HAPI provides a local-first framework for running AI coding sessions (Claude Code, Codex, Gemini) on any machine. It enables remote monitoring and control via Web, PWA, or Telegram Mini App, offering a flexible alternative for developers needing to manage AI coding workflows across devices.

How It Works

HAPI adopts a local-first architecture, allowing AI coding sessions to run directly on user hardware. Core components include a server for managing sessions and AI backends, and clients (Web/PWA/Telegram) for remote interaction. This design prioritizes data privacy and offline capability while enabling seamless control from any internet-connected device, including remote approval of tool usage.

Quick Start & Requirements

Installation is supported via Homebrew (brew install tiann/tap/hapi), npm/npx (npx @twsxtd/hapi), prebuilt binaries (download from Releases), or Docker (ghcr.io/tiann/hapi-server:latest). Running the server requires hapi server or npx @twsxtd/hapi server. CLI usage is hapi or npx @twsxtd/hapi. Prerequisites include the Claude CLI for Claude Code sessions and Bun if building from source. For remote access, exposing the server via HTTPS (e.g., Cloudflare Tunnel, Tailscale) is necessary if it lacks a public IP. Session tokens are managed via ~/.hapi/settings.json or environment variables. Links to relevant documentation include docs/WHY_NOT_HAPPY.md, cli/README.md, server/README.md, and web/README.md.

Highlighted Details

• Supports multiple AI backends: Claude Code, Codex, and Gemini.
• Enables remote session control, file browsing, and git diff viewing.
• Features remote approval/denial of AI tool permissions.
• Includes session progress tracking via todo lists.

Licensing & Compatibility

The HAPI CLI is licensed under MIT, while the server and web components fall under LGPLv2. The LGPLv2 license may impose copyleft requirements on modifications to the library code when dynamically or statically linked into proprietary applications.

Limitations & Caveats

HAPI is positioned as a local-first alternative addressing differences from the "Happy" project. It requires the Claude CLI for specific AI backends and necessitates HTTPS exposure for remote accessibility if the server is not publicly reachable. The LGPLv2 license for core components warrants careful review for integration into closed-source commercial products.