In the rapidly expanding landscape of AI agent orchestration platforms, Mindstone’s Rebel, a new offering from the London-based AI transformation startup, stands out as a particularly promising contender. Launched officially this week, Rebel operates as a local-first, agentic AI operating system distributed under a “Fair Source” license. This model grants teams of up to 100 users the freedom to adopt and customize the system without charge, while larger organizations will require an enterprise license.
The platform’s core strengths lie in its simplicity and extensive customizability, designed to accommodate unique or highly specific team workflows. Central to its architecture is the use of markdown (.md) files, a common and accessible open-source standard, for managing agent memory and instructions. This approach fosters an organizational memory layer, ensuring that agents consistently utilize the enterprise’s preferred AI models for various tasks and subtasks. Rebel dynamically switches between local and cloud-based models, providing predictable and visible operations that optimize costs while maintaining data privacy and security.
“Shared memory is the most empowering thing you could possibly do with a knowledge-worker AI,” stated Greg Detre, Chief Technology Officer at Mindstone, during a recent interview. “You get this feeling of being a super-organism as a company that just gets smarter and smarter.”
Rebel is currently available for macOS on both Intel and Apple Silicon architectures, as well as for Windows. Support for Linux is under active development.
Mindstone has secured $5 million in funding from investors including Pearson Ventures, Moonfire Ventures, and Zanichelli Venture.
A Distinctive, Local-First Architecture Built on Markdown Files
Rebel’s local-first architecture is a key differentiator. Unlike developer-centric agent frameworks such as LangGraph, CrewAI, and AutoGPT, which necessitate teams to manually integrate databases, cloud infrastructure, and state-management logic, Rebel centralizes its core agent memory and instructions within local markdown text files. This method leverages arguably the simplest and most widely adopted format for guiding AI agents, embraced by AI developers and power users globally.
Mindstone reports that Rebel stores its operational state, prompts, task instructions, and memory hierarchy in these markdown files, enabling users and organizations to easily inspect, migrate, or modify them. A primary configuration file, `agents.md`, serves as the agent’s foundational instruction set and runtime boundary.
This architectural choice addresses cost considerations directly. Mindstone posits that conventional office formats like Word documents and PDFs often introduce formatting and metadata overhead that consumes valuable model token context, thereby increasing API expenses. Markdown, by contrast, keeps information closer to raw text, allowing the AI model’s context window to be utilized more effectively for the actual task rather than document structure.
Furthermore, the company positions this approach as a safeguard against vendor lock-in. By storing agent instructions, automations, and memory locally as plain text files, organizations are not confined to a single SaaS provider’s interface or database. This is increasingly critical as enterprises grant AI systems broader access to sensitive corporate resources such as email, calendars, documents, and internal workflows.
Rebel also facilitates the creation of repeatable AI workflows. “Skills” are saved multi-step procedures that agents can reuse. “Operators” refine agent behavior for specific tasks, such as critically reviewing a pitch deck from an investor’s viewpoint or assessing work through a security lens. “Automations” enable scheduled background tasks, including scanning communications or files, identifying relevant updates, drafting responses, or preparing information for employee access.
Dynamically Selecting the Optimal AI Model for Every Task
A standout feature of Rebel is its multi-model orchestration capability. The system can deconstruct a task into constituent parts and route different steps to distinct AI models, including a strategic split between local and cloud-based resources. This routing is governed by factors such as the sensitivity of the data or enterprise policy mandates.
A more powerful model can be assigned complex reasoning or planning tasks, while a less expensive model handles routine operations. Sensitive operations or approval checks can be delegated to local models. This flexibility is paramount for enterprises seeking granular cost control and operational adaptability; not every task necessitates leveraging the most expensive cloud model, and specific corporate policies may prohibit sensitive data from leaving local infrastructure.
“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,” Detre elaborated.
This model-agnostic framework empowers companies with enhanced control over both expenditure and security. Data-intensive processes can leverage more cost-effective models like Llama or DeepSeek, while high-level analytical tasks can be reserved for premium models. Critical operations involving sensitive information can be processed via a local model running directly on the user’s machine, ensuring data containment.
This strategy also enables enterprise teams to seamlessly integrate both cloud and local inference capabilities without resorting to an all-or-nothing approach.
By moving away from centralized, monolithic cloud interfaces towards a local, file-driven architecture, Mindstone is pioneering a new paradigm for how enterprise technical decision-makers orchestrate autonomous workflows while preserving data sovereignty and operational predictability.
Operationalizing AI: A Tiered Memory System
Mindstone CTO Greg Detre designed Rebel’s memory system to circumvent a common challenge in enterprise AI deployments: the practice of ingesting vast amounts of company data into a database with the implicit hope that search functionality will later retrieve the correct context. Rebel employs a hierarchical memory structure.
Upon an interaction, the system assesses the probability of that information being relevant for future use. Data anticipated to be highly valuable is recorded in a local `readme.md` file associated with a specific project. Information deemed moderately valuable is converted into reference links pointing to more comprehensive historical records. Lower-priority data is archived in an indexed memory directory, remaining accessible but dormant until a relevant task necessitates its retrieval.
An ROI Dashboard for Enterprise Procurement
For larger organizations, Mindstone Pro introduces an “Impact Dashboard.” This feature is engineered to illustrate the time and cost savings generated by Rebel across various business units. Mindstone states that the dashboard utilizes a separate, closed-source Large Language Model (LLM) to analyze telemetry data and quantify business impact. The system is reportedly calibrated conservatively, employing the lower end of estimated performance gains to ensure realistic productivity metrics.
This dashboard addresses a critical concern for enterprise AI buyers: the need to demonstrate tangible value without resorting to invasive employee monitoring. Mindstone assures that the dashboard operates independently of individual workspaces, enabling IT and business leaders to assess adoption and return on investment without accessing employees’ private agent activity logs.
Fair Source Licensing: Mitigating Platform Risk
Mindstone is rolling out Rebel under a “Fair Source” license, a licensing model positioned between fully proprietary Software-as-a-Service (SaaS) and permissive open-source software. This license permits viewing, auditing, modification, and deployment of Rebel’s codebase. Individuals and organizations with up to 100 concurrent users can utilize the platform at no cost. Exceeding this user threshold necessitates the acquisition of a commercial Mindstone Pro license.
The license also incorporates a two-year sunset clause. Twenty-four months following the release of a specific version, that version automatically transitions to the MIT open-source license.
For enterprise clients, the primary advantage of this licensing model is the reduction of vendor lock-in risk. With all automations, memory files, and agent instructions stored locally in markdown format, companies retain the flexibility to migrate their data and workflows to alternative solutions if required. While the product is commercial, its foundational elements are designed for inspectability and portability.
Security Discussions Center on Local Approvals and Shared Memory Protocols
Rebel’s recent debut on the product-sharing platform Product Hunt sparked technical inquiries regarding permission handling, safety protocols, and shared memory management in a local-first agent environment. Developer Nikita Pokryschko questioned whether approval checks for sensitive actions could be executed entirely by a local model, or if the gating logic still mandated a cloud interaction.
Detre responded by detailing Rebel’s architectural separation of planning, execution, and background safety logic. He clarified that companies can configure Rebel to rely exclusively on a local model for critical gating decisions, a point emphasized by Mindstone’s CEO Joshua Wöhle and CTO Greg Detre.
This distinction is crucial for corporate security and compliance teams. Autonomous agents frequently require extensive permissions to access files, draft communications, or interact with internal systems. If the ultimate approval checkpoint relies on an external cloud model, some organizations perceive this as a significant compliance vulnerability. Mindstone asserts that Rebel can maintain these approval boundaries within the local environment.
A secondary discussion revolved around Rebel’s methodology for determining data shareability. Product developer Clement Morel inquired whether shareability is contingent on content, user-defined settings, or learned behaviors, and how the system addresses potential inaccuracies. Detre explained that Rebel utilizes the user’s local “Chief-of-staff README” and defined workspaces to delineate private, team-specific, and company-wide information. In instances of ambiguous context, the system prompts the user for explicit approval before proceeding.
This emphasis on transparency underpins Mindstone’s broader critique of opaque agent systems. As CEO Joshua Wöhle articulated in a LinkedIn post: “If an agent is going to sit inside your workspace, remember your context, and ask permission before changing the world, you should be able to see how it works. Not because everyone will read the code, but because someone can.”
Mindstone Cites Customer Rollout as Early Validation
Mindstone reports that Rebel has already been implemented across the 250-person workforce of its customer, Epignosis, serving sales, engineering, product, finance, and customer success departments. “The entire organization is operating on Rebel today,” Wöhle confirmed.
During a 12-week deployment period, Mindstone claims Epignosis achieved a productivity equivalent to reclaiming the capacity of eight full-time roles. The company attributes the organic adoption rate to employees witnessing colleagues automate time-consuming tasks, a phenomenon reportedly dubbed the “potatoes effect.”
The Epignosis case is central to Mindstone’s thesis that enterprise AI should transcend isolated, individual tools. Rebel’s shared-memory architecture is designed to facilitate workflow continuity across teams and promote cumulative improvements as more employees engage with the system.
“The border between learning and doing is fading out – and that changes everything about how you scale,” stated Epignosis CEO Dimitris Tsingos in a statement provided by Mindstone.
Background on Mindstone
Mindstone Learning Limited, headquartered in London and founded in 2020, is led by CEO Joshua Wöhle, a co-founder of the digital child safety firm SuperAwesome. Initially focused on consumer education technology, the company developed a digital curation platform described as a “Spotify for learning,” incorporating principles of compound learning. However, with the widespread commercialization of generative AI technologies between 2022 and 2024, Mindstone pivoted to B2B enterprise enablement. The leadership identified a critical “last-mile” challenge: while AI tools promised significant productivity gains, traditional corporate training methods were insufficient for effectively integrating them into daily operations.
Currently, Mindstone operates as a holistic enterprise software and training ecosystem aimed at maximizing corporate ROI on existing AI licenses. Its product architecture systematically addresses various organizational levels through highly contextualized, practical software applications rather than abstract theoretical presentations. Mindstone employs a hybrid capitalization strategy, blending institutional venture capital from firms like Moonfire Ventures and Pearson Ventures with community-driven equity crowdfunding on platforms such as Seedrs and Crowdcube.
The company has successfully established a presence in the enterprise market, securing commercial agreements with major corporations including The Home Depot, Hyatt Hotels Corporation, Pearson, and Ernst & Young. Mindstone positions itself as a vital catalyst for overcoming corporate inertia, empowering organizations to build the internal competencies necessary for successful AI transformations.
Mindstone’s Strategic Positioning: Enterprise AI Demands Shared Memory, Not Just More User Seats
Rebel enters the market at a pivotal moment when businesses are transitioning from AI experimentation to full-scale AI operations. The initial phase of enterprise AI adoption largely focused on access – providing employees with chatbots, copilots, and model subscriptions. Mindstone’s strategy hinges on the prediction that the next wave will prioritize coordination and orchestration.
This involves developing capabilities such as shared memory, reusable workflows, local control mechanisms, flexible AI model routing, and measurable business impact reporting. Crucially, it also entails providing enterprises with transparency into the systems they are expected to trust. The company’s immediate challenge lies in execution. Local-first software can present greater management complexities than cloud SaaS solutions. Shared memory raises significant governance questions, and multi-model routing introduces inherent complexity. Furthermore, enterprises will require robust evidence that agentic workflows can deliver consistent productivity gains without compromising security or compliance.
Mindstone’s core proposition is clear: merely acquiring AI user licenses does not equate to establishing robust AI infrastructure. Rebel represents their ambitious endeavor to transform scattered individual employee experiments into a cohesive operational layer for the modern enterprise.
Business Style Takeaway: Mindstone’s Rebel platform addresses a critical gap in enterprise AI adoption by focusing on collaborative “shared memory” and flexible model orchestration, moving beyond simple user seat expansion. This approach offers businesses a path to enhance AI ROI through integrated workflows and data control, potentially reducing costs and mitigating vendor lock-in risks inherent in more centralized solutions.
Original article : venturebeat.com