SHEEP MANIFEST

0. Who I Am

I'm Filippo, 25 years old. My life has been shaped by physics and programming. Today, observing the frenetic evolution of Artificial Intelligence, I see an ecosystem that is incredible but fragile.

What drives me is a concern: we are heading toward a future where AI power is concentrated in the hands of a few. The intelligence is there, but it's locked behind interfaces we don't control, policies we didn't choose, and infrastructure we can't inspect. This scares me — and it's why I started building.

Sheep is my attempt to change the direction, one primitive at a time.

1. The Problem: Access is not Control

We are at a historic fork in the road. This isn't just the classic "Open Source vs. Closed Source" debate; it's a question of engineering viability.

The LLMs themselves — whether proprietary or open-weight — are not the core issue. Excellent models exist and will keep improving. The real problem is that the applications built on top of them are all closed and proprietary. If you want an AI assistant today, you must accept someone else's interface, their policies, their data handling. The intelligence is accessible, but the control layer isn't.

The centralization of social networks taught us what happens when we don't own the infrastructure. Entrusting sensitive data — medical records, financial information, proprietary code — to external black boxes is not just an ethical risk; it's an unacceptable operational risk.

2. The Thesis: The Model is the Engine, You Build the Car

Here's the core idea: the competitive advantage of tomorrow won't only lie in accessing the smartest LLM (that's becoming a commodity). It will increasingly depend on your ability to integrate intelligence into your own systems through architecture you control.

A genius model is useless if it can't read your internal database, can't respect your compliance requirements, or leaks your IP to an external cloud.

What's missing is Cognitive Middleware: a layer that decouples Intelligence (the Model) from Execution (the Action). Not another chatbot. Infrastructure.

3. What Sheep Actually Is

Let's be concrete.

Sheep is a collection of open-source libraries designed for agentic development. They're built to be wrapped by MCP servers and integrated into existing company infrastructure.

We don't build LLMs. Training foundation models requires resources and expertise we don't have — and others do it well.

We don't sell pre-packaged "AI employees." We're not in the business of magic agents that promise to replace your team.

We build primitives. Tools that give LLMs the ability to see your data, act on your systems, and remember context — all running on infrastructure you control.

These libraries are designed to work with frameworks you might already use (LangChain, LangGraph, CrewAI, etc.), not to replace them. Think of Sheep as the specialized tooling layer beneath your agent orchestration.

Example use cases:

• A physicist who wants to run complex data analysis pipelines with AI assistance, without uploading datasets to external services.
• A healthcare company that needs AI-powered document processing while guaranteeing GDPR compliance and data residency.
• An enterprise that wants to build an internal Copilot for their codebase, avoiding per-seat costs and keeping proprietary code in-house.

4. Why Open Source

An open-source model without open tools to orchestrate it is like an engine without a steering wheel.

Sheep is open source because:

• Transparency matters. If you're trusting software with sensitive data, you should be able to audit every line.
• Modularity wins. Take what you need, ignore what you don't, extend what's missing.
• Independence is the point. Your system shouldn't break because a vendor changed their pricing or ToS.

We're in the early stages. The libraries exist, the vision is clear, but we're still building traction and community. If you believe in this direction, there are ways to get involved (see below).

5. What's Next

We're exploring sustainability paths: R&D partnerships with companies who need custom tooling, managed hosting for MCP servers, and consulting for teams building agentic systems. The model is evolving — what won't change is the commitment to keeping the core open.

The name "Sheep" isn't about following the herd. It's about swarm intelligence: small, modular, numerous units that — when well-orchestrated — solve problems better than a single monolith.

We want to enable a new generation of System Architects. People who don't just prompt, but design. We build the foundations. You bring the vision, the data, the domain expertise.

Get Involved