Boxxy

A self-improving Linux terminal powered by AI characters.

Boxxy is a full terminal emulator with an agentic AI layer — BoxxyClaw — built directly into it. Unlike chat-window add-ons, BoxxyClaw has eyes on your shell, reads your terminal buffer, remembers your preferences, and can autonomously fix broken dependencies, manage system configs, and run your scripts. Just press Ctrl+/ and you're in.

---

Install

Boxxy is currently in Preview.

Native Nightly (Requires GTK 4.24, libAdwaita 1.10)

curl -sSf https://raw.githubusercontent.com/boxxy-dev/boxxy/main/scripts/install.sh | sh

Flatpak Nightly

curl -O https://boxxy-dev.github.io/boxxy-flatpak-remote/boxxy.gpg && \
  flatpak remote-add --user --if-not-exists --gpg-import=boxxy.gpg \
  boxxy https://boxxy-dev.github.io/boxxy-flatpak-remote/repo && \
  flatpak install --user boxxy dev.boxxy.BoxxyTerminal

---

Architecture

Crate	Role
boxxy-app	GTK4/Adwaita UI
boxxy-agent	Host-level daemon — PTY management, shell tools, character claims, D-Bus IPC
boxxy-vte	Headless terminal engine written in pure Rust
boxxy-claw	Agentic intelligence layer — characters, memory, skills, toolbox

---

Supported Providers

Provider	Support
Google Gemini	Full Gemini model catalog (with configurable thinking levels & native Google Search)
Anthropic	Full Claude model catalog (with configurable thinking levels & native web search)
OpenAI	Full OpenAI model catalog (with configurable reasoning effort & native web search)
DeepSeek	Full DeepSeek model catalog
Moonshot (Kimi)	Full Kimi model catalog (kimi-k3, kimi-k2.7-code, etc., with configurable thinking levels)
xAI	Full Grok model catalog (grok-4.7, grok-4.6, etc., with configurable reasoning effort)
Z.ai	Full GLM model catalog (glm-5.3, glm-5.3-flash, etc., with configurable thinking levels)
Mistral	Full Mistral & Codestral model catalog (mistral-large-latest, codestral-latest, etc.)
MiniMax	Full MiniMax model catalog (MiniMax-M3, MiniMax-M2.7, etc.)
Groq	Ultra-fast Groq model catalog (gpt-oss-120b, qwen3.8-27b, llama-3.3-70b, etc.)
Ollama	Any locally installed model (ollama run ...)
OpenRouter	Any OpenRouter model ID
OpenAI-Compatible Endpoints	Any custom /v1/chat/completions or /v1/responses endpoint (OpenCode Go, OpenCode Zen, LiteLLM, vLLM, llama.cpp, LM Studio, etc.) with live /models discovery, configurable context size, and thinking modes

Need support for another provider? Please open an issue to request it!

---

Documentation

Full docs at boxxy.dev

• Getting Started
• How It Works
• Characters
• Skills
• MCP Support
• Development

---

License

MIT