it, and then media attention follows. Claude Code's trajectory from quiet beta to industry standard is the canonical example.

The list deliberately excludes third-party tools like Aider, Continue, and Cursor — not because they're worse (several are excellent), but because the scope is specifically vendor-maintained first-party tools. It also excludes IDE plugins that wrap vendor APIs, community forks, and tools that only offer API access without agentic capabilities like tool use and MCP support. The repository includes an overview of benchmarks used to evaluate these agentic coding tools, and given how fast this space moves, it's maintained with periodic snapshots rather than claiming to be exhaustive at any given moment. Check it out on GitHub.