Meet xllama

xllama is a terminal LLM agent: a rich Bubble Tea TUI on top of a chat-completions engine, running locally through llama.cpp by default or through any OpenAI-compatible provider.

The 30-second model

you ─▶ Chat (history) ─▶ Engine (llama.cpp | remote) ─▶ stream
             ▲                    │
             └── results ◀── Tools + MCP + Agents + Skills

CHAT

Keeps the running conversation and drives the ReAct tool loop.

ENGINE

Swaps between local llama.cpp and OpenAI-compatible endpoints.

TOOLS

Act on files, shell and web with explicit consent for dangerous work.

EXTENSIONS

Add MCP servers, skills, agents, plugins and custom commands at runtime.

Install in one line

curl -fsSL https://raw.githubusercontent.com/Valerio357/xllama/main/install-remote.sh | sh

The installer downloads a prebuilt binary for macOS, Linux or Android/Termux and falls back to a source build when Go is available. It installs to ~/.local/bin by default.

xllama setup --model <owner/repo:quant | /path.gguf>
xllama

Where everything lives

~/.xllama/
  config.json
  keys/<provider>
  plugins.json
  sessions/*.json
  agents/<name>.md
  skills/<name>/SKILL.md
  commands/<name>.md
  marketplaces/<name>/
  plugins/<name>/