GG Coder CLI
Terminal coding agent with a short system prompt, twelve providers and switching models in mid-conversation
- 293 ★ on the repository
- Last commit
- AGPL-3.0
Broad access
Why this level
- The agent edits files and runs commands in the project directory, and Ken Autopilot mode keeps going without stops until the result is confirmed
- Anthropic subscription sign-in goes through OAuth from a third-party client, so compare your subscription terms with the vendor rules: an account block is possible
- OAuth tokens and keys live in `~/.gg/auth.json`, shared with the desktop app
Install
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/gg-coderDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent edits files and runs commands in the project directory, and Ken Autopilot mode keeps going without stops until the result is confirmed
- Anthropic subscription sign-in goes through OAuth from a third-party client, so compare your subscription terms with the vendor rules: an account block is possible
- OAuth tokens and keys live in `~/.gg/auth.json`, shared with the desktop app
This is third-party code. Review the repository files before installing.
What it does
GG Coder starts in the terminal with `ggcoder` after `npm i -g @kenkaiiii/ggcoder`. Its system prompt is about 1,100 tokens, against roughly 15,000 for Claude Code by the author's measurement, it ships no built-in MCP servers and keeps a narrow toolset. For checking work against live code it has its own `steroids` tool, which finds similar production repositories and loads only when the agent asks. You change the model with `/model` in the middle of a conversation. Providers are Anthropic, OpenAI, Gemini, xAI, Moonshot, GLM, MiniMax, Xiaomi, DeepSeek, Sakana, OpenRouter, Hugging Face and local models. Anthropic, OpenAI and Gemini sign in through OAuth, the others need an API key, and tokens live in `~/.gg/auth.json`. The same engine runs in a Tauri desktop app with Code, Chat and Motion modes for video. You add your own MCP servers through `~/.gg/mcp.json` or `.gg/mcp.json`.
Who it is for. Developers who pay for AI subscriptions and want an agent with minimal overhead in context and a free choice of model mid-session.
Good fit when
- A session hits subscription limits fast and you want to cut extra instructions and tools from the context
- Work starts on one model and continues on another without losing the conversation
- Cloud and local models should work inside one agent
- Code is written against an unfamiliar library and the agent benefits from checking idioms against real repositories
Not a fit when
- The agent has to embed into Cursor or Claude Code as a plugin: GG Coder starts as a separate program
- You want a set of ready MCP servers out of the box: none ship by default and you add them by hand
- The provider subscription terms forbid signing in through a third-party client
Example request
Find the places in the project that use the outdated client call, fix them, run the tests and finish by listing what is still unconfirmedLimitations
The terminal version needs only Node.js and the `@kenkaiiii/ggcoder` package, and the desktop app downloads separately for macOS on Apple Silicon, Windows or Linux x86_64. OAuth sign-in requires a subscription with the provider itself, and the others work with an API key. The author took the comparison with 19 other agents in one race on 25 October 2026: 7 tasks, 5 runs each, on Claude Haiku 5.5. Subscription usage figures depend on your plan and tasks, and you can reproduce them only with your own measurements.
How to disable. Remove the package with `npm uninstall -g @kenkaiiii/ggcoder` and the `~/.gg/` folder that holds sign-in data and settings. Uninstall the desktop app with your system tools.
FAQ
How do I install GG Coder in the terminal?
Run `npm i -g @kenkaiiii/ggcoder`. Then `ggcoder login` to choose a provider and sign in, and `ggcoder` to start in the project directory.
Which models are supported?
Twelve providers: Anthropic, OpenAI, Gemini, xAI, Moonshot, GLM, MiniMax, Xiaomi, DeepSeek, Sakana, OpenRouter and Hugging Face, plus local models. You switch with `/model` in the middle of a conversation.
Why are there no built-in MCP servers?
The author holds that descriptions of a dozen servers overload the context and blur tool choice. GG Coder has the `steroids` tool instead, and you add your own servers to `~/.gg/mcp.json` or `.gg/mcp.json`.
How does the terminal version differ from the desktop app?
The engine is shared and both use the sign-in data in `~/.gg/auth.json`. The app adds the Code, Chat and Motion modes and GG Goals, where a team of workers sees a goal through.
Where do the comparison figures against other agents come from?
From the author's own run of 25 October 2026: 7 tasks, 5 runs each, every agent on Claude Haiku 5.5. In his table GG Coder finishes the suite in 99 seconds against 141 for Claude Code, with lower subscription usage.
Discussion
No questions or reviews yet.
Related
CLI toolsA terminal coding agent in Go: GLM, DeepSeek and local Ollama models, 59 built-in tools and MCP
Gajae-Code is a terminal coding harness that logs into a coding plan you already pay for and plans changes before mutating code.
GigaCode Proxy for editors: Cline, Roo Code, Continue
GigacodeOpenaiProxy
A dependency-free proxy connects GigaCode to Cline, Roo Code, Continue and Open WebUI, repairs tool call cuts and signs in with SberID
GigaCode project flows with hooks, gates and OpenSpec
gigacodeAgents
A monorepo of flows for the GigaCode CLI: development with stages and hard gates, a work progress observer and reverse analysis of code into a spec
A Python terminal coding agent in the manner of Claude Code: reads the repository, edits files, runs commands, with memory, sub-agents and MCP
A desktop board turns a card into an agent session in its own git worktree, flags the agent waiting on you, offers an MCP server
Comparison with similar tools
Cards that share the main search query, and how they differ.
README in short
The main page presents the desktop app, a benchmark table for 20 agents and download links for three systems. Below it describe the `gg-ai`, `gg-agent`, `gg-core` and `ggcoder` packages for building your own agents and invite readers to the author's community.