The Personal Context Map: What to Include and How to Structure It
People are building personal context documents that run to nineteen pages. Six-section READMEs,...
People are building personal context documents that run to nineteen pages. Six-section READMEs, context routers, hand-maintained maps of who they are — all assembled from scratch, because there's no shared answer to a basic question: what actually goes in one?
So the documents sprawl. Half of what's in them is trivia the model could infer. Half of what matters is missing. The result is long, hard to maintain, and not much more useful than the bio it replaced.
A personal context map fixes that — not by adding more, but by giving you a structure to fill and a reason to cut.
What is a personal context map?
A personal context map is a structured inventory of what an AI should know about you — organized so a tool can read it precisely, not interpret it loosely. It's the difference between handing someone a filing cabinet and handing them a shoebox of receipts.
It is not a bio. A bio is prose, and prose forces the model to guess what matters. It is not a document dump either — pointing an AI at every note you've ever written buries the signal under everything else.
This is also distinct from how you build one. The method — extracting from your real sources instead of typing from memory — is its own subject, covered in How to Build Personal Context for AI. This piece is the spec: what belongs in the map once you're building it, and how to organize it.
How should you structure it? The four layers
Structure it in four layers, each holding a different kind of information and updating at a different rate: Identity, Profile, Mind, and Environment. The layers exist because not all context is the same. Your name doesn't change; your current project changes weekly. Mixing them into one flat document is what makes context maps go stale and bloated at once — you can't maintain a fast-changing fact buried next to a permanent one.
This is the same model behind what personal context actually is, used here as a build framework. When the map is served to a tool — over MCP or as a file export — these layers are what the tool reads.
Identity — the facts that don't change
Name, role, company, location. The orienting facts every tool needs before it's useful for anything, and the ones that change least.
Keep it to what's true year over year. Identity is small by design — if you're writing paragraphs here, something belongs in another layer.
Profile — how you work
Your background, expertise, communication style, and long-term goals. This is the layer that makes AI responses feel relevant rather than generic — it's the difference between an answer pitched at the average user and one pitched at you.
Include the things that change what a good answer looks like: your domain depth, how blunt you want feedback, the formats you actually use. Leave out a full CV. The model needs your operating manual, not your résumé.
Mind — what you're working on now
Active projects, current priorities, decisions in progress. The most dynamic layer, and the one most people skip — which is why an AI can know who you are and still have no idea what you're doing.
This is where a context map earns its keep, and where it rots fastest. A finished project that still reads as active is worse than no entry at all. Mind is the layer you revisit most often.
Environment — the people, companies, and tools around you
The people you work with, the companies you deal with, the tools in your stack, the meetings on your calendar. Context that separates a specific answer from a generic one — an AI that knows who your co-founder is and what stack you ship on doesn't have to ask.
Include the relationships and tools a task actually depends on. Leave out your entire contact list. Environment is a working set, not a directory.
What should you leave out?
Leave out anything the model can already infer, anything that's gone stale, and anything that's true but irrelevant to how you work. This is the half of the skill nobody teaches — and the reason most hand-built context documents underperform. A bloated map is as unhelpful as an empty one, because every irrelevant line dilutes the lines that matter.
Three things to cut:
- Inferable detail. You don't need to tell a capable model what a product manager does. State your role; let the model fill the generic parts. Spend your space on what's specific to you.
- Stale detail. Last quarter's priorities, a project that shipped, a preference you've since changed. Old context doesn't just sit there harmlessly — it actively misleads.
- True-but-irrelevant detail. Your hobbies rarely change how an AI should answer a work question. If a fact doesn't shift what a good answer looks like, it's noise.
The test for any line: does this change the answer? If it doesn't, it's not earning its place in the map. For a fuller treatment of this judgment call, see What Information Should You Add to Your AI Context?.
How do you keep the map from going stale?
You keep it current by treating the map as a living document, not a one-time setup — the Mind layer especially needs regular revision, since it tracks what changes weekly. A perfectly structured map built once and never touched degrades into the same misleading mess as no map at all.
That's a problem worth its own treatment, because stale context fails quietly: the AI sounds confident and works from facts that expired months ago. The full case for maintenance — why it matters and what it costs to skip — is in Why It's Worth Building (and Keeping) Your Personal Context for AI. The short version: a map is only as good as its freshest layer.
Build the map once, structure it right
A personal context map is what turns "give the AI more about me" into something an AI can actually use. Four layers, each holding the right kind of fact, with everything inferable, stale, or irrelevant left out. Structure is what makes it usable; restraint is what keeps it useful.
Unabyss builds your context map from your real sources and serves it to any AI tool, structured into these layers automatically — so you're not hand-maintaining a nineteen-page document that drifts the moment you close it. Connect your sources, and your map is generated in under 90 seconds.
[Get started free →]
