LLMs Remember Your Code, Not Your Life: Building a Portable Personal Context Layer
September 8, 2026 – 6:04 pm
Personal context should reside in a store you own; an open protocol in the middle keeps store, assistant, and device choices independent.
Summary
Context management for AI-assisted coding has evolved into a rich ecosystem: convention files, repository maps, memory banks, and full-fledged memory frameworks all compete to ground coding agents within your codebase. However, when you step outside the terminal and ask an assistant about your health or finances, the tooling nearly disappears. Your personal context currently lives in a vendor’s memory feature, locked to that vendor, or nowhere at all.
This article argues that the missing piece is not a smarter memory system but a separation of concerns: place personal context in storage you own, independent of any AI vendor, and let every assistant connect to it. It describes an open-source implementation of this principle, a Git-hosted Markdown vault served to multiple LLM surfaces through a single remote MCP endpoint, along with the trade-offs this design accepts.
Running this system on edge infrastructure also revealed a problem that extends beyond it: edge platforms prevent runtime code generation, silently disabling the fast path of the TypeScript ecosystem’s standard validation library.
The article covers the fix: an open-source ahead-of-time schema compiler that grew out of this project and now validates every request the system serves. It then explores future directions as external services expose MCP interfaces, transforming chat into a space where real-world tasks get done. A user-owned context layer, independent of any device, is what makes those actions personal.
For coding, context management is nearly a solved problem
If you write code with an LLM today, you have a plethora of options for feeding it context:
- Convention files: CLAUDE.md for Claude Code, AGENTS.md as a cross-tool standard, Cursor rules, GitHub Copilot’s custom instructions.
- Structural tools: Repository maps that compress a codebase’s shape into the context window, plus codebase indexers and code-search integrations.
- Persistence layers: Cline’s Memory Bank pattern, Cursor’s session memories, and Claude Code’s memory directory.
For teams building custom agents, general-purpose memory frameworks like mem0, Letta, and Zep offer retrieval pipelines, ranking, and temporal knowledge graphs tracking fact validity.
Everything else is an afterthought
Now, step out of the terminal. The other half of LLM use, perhaps the larger half, is conversational: you pose a question or present a problem in a chat interface and expect an answer relevant to your situation. What did my bloodwork trends look like before I changed my diet? Given my actual portfolio, what impact does this market move have on me? What did I decide the last time I evaluated this vendor?