Tutti: A Unified Operating System for AI Agents
Tutti, a recent open-source project on GitHub, introduces an innovative concept aimed at solving the interoperability challenge among AI Agents. The project positions itself as an “operating system for agents,” with its core objective being to create a shared runtime space to break down the “information silos” that currently exist between different agent products like Codex and Claude Code. The name “Tutti,” derived from the musical term for “all together,” aptly summarizes its core mission of enabling all agents to work in concert.
Achieving Seamless Cross-Agent Context Sharing
Currently, the biggest pain point for users working with multiple AI Agents is workflow disruption. For example, code or documents generated in one agent must be manually copied and pasted into another for further processing, leading to context loss and reduced efficiency.
Tutti solves this by providing a unified virtual environment where users can run multiple different agents simultaneously. The key is that these agents can access a shared file system and session history. For instance, a user could generate a design mockup in Codex while having Claude Code write product documentation based on meeting minutes. Later, when developing a website in Codex, they can directly reference the documents and related conversations from Claude Code using the @ symbol, seamlessly connecting the two separate workflows. This design ensures that all information and context are naturally shared within the environment, eliminating the need for manual import/export.
Technical Implementation: Securely Reusing Local Agent Capabilities
Tutti’s implementation is both clever and security-conscious. It doesn’t integrate third-party agents through special APIs or reverse engineering. Instead, it works by calling and managing the runtimes of agents already installed locally.
For example, to integrate Codex, Tutti locates the command-line executable within the locally installed Codex desktop application (e.g., at /Applications/Codex.app/Contents/Resources/codex). It then launches the agent via the command line, creating a temporary configuration directory for it. To enable login-free access, Tutti links the user’s existing authentication files (e.g., login credentials in ~/.codex) to this temporary directory. This allows the agent instance launched by Tutti to inherit the user’s existing login status on the local machine, with all data processed locally and no user account credentials being uploaded.
Tutti currently supports Codex and Claude Code, with plans to integrate more major agents like OpenClaw, Gemini, and Hermes in the future.
Ecosystem and Future: Built-in Apps and Cloud Collaboration
Tutti’s vision extends beyond just connecting agents. It also offers a suite of built-in, scenario-based applications on top of this unified environment, such as an AI slide generator, a product prototype designer, and an AI document processor. These applications themselves aren’t highly intelligent, but they can invoke the capabilities of underlying integrated agents like Codex or Claude Code, providing a more user-friendly Graphical User Interface (GUI) for them. Users can perform specific tasks, like generating a slide deck or a product prototype, through simple interface interactions without needing to write complex prompts.
Furthermore, the project plans a cloud collaboration version called Tutti·VM. This version aims to sync the local working environment—including all agent sessions, files, and task statuses—to the cloud in real-time. This would not only allow users to switch seamlessly between devices but also enable them to invite colleagues into the exact same collaborative environment by sharing a link, facilitating real-time teamwork with no loss of context. This feature will expand the scope of agent collaboration from individual productivity to the team level.
