INSTALL ↓

local-first agent runtime

xllama

A simple harness for your agents.

A fast terminal agent with a rich TUI, powered by llama.cpp and built for any GGUF model—or any OpenAI-compatible provider.

$curl -fsSL https://raw.githubusercontent.com/Valerio357/xllama/main/install-remote.sh | sh
Read the wikiView on GitHub
Pixel-art xllama agent mascot
BUILD
READY
STREAMING◆TOOL CALLS◆PLAN MODE◆MCP◆SKILLS◆AGENTS◆PLUGINS◆OPENAI-COMPATIBLE API

// ONE ENGINE. MANY MISSIONS.

Small enough to understand.
Powerful enough to ship.

01

Run local. Stay yours.

Start and manage llama.cpp automatically, or point xllama at a local GGUF model. Your sessions, config and keys live on your machine.

02

Give agents real tools.

Read, search, edit and run commands with consent controls, visible diffs and one-command undo built into the loop.

03

Extend the harness.

Connect MCP servers, load reusable skills, delegate to specialist agents and install complete plugin bundles without leaving the TUI.

// THE 30-SECOND MODEL

Your prompt enters.
Work comes out.

xllama keeps the conversation, streams the model response, executes approved tools and feeds results back until the mission is done.

Explore the architecture →
01YOUprompt
message
02CHATagent loop
stream
03ENGINElocal or remote
TOOLSread · edit · bash · web · MCP↻ results
~/your-project — xllamaBUILD
llama.cpp · local runtimegpt-oss-20b

type a message, or / for commands · /mcp /plugins /agents /skills

YOU

Build the feature and run the tests.

◆ Thinking

Inspecting the project and planning the smallest safe change…

▸ edit_filesrc/agent.ts+24 −7
ASSISTANT

Feature implemented and verified. All tests pass.

● readyctx 18%   BUILD
❯ask anything

// BUILT FOR THE TERMINAL

See every move.
Stay in control.

  • 01 Streaming answers and reasoning
  • 02 Collapsible tool and subagent blocks
  • 03 Consent gates, diffs and undo
  • 04 Persistent sessions per project

// PRESS START

Give your local models
a mission.

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

macOS & Linux · installs to ~/.local/bin