← Articles
Field Notes

The Shell Layer: An Invitation to Build Portable AI Context

Mike Purvis · AI Solutions Architect & Founder, Context Arcana · · 4 min read
A glass terrarium of hermit crabs wearing different shells on sand, with driftwood, moss, and snails — a living metaphor for portable context moving between shells.

I'm inviting developers, consultants, AI builders, workflow architects, and agentic systems people to start experimenting with CAS — Context Architecture System.

CAS is my open standard for portable, user-owned AI context.

The core idea is simple:

The model is the body.
The context is the shell.

As AI tools change — ChatGPT, Claude, Cursor, Gemini, local models, agents, MCP tools, RAG systems, and whatever comes next — people should not have to rebuild their context from scratch every time.

Their identity, rules, workflows, project history, decisions, knowledge, prompts, memory, and operating style should be portable.

One Shell. Any Body. Forever.

Getting started

The easiest way to start is with the CAS Shell Template:

github.com/context-arcana-llc/cas-shell-template

Install it on the machine where you do your work. Open the SETUP_PROMPT.md. Paste it into the LLM or coding assistant you are using — Claude, ChatGPT, Cursor, or another tool — and ask it to create a shell for a real project.

At the end of a session, type something like:

"end of session"

Then have the LLM summarize the work, decisions, open issues, and next steps into a session file.

In the next session, even with a different model, load the shell again and continue.

That is the basic workflow.

Over time, the shell becomes a portable project brain.

It can document your workflow, preserve decisions, expose automation opportunities, and create a structure that future tools can index, search, transform, or extend.

Open by design

The main open-source CAS project is here:

github.com/context-arcana-llc/context-architecture-system

I published CAS under Apache 2.0 intentionally.

That means developers, consultants, agencies, educators, and AI builders can extend it commercially.

You can build tools on top of it. You can build RAG layers for it. You can create client workflows with it. You can make your own shells. You can build integrations. You can create training, courses, or implementation services around it.

That is not a loophole.

That is the point.

CAS is meant to be open infrastructure.

The commercial layer I am building will live through Context Collection:

contextcollection.com

I see this becoming a shell marketplace: a new kind of digital product format for the AI-native era.

Shells for business. Shells for productivity. Shells for learning. Shells for creative work. Shells for agents. Shells for workflows. Shells for interactive books and LLM-based entertainment.

This is very early. The template needs testing. The standard needs contributors. The tooling needs to become more solid. But the direction is clear.

The 10 principles guiding CAS

  1. User ownership. The user owns their context, not the platform.
  2. Portability. Context should move across models, tools, agents, and vendors.
  3. Model freedom. The shell should survive as models change.
  4. Vendor neutrality. CAS should not depend on one AI company, API, product, or ecosystem.
  5. Interoperability. Context should be structured so different tools can read, write, import, export, and transform it.
  6. Human agency. AI should support human judgment, not replace ownership, consent, or responsibility.
  7. Transparency. The structure should be readable, inspectable, and understandable.
  8. Modularity. Systems should be component-based so parts can be swapped, upgraded, or replaced.
  9. Practical usefulness. CAS should solve real workflow problems, not exist as abstract architecture.
  10. Ethical adoption. CAS should move the market away from manipulative capture funnels, hype-driven AI selling, and platform lock-in, and toward honest, useful, ownership-based systems.

That last principle matters a lot to me.

I do not want CAS to become another way to trap people inside someone else's platform.

I want it to become a way for people and businesses to carry their intelligence, memory, knowledge, and workflow across systems they choose.

An invitation

If you are working with AI agents, RAG, MCP integrations, automation, consulting, AI-assisted coding, education, digital products, or workflow design, I invite you to try the template in a real workflow.

Use it. Break it. Improve it. Fork it. Build on it. Contribute to it.

This is the beginning of the shell layer.

One Shell. Any Body. Forever.

← Back to all articles