When five notes aren’t enough: build a context system
Once you have several projects and repeated AI conversations, the problem changes. Capturing notes is no longer the hard part. The hard part is knowing which current background belongs in which decision.
A personal Context OS is a maintained set of notes and rules for that routing. “OS” describes the operating pattern, not a separate app and not an assistant that remembers everything for you.
What it contains
At minimum, the system needs:
- current goals and priorities;
- constraints and commitments;
- decisions with their reasons;
- project-specific current context;
- standards or preferences that genuinely change an answer.
It also needs boundaries: what can be shared, what must be blurred, what stays local, and what never enters an AI workflow. The four buckets are part of the system, not an afterthought.
What makes it an operating system
The value is in repeatable movement. A new fact enters through Capture. It is placed with the project or decision it affects. A consultation draws only the relevant notes. The resulting decision and next action return to the source.
Without that return path, the notes gradually describe an old version of your work. Maintenance is therefore a core function: review current context, retire stale claims, and keep the reason behind changed decisions.
When to build beyond the basics
Do not start here if five notes and one project briefing are enough. Build a wider Context OS when the same routing problem repeats across projects, when stale context causes confusion, or when you need clearer privacy boundaries between areas of work.
Automation can help retrieve or format context, but it should remain inspectable and reversible. It should not silently decide which sensitive material leaves your vault or convert AI suggestions into commitments.
Add three layers when one briefing stops scaling
An advanced context system needs separation more than volume.
| Layer | Question it answers | Example note | Update trigger |
|---|---|---|---|
| Durable | What stays true across projects? | standards.md, working-principles.md | a principle or quality bar changes |
| Project | What is true for this body of work now? | projects/studio-site/current-context.md | scope, constraint, owner, or next milestone changes |
| Decision | What does this one choice require? | decisions/new-client-pricing.md | the choice is made or its review condition fires |
Do not copy every durable rule into every project. Link to the source locally, then assemble only the lines needed for a handoff. This keeps one correction from becoming five inconsistent edits.
Use a routing receipt for every AI handoff
When multiple projects and privacy boundaries are involved, record what was selected and what was deliberately left out. This makes retrieval auditable without saving the entire AI conversation.
# Context routing receipt
Decision: [one decision]
Prepared: YYYY-MM-DD
Included:
- [source note] — [selected claim]
- [source note] — [selected constraint]
Blurred:
- [detail] → [safe generalization]
Kept local:
- [material omitted and why]
Freshness checks:
- [claim] verified on [date]
- [claim] must be reviewed when [condition]
Question sent:
- [exact question]
After consultation:
- Decision:
- Reason:
- Source note updated:The receipt is useful when the same system serves several clients or work areas. It shows the boundary used for one consultation and provides a return path to the source. It is not a license to log secrets; describe omitted material by category, not by copying it.
Know when automation is ready
Automate a routing step only if all four statements are true:
- the manual selection has worked repeatedly;
- the source of each included line is visible;
- sensitive categories have an explicit default action;
- a human reviews the final handoff before it leaves the system.
If routing rules change every week, automation will encode confusion faster. Keep using a manual Current Context note until the repeated boundary is obvious.
Begin with a minimal vault, then choose one repeated workflow. The larger system should grow from those working pieces.
Join the Starter Kit waitlist if you want one notification when the kit is ready.