keen

Minimal terminal AI coding agent with pluggable LLM providers

TLDR

Start the agent in the current directory

$ keen

Run without a global install via npx

$ npx keen-code

Check the installed version

$ keen --version

Switch to a different LLM provider mid-session (inside the agent)

$ /model

Adjust the extended-reasoning effort level (inside the agent)

$ /thinking

Compress the conversation when nearing the context limit (inside the agent)

$ /compact

keen (distributed as the keen-code npm package) is a small, terminal-based coding agent written in Go. It treats the AI model as a generic backend and ships a deliberately tiny tool surface: read_file, write_file, edit_file, glob, grep, and bash. That is the entire toolbox the agent uses to navigate, edit, and run code, by design favouring simplicity over feature breadth.Invoking keen with no arguments drops you into an interactive session anchored at the current working directory. The agent reads files, edits them in place, and runs shell commands on your behalf, conversing through the terminal. Slash commands such as /model, /thinking, and /compact control provider selection, reasoning depth, and context compaction without leaving the session.The agent is provider-agnostic: it can drive Anthropic, OpenAI (including ChatGPT OAuth), Google AI (Gemini), DeepSeek, Kimi, GLM, MiniMax, Codex, and the Go port of OpenCode. Switching providers is done at runtime with /model rather than via separate CLI flags, so a single keen install covers every supported backend.A notable trait of the project: every line of code in the repository is reported to have been written by an AI agent, with the prompts and design decisions archived under .ai-interactions/.

PARAMETERS

--version

Print the keen-code version and exit. Also reachable as npx keen-code --version for users who skip the global install.

SLASH COMMANDS

/model

Switch between configured LLM providers without restarting the session.

/thinking

Adjust the extended-reasoning effort level for providers that support it.

/compact

Compress the running conversation when the context window fills up.

INSTALLATION

Via npm (recommended, installs the keen binary on PATH):

$ npm install -g keen-code

Via the install script (drops the binary in /usr/local/bin or $HOME/.local/bin):

$ curl -fsSL https://raw.githubusercontent.com/mochow13/keen-code/main/scripts/install.sh | bash

One-off run without installing:

$ npx keen-code

CAVEATS

The minimal six-tool design is intentional and means keen does not expose specialised helpers other agents bake in (web fetching, structured planning tools, sub-agents). Workflows that depend on such features may need a different agent.Each LLM provider must be configured with its own credentials. Costs and rate limits accrue against whichever provider is currently selected via /model.

HISTORY

keen-code is a 2026 project (latest release v0.20.3 as of June 2, 2026) published under the MIT license by mochow13 on GitHub. It is part of the broader wave of minimalist terminal coding agents inspired by Claude Code, Codex, and OpenCode, distinguishing itself by limiting the tool surface and by being entirely AI-authored.