01 · MYCONTEXT
MyContext
Personal context infrastructure for AI
Accumulated personal context — ideas, decisions, research, projects, conversations — preserved across AI sessions with clear provenance. Not a profile store. A record of what you've built and thought about.
02
What it is.
MyContext is personal context infrastructure for AI. It gives a single engineer a structured way to store, retrieve, and reason about accumulated knowledge — codebases, decisions, notes, research findings, plans, conversations — in a format that AI agents can actually use.
The key distinction: this is not a user-profile memory store. It preserves the full intellectual history of what you've worked on, thought about, and decided. Ideas, suggestions, advice, tasks, projects, conversations — all of it. And crucially, it tracks provenance: AI-generated proposals and inferences are distinguishable from user-confirmed facts and decisions.
Unlike generic RAG systems, MyContext is built for a single user's actual workflow. Projects are units of context. Decisions are first-class objects. Everything is related.
03
Why I built it.
Every AI coding agent I use starts each session with no memory of what I told it yesterday. I keep repeating myself — project constraints, architectural decisions, coding conventions — because nothing persists between sessions.
Most solutions treat this as a "memory" problem: dump everything into a vector store and hope retrieval works. But the real problem is more specific. I need my context to survive session boundaries, to be queryable by an agent, and to carry provenance — so the agent knows whether something was my decision or its own inference.
MyContext is built for that. Filesystem-first so I can browse and edit with normal tools. SQLite-backed so queries are fast and relationships are explicit. Content-addressed so integrity is verifiable. And with a Source Layer that preserves conversation history with external AI systems as immutable evidence — not as part of my semantic memory, but as proof of where each idea came from.
04
How it works.
The repository lives at ~/.mycontext/. It combines
three storage layers:
The filesystem tree is the primary organization.
Context lives under human-readable paths — projects/,
finance/, health/ — so you can always
browse, edit, or back it up with normal file tools.
The SQLite metadata layer stores records, revisions, relationships, and permissions. It provides ACID guarantees for writes and supports as-of reads on the revision chain.
The content-addressed object store holds the actual data blobs, keyed by SHA-256. This enables transparent deduplication and corruption detection.
The Source Layer (Phase 3B.3) preserves conversation history with external AI systems as immutable evidence. Imported conversations are stored raw — open string provider, role vocabularies, per-entity canonical envelopes. Importing never creates semantic ContextRecords; the source material is only accessible through the referencing ContextRecord's hosting node path.
05