Welcome to the first issue of The Context Layer.
This newsletter explores the emerging tools and systems of context that help AI produce high-quality, high-fidelity, humanized output.
I'm Ryan Charleston, the creator and the first author here. I'm a semi-technical creative with 15+ years of experience in marketing, sales, and product development.
I've been working and building with AI since ChatGPT launched in late 2022. I was the first AI-native hire at an enterprise blockchain startup in 2024, and I've always been fascinated by business intelligence, analytics, and automation.
Over the past two years, I've become increasingly interested and involved in context engineering and building the systems that actually help people and businesses get productive output from AI.
Below is how I've mapped the AI context layer so far. I expect this to evolve as the space grows, along with my understanding of it.
A working map of the AI context layer (October 2026)
The map above is a working map, with examples, not rankings. Sources are on the left. Next come the stores: knowledge tools that hold what you wrote, memory tools that hold what happened, and the databases underneath both. Then the conventions and protocols that carry context to a model. On the right are the tools that do the work and write back into your stores.
In the middle sits the switchboard: who you are, and what each tool is allowed to read. That switchboard is what I call your vibegraph. It's what helps AI tools produce more human results: sounding like you when they write, and acting as you when they work on your behalf. I saw nobody building this part well, so I built it myself.
People in the field call this whole practice context engineering. In plain terms, it means organizing information in a way that allows AI to know about you or your business (as context) before you engage with it.
Putting an end to AI slop
If you're already an AI power user, you know the frustration of AI slop, hallucinations, and sycophancy, especially if you work in marketing or sales, or create content for a living.
You may also know how annoying it is to keep pasting the same context, prompts, and data into one-off chats to get better output.
It feels like you start from zero every time, and it's inefficient. If you're new to AI and not sure what I'm talking about, trust me, you'll run into this problem eventually.
The industry has made some progress addressing this problem over the past few years, with better agent harnesses (Codex, Claude Code, Cursor), better memory systems (internal and external), skills, plugins, MCP, and AGENTS.md. All helpful, and I use them every day.
What's missing, from my perspective, is something that codifies the unique identity, personality, purpose, taste, and judgment of the human operator. It matters more than people realize, and I've experienced this firsthand.
To solve this problem, I built vibegraph.md, an open-source context engineering framework that defines your human "vibes" as rich context for LLMs and AI agents. I'm also building vibegraph.ai, a web app that gives less technical users an easy way to build a vibegraph and connect it to AI tools like Claude and ChatGPT, or any AI agents they use.
This is the space I explore in this newsletter. Every Wednesday I break down one part of the context layer. I define terms, show you what I'm building and what others are building, say what's working and what's not, and give you specific steps to build a rich context layer for yourself or your business.
My goal is to reveal something useful every week, in under 7 minutes, that you can act on to get more from AI in your personal life or business.
With that, let's dive into the word of the week.
Term settled: what is a vibegraph?
Definition: a vibegraph is the network of identity, context, knowledge, and memory that governs how AI thinks, writes, and acts for one human, and any businesses they own or operate.
Vibegraph whitepaper (October 2026)
For the record, "vibes" here means subjective human traits: personality, taste, voice, judgment, values, purpose, and aesthetic. It covers everything an AI gets wrong by default when it doesn't have this context.
You may already have the seed of a vibegraph
If you use AI for real work, part of your vibegraph may already exist. It's just scattered across a notes app, a folder of docs, three platform memories that aren't easy to see or audit, a prompt you pasted into a project last month, a few skill files, and ideas trapped in your own head. Most people don't have these things codified, connected, or kept up to date.
Your seed is like a shoebox full of random photos and notes. Everything is there. Nothing is organized or connected. This is why your AI tools keep producing slop. They start from zero, or they dig through your pile and draw conclusions from messy or incomplete context.
What turns your messy shoebox into an album
It takes two things.
An identity core. A short, stable statement of who you are and how you sound, grounded in something firmer than five adjectives you jotted down over coffee with a friend. A vibegraph builds on established personality tests, like the Big Five and the Enneagram, and on the Japanese concept of Ikigai, which covers purpose and meaning.
A root file. One plain-text file, VIBEGRAPH.md, that says who the work is for, where everything else lives, and what each tool is allowed to read. It's the switchboard that points AI to the right parts of your context layer.
A vibegraph has five parts:
Identity core: who you are (personality and Ikigai).
Brands: how you, and any business you own, show up.
Areas: your personal context, like finances, career, health, and journals.
Components: the tools you use, each registered with a kind and a scope.
Root file: the index that ties the parts together and sets the read order.
Every link in a vibegraph has a kind and a scope, so nothing private gets exposed just because you forgot to list it.
What it is not
A vibegraph isn't a notes app, a memory layer, a second brain, or an AI operating system. It complements all of them. Your Obsidian vault, for example, could be a knowledge store. ChatGPT or Claude memory is a memory store. Each earns its place in the context layer, and gets a kind and a scope inside your vibegraph. None of them, on its own, is the switchboard.
The short version:
Memory layers remember what happened. A vibegraph defines who you are.
Why a new word
In May 2007, Mark Zuckerberg popularized the term "social graph" at the first Facebook F8 conference. He used it to describe the network of real-world connections and relationships through which people share information and communicate.
Five years later, Google announced the Knowledge Graph, and within a couple of years "knowledge graph" was a common noun. It describes an interconnected network of entities, such as people, places, and concepts, and the real-world relationships between them, enabling search engines and AI to understand information as context rather than just isolated keywords.
In 2025, AGENTS.md took off with support from OpenAI and others as an open, vendor-neutral standard. It's a single Markdown file at the root of a code repository that gives AI coding agents context, rules, and constraints.
Carrying this forward: where AGENTS.md tells an agent how to work on your code, a VIBEGRAPH.md file tells an agent how to work not just for you, but as you, in your likeness and with your unique vibes.
The word "vibegraph" (lowercase) can be used freely as a common noun. You don't need an app to have one or build one. You only need to recognize what context is important relative to your use of AI, then use the framework to organize it.
Convention watch: the October count
To keep the trends visible, I track the same four rows every month.
Convention | What it is | Count |
|---|---|---|
AGENTS.md | How an agent should work on your code | 60,000+ open-source projects |
SKILL.md | A task an agent can run | 46 tools read it |
llms.txt | A site's map for language models | 8.6% of the top 100,000 sites |
VIBEGRAPH.md | Who the work is for | 0 public files, not counting my own or the example |
Method: agents.md's own count, the client list at agentskills.io, SEOmator's September 13 scan of the top 100,000 domains, and a GitHub code search on October 1, 2026. The first three show how fast a plain file spreads once it pays off. One honest caveat: Ahrefs found that 97% of llms.txt files got zero AI requests in May. A file spreading isn't the same as a file getting used. Also, many VIBEGRAPH.md files may stay private, on their creators' own drives or servers.
Try this week: the cold-start test
I'll end this first issue with a test for you.
Open a fresh chat in any AI tool you use (ChatGPT, Claude, Codex, Cursor). Ask it three things it should know about you or your business: who you are, what matters to you, how you write or work, or what you're working on. Is the answer accurate? Notice what it gets right and what it makes up. The gap between the two is what this newsletter will help you close.
I'd also like to hear from you. What are you working on, how are you building your context layer, and where has context engineering worked or not worked for you? Hit reply and let me know. I don't have all the answers myself, but I'm excited to learn this space with you and share what we find along the way.
Context engineers and AI startups
If you're building something in AI context, context engineering, or AI memory, or you have ideas or tools to share, reply to this email or connect with me on X or LinkedIn. I'll occasionally interview and feature the people and companies building across the context layer, and I'm always interested in what's new and emerging.
What's in your vibegraph?