Background: The Adoption Gap for AI Coding Agents
Since 2025, coding agents represented by Claude Code and Codex CLI have demonstrated powerful capabilities far beyond simple code completion. However, in practice, about 95% of knowledge workers are unable to effectively utilize them due to three core obstacles. Firstly, these agents commonly rely on command-line interaction, which presents a barrier for users unfamiliar with concepts like the terminal and file paths. Secondly, configuring complex development environments such as Python and Node.js on a local machine is a major challenge for non-technical personnel. Lastly, uploading sensitive business data, such as financial or legal information, to third-party cloud services raises significant data privacy and security concerns.
Multica: The Middleware Connecting Humans and AI
To solve these problems, the open-source project Multica was created. It provides a native desktop client that brings the powerful capabilities of coding agents to every knowledge worker through a unified visual interface. The project’s name, Multica, is inspired by the pioneering 1960s operating system Multics, reflecting its goal of solving the current challenges of multi-modal and multi-agent collaboration.
Multica’s core vision is to become an extensible middleware layer, similar to Obsidian, where users can customize the application for their specific workflows via a plugin system in the future. For example, by loading a development plugin, Multica can become a professional coding assistant; with a design plugin, it can aid in creative design; and with research-oriented plugins, it can transform into a knowledge management and analysis tool akin to NotebookLM.
Core Features and Technical Architecture
As a local-first macOS application, Multica ensures that user data always remains on the local device, sending only task commands to cloud-based AI agents. Its main features include:
- Native Desktop Experience: Offers a clean and intuitive graphical user interface.
- Multi-Agent Support: Through the Agent Client Protocol (ACP), it currently supports various agents like Claude Code, OpenCode, and Codex CLI (ACP).
- Local-First Data: All session history and user files are stored locally (
~/.multica/sessions/), resolving data privacy issues.
- Sandboxed Environment: Comes with a pre-configured sandboxed environment with all dependencies handled, enabling an “out-of-the-box” experience and eliminating the hassle of manual setup for users.
Its technical architecture is based on Electron. The Conductor module within the main process is responsible for communicating with and managing various AI agents running as child processes via the ACP protocol. The renderer process, built with React, constructs the user interaction interface, ensuring the application’s stability and high performance.
Quick Start and Usage
For advanced users and developers, Multica also provides a comprehensive command-line interface (CLI) that supports both interactive mode (REPL) and single-shot commands. Users can check the installation status of agents with pnpm cli doctor or interact directly with an agent using pnpm cli prompt "Your request". The project uses pnpm as its package manager. Developers can quickly start the application and contribute to its development with pnpm install and pnpm dev. The project is open-source on GitHub and provides build commands for macOS, Windows, and Linux platforms.
