AI agents are becoming a cornerstone of enterprise workflows, but most platforms force teams into a one-size-fits-all approach. London-based startup Mindstone is changing that with Rebel, a new AI operating system designed to let agents dynamically select the best model for each task—whether it’s a local, cost-effective option or a powerful cloud-based system.
A local-first agent OS built for flexibility and control
Rebel launched this week as a local-first AI agent orchestrator, operating under a Fair Source license. This means small teams (under 100 users) can adopt it for free, while larger organizations can opt for an enterprise license. Unlike many agent frameworks that rely on complex cloud infrastructure, Rebel stores its core memory, instructions, and configurations as markdown files—.md—on the user’s device.
This local-first design ensures data stays where it belongs: under the company’s control. By avoiding reliance on centralized databases or vendor-specific interfaces, Rebel reduces the risk of vendor lock-in and gives teams full visibility into how their AI agents operate. The system also cuts costs by minimizing unnecessary API calls, as markdown files require fewer tokens than formatted documents like Word or PDFs.
The architecture centers around agents.md, a primary configuration file that acts as the agent’s instruction manual. Users can inspect, modify, or move these files without navigating opaque cloud dashboards, making customization and troubleshooting straightforward.
Dynamic model routing for cost and security optimization
One of Rebel’s standout features is its ability to automatically route different parts of a task to the most suitable AI model. Instead of funneling every step through a single, expensive cloud-based model, Rebel can break work into subtasks and assign each to the optimal tool—local or cloud.
For example:
- A high-cost model could handle complex reasoning or strategic planning.
- A lightweight, open-source model like Llama or DeepSeek might process routine data entry.
- A local model could manage sensitive tasks, such as reviewing confidential documents or approving internal workflows.
This approach aligns with enterprise priorities around cost efficiency and data security. Companies can enforce policies that prevent sensitive data from leaving local infrastructure while still leveraging cloud models for less critical work. As Mindstone CTO Greg Detre explained, "I want to be able to say, ‘Help me with this,’ and it knows what’s personal, what’s sensitive, and what can be shared with the whole company."
The system also supports multi-step workflows called "skills," reusable procedures agents can execute on demand. Operators let teams define behavioral rules—such as reviewing documents from a security or investor perspective—while automations handle scheduled tasks like scanning messages or drafting responses before employees even open the app.
Tiered memory for smarter, context-aware agents
A common pitfall in enterprise AI is dumping vast amounts of company data into a central database and hoping the system retrieves the right context later. Rebel avoids this by using a tiered memory structure that prioritizes relevance.
When an interaction occurs, the system evaluates how likely the information is to be useful again. High-priority data is stored in a way that ensures quick retrieval, while lower-priority details are archived or discarded. This not only improves efficiency but also reduces clutter in the agent’s context window.
Available now for macOS (both Intel and Apple Silicon), Windows, and Linux (in development), Rebel is already attracting attention from organizations seeking a more transparent and adaptable AI orchestration platform. With $5 million in funding from investors like Pearson Ventures and Moonfire Ventures, Mindstone is positioning Rebel as a practical alternative to complex, cloud-dependent agent systems.
As enterprises continue to integrate AI into their operations, tools like Rebel highlight a growing trend: autonomy doesn’t have to come at the cost of control. By prioritizing local-first design, dynamic model routing, and user-friendly customization, Mindstone is paving the way for AI agents that work smarter, not harder—without compromising security or predictability.
AI summary
Mindstone’un Rebel aracı, işletmelerin AI ajanlarını yerel markown dosyalarıyla yönetmesini ve tercih edilen modelleri otomatik seçmesini sağlıyor. Veri gizliliği, maliyet kontrolü ve esneklik sunan bu yenilikçi sistem hakkında detaylar.
