Mindstone Rebel: The Local-First Agentic AI Operating System

What happened?

London-based startup Mindstone has unveiled Rebel, an agentic AI operating system designed to run locally and based on Markdown files. The product, officially launched this week, is available for macOS (Intel and Apple Silicon) and Windows, with Linux support in development. Mindstone has raised $5 million from investors including Pearson Ventures, Moonfire Ventures, and Zanichelli Venture. According to VentureBeat, Rebel stands out in a crowded landscape of agent orchestration platforms, being described as "one of the most promising" for its local-first approach and Fair Source license.

Local-first architecture and shared memory

Rebel distinguishes itself from frameworks like LangGraph, CrewAI, or AutoGPT with its local-first architecture. Instead of relying on cloud databases or complex infrastructure, agent memory and instructions are stored in Markdown text files (.md). This includes a main configuration file, agents.md, which acts as a central instruction layer and execution boundary. According to Greg Detre, CTO of Mindstone, in a statement to VentureBeat: "Shared memory is the most empowering thing you can do with AI for knowledge workers. You get the feeling of being a superorganism as a company that becomes increasingly intelligent." Unlike platforms like LangGraph, which require integrating databases and state logic in the cloud, Rebel simplifies the process by using plain text files, reducing technical complexity and making data inspection and modification easier. This approach also enables an organizational memory layer that ensures agents use the company's preferred AI models for each task, dynamically switching between local and cloud models in a predictable and visible way to optimize costs and privacy.

Key advantages: costs, privacy, and customization

Using Markdown reduces token consumption compared to formats like Word or PDF, as it eliminates unnecessary metadata and formatting, resulting in lower API costs. Additionally, storing everything locally avoids vendor lock-in and facilitates data inspection and modification. Rebel allows creating "Skills" (reusable procedures), "Operators" (behavior adjustments for specific tasks), and "Automations" (scheduled background tasks). This customization is key for teams with unique workflows. Compared to platforms like AutoGPT, which often require advanced technical knowledge and rely on the cloud, Rebel offers a more accessible alternative. According to VentureBeat, simplicity and adaptability are its main features, allowing any team, regardless of how specific their processes are, to implement AI agents locally.

Fair Source license: accessible for small teams

Rebel is distributed under a Fair Source license, which allows free use for teams of fewer than 100 users. Larger organizations must purchase an enterprise license. This strategy aims to foster adoption in startups and SMEs, while offering a monetization path for corporate clients. The Fair Source license is an intermediate model between open source and proprietary software, allowing free use for small teams but requiring payment for enterprise use. This contrasts with frameworks like LangGraph, which are fully open source, or commercial solutions like OpenAI's, which charge per use. Mindstone's decision could facilitate adoption in the startup ecosystem, where resources are limited, while generating revenue from large companies.

Why is this important?

Agentic AI orchestration is a rapidly expanding field, but most current solutions require cloud infrastructure and advanced technical knowledge. Rebel bets on simplicity and local control, which could democratize access to customized AI agents. Its focus on shared memory also addresses a critical problem: the lack of continuity and context in current agents. At a time when companies like Microsoft and Google are pushing cloud-integrated AI assistants, Rebel offers an alternative for those prioritizing data privacy and infrastructure control. Moreover, using Markdown as a universal format could standardize interaction with agents, similar to how JSON has become the standard for structured data. According to VentureBeat, the Markdown format is being widely adopted by developers and advanced AI users, suggesting Rebel could capitalize on this trend.

Consequences and outlook

If Rebel achieves significant adoption, it could pressure cloud platform providers to offer lighter, local options. However, it still needs to demonstrate scalability and performance in complex enterprise environments. Dependence on Markdown files may be a limitation for teams requiring deep integrations with databases or external APIs. Additionally, competition is fierce: LangGraph, CrewAI, and AutoGPT already have established communities, and giants like Microsoft are investing in local agents with Copilot. Nevertheless, Rebel's value proposition—simplicity, privacy, and shared memory—could attract niches such as startups, research teams, and companies with strict regulatory compliance requirements. The $5 million raise from investors like Pearson Ventures suggests confidence in the model, but success will depend on real adoption and Mindstone's ability to iterate quickly.

What readers should know

• Rebel is available for free for teams of up to 100 users (Fair Source license).
• It runs on macOS and Windows; Linux is in development.
• Memory and instructions are stored in Markdown files, facilitating transparency and control.
• Mindstone has raised $5 million and is backed by renowned investors such as Pearson Ventures, Moonfire Ventures, and Zanichelli Venture.
• It is an alternative to frameworks like LangGraph or AutoGPT, but with a simpler, local approach, ideal for teams prioritizing privacy and customization.
• According to VentureBeat, Rebel allows dynamic switching between local and cloud models to optimize costs and security.