Ho!

Agentic desktop launcher — MCP tools, LLM agents, and embedded scripts behind a global hotkey.

Ho is a Tauri app (React + Rust). Open the launcher, pick a mode, run tools or an agent against connected Model Context Protocol servers.

Headless CLI: same ho binary supports operator subcommands (doctor, agent, runs, …). See CLI.

Features

• Launcher — local keyword tool search and direct MCP execution
• Agent — multi-step runs with guardrails, user approvals, durable event timeline, conversation history
• Vibe — natural language → LLM-generated Python or JavaScript (in-process runtimes)
• MCP — stdio, HTTP, and built-in OS Capabilities; OAuth for remote servers
• LLM — OpenAI, Azure OpenAI, Ollama, Thinkers
• Desktop shell — system tray, transparent always-on-top window, 32 themes, EN/ZH i18n

Default hotkey: cmd+` (macOS) / ctrl+space (Windows, Linux).

Requirements

• Node.js 18+
• Rust (stable)
• Platform toolchain for Tauri (Xcode on macOS, etc.)

Quick start

git clone https://github.com/thinkers-cool/Ho.git
cd Ho
npm install
npm run dev

Optional: copy .env.example → .env for build-time analytics and landing-page footer overrides. Leave unset for OSS builds.

Variable	Purpose
APTABASE_APP_KEY, APTABASE_HOST	Backend product analytics (compile-time)
VITE_CLARITY_PROJECT_ID	Microsoft Clarity on landing build only
VITE_LANDING_COPYRIGHT, VITE_LANDING_LICENSE_URL	Landing page footer text and license link

Scripts

Command	Purpose
npm run dev	Full desktop app with hot reload (Tauri)
npm run dev:web	Frontend Vite dev server only
npm run build	Production bundle for current OS
npm run typecheck	TypeScript check (frontend)
npm run lint	Frontend dependency-cruiser + Rust clippy
npm test	Frontend Vitest + backend cargo test --lib
npm run test:e2e:ui	Browser E2E via Playwright (no LLM)
npm run test:e2e:cli	Headless CLI E2E (live LLM, needs API key)
npm run ci	typecheck + lint + test (CI gate)

Frontend-only commands (run from repo root with -w frontend, or cd frontend): dev:landing, test:unit, test:integration, test:watch, dev:e2e.

Project layout

Ho/
├── backend/     Rust/Tauri backend, prompts, capabilities
├── frontend/    React UI
├── docs/        Documentation
├── test/        CLI E2E scripts and fixtures
├── build/       Platform build scripts
├── scripts/     Version and verify helpers
├── app-data/    Local config, runs, cache, and logs (debug builds)
└── EULA/        End-user license

Configuration

On first run, settings and local data are created under the app data directory:

Build	Path
Debug	./app-data/
Release	OS config dir → thinkers/ho/

Notable paths under app data: settings.json, split registry JSON files, runs/, events/, conversations/, sessions/, pending-runs/, cancel-requests/, resume-requests/, app_cache.json (App Launcher metadata), token-ledger.jsonl, logs/ho.log.* (rotating trace logs).

Configure LLM providers, MCP servers, agent limits, skills, and guardrails in the in-app Settings panel (or edit JSON directly). API keys support plaintext or env:VAR references.

Documentation

Doc	Contents
Architecture	System overview, state model, modules
Design	Module map and persistence reference
Workflow	Startup and runtime flows
CLI	Headless operator commands
MCP	MCP server setup
Agent	Agent settings
Authorization	Credentials and approvals
Integration	External systems and APIs
UI	Frontend structure
Test	Test commands and coverage
Memory	Conversation and event storage
Skills	Skill registry
Budget	Token limits
Parallelization	Concurrency model
Terminology	Glossary

Contributing

1. Fork the repo
2. Create a branch from main
3. Make changes; run npm run ci (or npm test for a faster check)
4. Open a pull request with a clear description

License

MIT — Copyright (c) 2026 Shanghai Thinkers Technology Co., Ltd.

End-user terms: EULA (English) · EULA (简体中文)