CLI harness

LM Studio

Agent harness in your terminal

Caprigo embeds the agent against LM Studio — multi-pane HUD, HOME missions, desktop body, and tool loops that finish work on your machine. No gateway. No browser dashboard required.

RuntimeIn-processEmbedded agent

BackendLM StudioOpenAI-compatible

SurfaceCLI HUD5-pane TUI

Get started Download GitHub

Caprigo — embedded HUD

Caprigo CLI HUD with LM Studio, tool cards, and context caps

Why Caprigo

A harness that runs tools, not a chat wrapper.

Most AI products hide execution behind a message bubble. Caprigo shows every tool call, cap status, and mission verification in a terminal HUD built for operators.

Embedded runtime

Agent + LM Studio backend in one process. No server to babysit — `caprigo` is the product.

HOME missions

Harness-owned playbooks bootstrap before the first model turn and auto-drain when models ignore tools.

Multi-pane HUD

Agents, live session log with tool cards, Context caps, and double-Enter STEER input.

Digital body

Windows desktop_*, WinRT OCR, Playwright browser, shell execution — real tools on your machine.

Brain + stumble

Sticky lessons, failure recovery, and model dialect profiles so tool loops keep moving.

MIT · local-first

Your model, your files, your permissions. Optional Vibes marketplace imports when you want them.

Missions

HOME playbooks that finish real work.

The harness compiles intent into playbooks — notepad type, web answer, write file — and executes bootstrap steps before the first LLM reply.

Desktop

Notepad mission — HOME playbook

Harness compiles intent, runs desktop_focus → type → screenshot/OCR, verifies without waiting on the model.

Web

Search and fetch loop

web_search + web_fetch with live tool cards while the harness keeps the turn moving.

Write

Files + browser preview

write_file with mkdir parents, then browser_open to preview generated HTML.

Install

Four steps to a running HUD

Windows: setup.ps1 -LaunchHud wraps everything. macOS/Linux: npm install, npm run build, caprigo connect --launch.

Step 01

Clone and install

Node.js 18+. One command on Windows wraps install, build, and guided .env setup.

git clone https://github.com/doteyeso-ops/caprigo
cd caprigo
.\setup.ps1 -LaunchHud

Step 02

Start LM Studio

Enable Local Server (port 1234) and load a tool-capable model.

# LM Studio → Local Server ON
# Load e.g. qwen2.5-coder-7b-instruct
caprigo connect --launch

Step 03

Configure .env

OpenAI-compatible defaults for the embedded harness.

CAPRIGO_LLM_PROVIDER=openai_compatible
OPENAI_BASE_URL=http://127.0.0.1:1234/v1
DEFAULT_MODEL=qwen2.5-coder-7b-instruct
CAPRIGO_HARNESS_MODE=1

Step 04

Launch the HUD

Embedded agent runs in-process — no gateway, no browser dashboard.

.\launch-hud.ps1
# or: caprigo tui
caprigo doctor

CLI helpers: caprigo doctor, caprigo connect, caprigo setup --interactive, /bug for handoff packs

FAQ

Common questions

Do I need a gateway or web UI?

No. The default path is the embedded CLI HUD. Agent runs in-process against LM Studio.

Which models work best?

Tool-capable models like qwen2.5-coder-7b-instruct. HOME auto-drain helps when models ignore native tools.

Windows only?

The HUD runs on macOS/Linux. Desktop body skills are Windows-first; browser and file tools work cross-platform.

How is this different from chat UIs?

Caprigo is a harness: structured panes, tool cards, mission verification, and operator caps — not a single chat box.