Rakazo: Open-Source AI Agent Framework for Local Multi-Model Computer Use

Following the buzz generated by Anthropic's Claude Computer Use and Elon Musk's Grok Bot, an open-source project called Rakazo is emerging as a new focal point in the developer community. Unlike other open-source solutions that still rely on cloud sandbox services, Rakazo achieves truly local deployment via Docker — enabling developers to build an AI workforce made up of Claude, GPT, and locally hosted models working in concert.

A Genuinely Local, Open-Source AI Agent Solution

While open-source Computer Use projects like Open Boat already exist, most still depend on cloud sandbox services that require developers to register for third-party API keys. Rakazo stands apart through its thoroughly local-first design — it uses Docker images to spin up a complete virtual OS environment on your own machine, with AI Agents executing tasks inside an isolated Chrome OS container.

This architectural approach delivers several notable advantages:

• Zero cloud costs: No need to purchase cloud VM services
• Full data sovereignty: All data stays entirely on your local machine
• Flexible environment configuration: Maintains the security isolation of a sandbox environment

For tasks that need to run for extended periods (e.g., a 72-hour continuous web scraper) or require specialized environments (biopharmaceutical experiments, hardware communication testing), developers can flexibly configure independent execution environments.

Multi-Agent Collaboration: From Solo to Team

Rakazo's most innovative feature is its "Group" concept. Developers can create multiple Bots with distinct roles and organize them into teams to tackle complex tasks collaboratively. In testing, a two-Agent group called "DV Group" was tasked with drafting a social media growth strategy.

Initially, only the earlier-created Agent (named AppleDex) was working, while the other (DSH) remained on standby. When asked why, the system responded that responsibilities hadn't been clearly assigned. After the user prompted both Agents to negotiate and fill in their respective skill descriptions, the team shifted into a genuine collaborative mode — tasks were automatically broken down and executed in a pipeline fashion.

This design philosophy draws from how human teams operate: each member has clearly defined expertise and responsibilities, completing complex projects through negotiation and division of labor. For enterprise use cases, this means you can assemble specialized AI teams — an R&D group, a content creation group, a data analysis group, and so on.

Technical Architecture and Multi-Model Support

Developed by an international team, Rakazo natively supports mainstream global models like OpenAI GPT and Anthropic Claude, while also being compatible with Chinese models including DeepSeek, Zhipu, Qwen, and Xiaomi. The project is released under the MIT license, giving developers full freedom to modify and commercialize it.

Beyond core Agent conversation and task execution, Rakazo offers a rich set of extension features:

• Voice interaction: Integrates with voice services like ElevenLabs for real-time conversation
• Intentions connector: A Zapier-like workflow mechanism that connects to third-party services such as Gmail and Notion
• Sandbox compatibility: Supports cloud sandbox services like E2B, as well as domestic sandbox products from Alibaba Cloud and Tencent Cloud

Deployment Challenges and Current Limitations

Despite its advanced architecture, Rakazo is still in its early stages. The project doesn't yet offer an official desktop client, meaning regular users need to compile the source code themselves. Testing also revealed some rough edges — for instance, the Chrome OS container lacks Chinese fonts, causing display glitches (though not affecting functionality), and there are occasional network connectivity issues.

These early-stage limitations don't diminish its potential, however. As cloud sandbox service costs continue to fall (with traditional VM vendors pivoting to sandbox offerings) and open-source sandbox projects mature, Rakazo is well-positioned to become core infrastructure for enterprise AI Agent applications. Developers can use it as a foundation to build:

• 24/7 uninterrupted automated workflows
• Complex decision-making systems powered by multiple models in concert
• AI assistants deeply integrated with internal enterprise systems
• Specialized domain applications requiring custom execution environments

Rakazo vs. Commercial Computer Use Products

Compared to Anthropic's Claude Computer Use (which relies on cloud containers) and the forthcoming Grok Bot, Rakazo's open-source nature and local-first advantage make it more compelling in certain scenarios. Enterprises don't need to worry about data leakage, nor are they subject to the pricing and constraints of third-party services.

For developers, the MIT license means you can build commercial products on top of Rakazo or deeply customize it for specific needs. As the project's community grows and its feature set matures, this could become one of the most important open-source infrastructure projects in the AI Agent space.

The current excitement around Computer Use technology is no accident — it represents a pivotal shift for AI from "conversational assistant" to "autonomous executor." By taking an open-source, local-first approach, Rakazo opens up a compelling new path worth watching closely.