Skip to content

Home

Jump to bottom
Kritarth-Dandapat edited this page Jul 10, 2026 · 2 revisions

context-kernel

context-kernel is a small, self-hostable context memory for LLMs. You hand-curate a set of Markdown files about yourself and your work; a Cloudflare Worker serves them to Claude (Claude Code, Desktop, and browser chat) over a single authenticated remote MCP connector. Agents can extend the memory through an append-only journal, but only you promote journal notes into the curated source of truth.

The problem it solves

If you run agentic sessions across several machines and browser chats, you end up re-feeding the same context every session: who you are, your goals, how you like prose written, how figures should look, how you want answers delivered. context-kernel puts that in one place you control, reachable everywhere Claude runs, that Claude pulls itself instead of you pasting it.

What it is

  • Infrastructure that answers one question: "what context should Claude know about me and my work right now?"
  • Hand-curated Markdown as the single source of truth, versioned in a private git directory.
  • A Cloudflare Worker that serves that Markdown as MCP tools over HTTP, gated by bearer tokens.
  • An append-only journal that agents can write to, and only the owner promotes by hand into curated content.

What it is not

  • Not a personal website, resume renderer, CMS, or chatbot.
  • Not a public, unauthenticated API. No HTML frontend.
  • Not a vector-store memory tool that extracts facts automatically. See Design-Decisions for why that distinction matters.

Where to go next

  • Architecture — the content -> build -> KV -> Worker -> MCP pipeline, and the KV key scheme.
  • Security-Model — the two-token read/write split, constant-time comparison, and the trust boundary around journal promotion.
  • Deployment — how to stand up your own copy on Cloudflare Workers and KV.
  • Design-Decisions — the settled architectural decisions (D1-D7) and why this project exists instead of an off-the-shelf vector-memory tool.
  • Contributing — repo layout, build phases, scripts, and the hard rules that govern any change to this codebase.

Project status

As of this writing, the engine (artifact generator, token auth gate, and the full MCP tool set — both read tools and the journal write tools) is implemented and tested. The interactive promotion script (scripts/promote.ts) and the personal Skill (skill/context-kernel/SKILL.md) are still stubs. See Contributing for the phase-by-phase build plan and what is done versus outstanding.

Clone this wiki locally