Mindstone Rebel: Local-First AI Agents with Shared Memory Reshape Enterprise Workflows

Mindstone Rebel: The Local-First Agentic OS That Remembers Everything

Enterprise AI has a memory problem—and not the technical kind. Most organizations have deployed chatbots, copilots, and model subscriptions, but these tools operate in silos. Each interaction starts from scratch. Knowledge gained by one team rarely transfers to another. The result: fragmented productivity gains and a growing stack of underutilized AI licenses.

Mindstone, a London-based AI transformation startup, is betting that the next wave of enterprise AI will be defined not by access to models, but by coordination. Its new product, Rebel, is a local-first, agentic AI operating system that stores organizational memory in plain markdown files, routes tasks dynamically across local and cloud models, and operates under a Fair Source license that lets teams inspect, modify, and own their agent infrastructure.

This is not another agent framework. Rebel is a bet that the future of enterprise AI is transparent, portable, and built on shared memory—and early results from a 250-person deployment at Epignosis suggest the bet may pay off.

Why Shared Memory Is the Killer Feature

Mindstone CTO Greg Detre calls shared memory “the most empowering thing you could possibly do with a knowledge-worker AI.” The logic is simple: when every agent interaction—every prompt, every decision, every workflow—is recorded and accessible, the organization becomes a learning system. “You get this feeling of being a super-organism as a company that just gets smarter and smarter,” Detre said.

Rebel’s memory system uses a tiered structure. High-value information is written directly into a project’s readme.md file. Medium-value data becomes a reference link. Low-priority material is stored in an indexed directory, dormant until a relevant task calls it back. This approach avoids the common pitfall of dumping everything into a vector database and hoping retrieval works.

For enterprise buyers, the implication is clear: Rebel turns scattered employee experiments into an institutional asset. Every workflow improvement, every prompt refinement, every model preference is captured and reusable. The company gets smarter over time without requiring manual documentation or centralized oversight.

Local-First Architecture: A Hedge Against Vendor Lock-In

Rebel’s architecture is fundamentally different from cloud-dependent agent frameworks like LangGraph or CrewAI. Instead of wiring together databases, cloud infrastructure, and state-management logic, Rebel stores its entire state—prompts, task instructions, memory hierarchy—in local markdown files. A primary configuration file, agents.md, acts as the agent’s core instruction layer and runtime boundary.

This design choice has three strategic consequences. First, it reduces token costs. Markdown files carry minimal formatting overhead compared to Word documents or PDFs, meaning more of the model’s context window is spent on actual tasks. Second, it makes the system auditable. Any developer—or compliance officer—can open a markdown file and see exactly what the agent is instructed to do. Third, it eliminates platform dependency. If a company decides to switch providers or bring workflows in-house, all agent instructions and memory are portable.

Under the Fair Source license, Rebel’s code is viewable, modifiable, and deployable. Teams of up to 100 concurrent users can run it for free. Beyond that, a commercial license is required. A two-year sunset clause automatically converts each version to MIT open-source after 24 months. This structure reduces the risk of being trapped in a proprietary ecosystem—a growing concern for enterprises deploying AI agents with broad access to internal systems.

Automatic Model Routing: Cost Optimization Without Compromise

One of Rebel’s most practical features is its ability to break a task into subtasks and route each step to the optimal model—local or cloud, cheap or powerful. A planning step might go to GPT-4 or Claude; routine data extraction could run on Llama or DeepSeek; sensitive approval checks can stay on a local model, never leaving the device.

This multi-model orchestration addresses a real pain point for enterprise buyers: the tension between cost, performance, and security. Sending every query to the most expensive cloud model is wasteful. Sending sensitive data to an external API is risky. Rebel’s approach lets companies define policies that automatically balance these factors, without requiring manual model selection.

For security teams, the ability to keep approval logic local is critical. Rebel can run gating decisions entirely on a local model, ensuring that sensitive actions—like drafting emails or modifying files—are authorized without any cloud call. This distinction matters for regulated industries where data sovereignty is non-negotiable.

The Epignosis Case: Eight Full-Time Roles Recaptured in 12 Weeks

Mindstone’s most compelling proof point is the deployment of Rebel across Epignosis, a 250-person company spanning sales, engineering, product, finance, and customer success. Over 12 weeks, Mindstone says Epignosis recaptured the equivalent capacity of eight full-time roles. Adoption spread organically after employees saw colleagues automate time-consuming work—a pattern the company calls the “potatoes effect.”

Epignosis CEO Dimitris Tsingos framed the impact in broader terms: “The border between learning and doing is fading out—and that changes everything about how you scale.” For enterprise buyers, this case suggests that Rebel’s shared-memory design can deliver measurable ROI in a matter of weeks, not quarters.

Who Gains, Who Loses

The winners in this shift are clear. Small and mid-sized teams get a powerful, customizable agentic OS for free. Enterprise clients gain a transparent, portable alternative to opaque cloud agents. Mindstone itself stands to capture a growing market for agent orchestration, backed by $5 million from Pearson Ventures, Moonfire Ventures, and Zanichelli Venture.

The losers include traditional RPA vendors, whose rigid automation scripts look increasingly obsolete next to adaptive, memory-driven agents. Closed-source AI agent platforms may also feel pressure as enterprises demand inspectability and portability. Rebel’s Fair Source license and local-first design set a new baseline for what enterprise buyers should expect.

Outlook: What to Watch in the Next 30 Days

Rebel’s success will depend on execution. Local-first software can be harder to manage than cloud SaaS. Shared memory raises governance questions—who decides what gets remembered? Multi-model routing adds complexity. And enterprises will still need proof that agentic workflows can deliver reliable productivity gains without creating security or compliance headaches.

Key indicators to watch: Linux support (in development), adoption metrics beyond Epignosis, and whether the Fair Source community contributes improvements. If Rebel can demonstrate repeatable ROI across diverse enterprise environments, it could become the operating layer for AI-driven work—not just another agent framework.

Source: VentureBeat