distributed-ai-memory-system

A personal, cross-provider memory server: one Cloudflare Worker that lets any MCP-capable AI client (Claude, ChatGPT, Gemini CLI, Claude Code) read and write structured memory for your projects, backed by D1 (registry) and R2 (per-project JSONL blobs). Ships with a small web UI for browsing and chatting with a project's memory via the Anthropic API.

Deployed at memory.mahmoudbebars.dev.

Stack

• Hono on Cloudflare Workers (no Express, no Next.js)

• D1 via Drizzle ORM — registry/index of projects

• R2 — raw memory.jsonl per project, append-only

• Zod validation at every route boundary

• React + Vite + Tailwind frontend, served as Workers Static Assets from the same deployment (no separate Pages project)

Related MCP server: SkyBrain

Local development

npm install
npm run db:migrate:local
npm run dev            # Worker on :8787

cd app && npm install && npm run dev   # frontend on :5173, proxies /api to :8787

First deploy

See DEPLOY.md.

Project structure

src/
  index.ts              Worker entry — mounts /api routes, falls back to ASSETS
  lib/bindings.ts        Shared Env/Bindings type (D1, R2, ASSETS, API key)
  db/schema.ts            Drizzle table definitions
  modules/
    projects/            4-file module: schema.ts, service.ts, routes.ts, index.ts
    chat/                 Same pattern — Anthropic API chat over a project's memory
migrations/               D1 migrations (wrangler-managed)
app/                       Vite/React frontend, built into app/dist and served by the Worker

Status

See PROJECT_UNDERSTANDING.md for the full architecture and what's still ahead (MCP tool layer, OAuth, sync CLI).