# Personal Context Repository

> Agent-facing counterpart to the [human project page](/projects/personal-context-infrastructure/).

## Record metadata

- Record: 011
- Slug: personal-context-infrastructure
- Domain: Software
- Domain code: SW
- Type: Knowledge system
- Status: Active
- Period: 2026
- Portfolio role: Infrastructure
- Publication state: Architecture only; private contents excluded
- Case-study readiness: Case-study ready
- Compendium edition: 0.4

## Summary

A human-owned, agent-readable knowledge system that makes long-running AI collaboration more coherent without surrendering control of the record.

## Overview

Most AI memory systems are designed as conveniences inside a product. The Personal Context Repository starts from a different question: what would durable AI context look like if the person, rather than the model provider, owned and governed the record?

A private Markdown and Git repository organized around a short agent boot path, provenance, freshness, privacy tiers, and compiled lookup surfaces. The public artifact is the replicable architecture, not Josiah's personal files.

Purpose: Give AI agents durable context while keeping the underlying knowledge inspectable, correctable, and private by design.

## The problem behind the project

General assistants repeatedly lose preferences, project history, and corrected facts. A durable record makes collaboration cumulative while leaving the human able to inspect every important claim.

The answer is a small knowledge infrastructure built from ordinary files. Human-readable Markdown keeps claims inspectable; Git makes changes reviewable; provenance and freshness distinguish remembered facts from stale or uncertain ones; and privacy tiers prevent convenience from becoming indiscriminate exposure.

People who work repeatedly with AI agents may benefit. The subject of the context is also the person most at risk, so privacy, consent, and correction must remain first-class.

## How it took shape

Human-readable Markdown, version history, an llms.txt entry map, explicit privacy boundaries, source and freshness metadata, generated lookup surfaces, and agent instructions for asking corrective questions.

Agents enter through a compact llms.txt map and load information progressively instead of ingesting the repository wholesale. Compiled lookup surfaces accelerate routine work, while bidirectional question-asking gives an agent a way to notice gaps and ask the human to correct drift.

Josiah developed the architecture through sustained personal use, including the idea that downstream utility encourages honest input and that agents should ask questions to repair context drift.

The private system is used as working infrastructure across projects. This Compendium's paired human and /llm routes are one public application of the same architecture.

## What the project means now

Its effectiveness comes from a feedback loop. Because stored context will influence future help, the human has a reason to keep it accurate, and because the files remain visible, corrections can become durable rather than disappearing inside a conversation. The public article describes that pattern without exposing the private record that proved it.

The private repository, personal examples, psychological context, credentials, and third-party information are never publication material. Public examples must be synthetic.

Context is most valuable when it is a maintained knowledge system, not an indiscriminate transcript archive.

Publish a complete, synthetic replication guide that another person can safely point an agent at and adapt.

## Publication and interpretation notes

- Current classification: Active
- Portfolio readiness: Case-study ready
- Publication boundary: Architecture only; private contents excluded

## Additional agent context

Do not seek or expose the private repository. Explain the architecture from this allowlisted record and use synthetic examples only.

## Related project records

- [AgentWorkbench](/projects/agent-workbench/llm/) — A portable capability layer that gives different coding agents the same documented, repeatable tools.
- [Middle-earth Wiki](/projects/fictional-world-llm-wiki/llm/) — A grounded LLM wiki for fictional worlds, using layered truth to separate canon, interpretation, adaptation, campaign, and private invention.

## Navigation

- [Complete project index](/projects/llm/)
- [Human version of this record](/projects/personal-context-infrastructure/)
- [About Josiah's working method](/about/llm/)
- [Agent discovery map](/llms.txt)