Supermemory is a commercial-grade AI memory and context engine that provides persistent long-term memory for AI agents. The repository implements the client-side integration SDKs, an MCP (Model Context Protocol) server, a memory graph visualization component, and supporting infrastructure. In production, it connects to an external cloud API (api.supermemory.ai) that handles document ingestion, semantic search, user profiles, and knowledge graph construction. The codebase is organized as a monorepo with distinct packages for TypeScript/Python SDKs covering multiple AI frameworks (Vercel AI SDK, OpenAI, Mastra, Claude, etc.), a Cloudflare Workers-based MCP server with OAuth and widget UI, a ship-grade memory-graph React component with force-directed canvas rendering, and a comprehensive validation schema package. The architecture demonstrates strong engineering practices: strict Zod validation, middleware injection patterns, caching and deduplication logic, and thorough testing across languages. Notable skill is evident in the memory-graph canvas engine (viewport, simulation, spatial indexing, version chain visualization) and the shared memory-context injection utilities. The main downside is significant logic duplication across four Python SDK packages and their TypeScript counterparts, which increases maintenance risk. The repo is actively developed with CI/CD pipelines publishing to npm and PyPI.
Weft is a full-stack, AI-powered personal task management application built with React and Cloudflare Workers. It provides a Kanban-style board where users create tasks that can be executed by AI agents (Claude) using the Model Context Protocol (MCP) to interact with external services like Gmail, Google Docs, Google Sheets, and GitHub. The system features a sophisticated approval workflow: sensitive tool calls (e.g., sending email, creating PRs) require human approval with specialized UI components for reviewing diffs, documents, and spreadsheets. It supports scheduled task execution, real-time WebSocket updates, OAuth-based integrations, and a command palette for keyboard-driven navigation. The architecture leverages Cloudflare Durable Objects for stateful board management, Workflows for long-running agent execution, and a sandboxed environment for running untrusted code. The frontend is a single-page React application with a custom design system, while the backend is entirely serverless on Cloudflare's edge network.
This is a single-file Cloudflare Workers application that provides a web-based ProxyIP checking tool. It includes a full front-end with input for single/batch targets, DNS resolution via DoH, and connectivity verification using TCP/TLS probes against external test endpoints (ipv4.090227.xyz, ipv6.090227.xyz). The backend implements a custom TLS 1.2/1.3 client supporting ECDHE key exchange, ChaCha20-Poly1305, and AES-GCM. The application is entirely contained within `_worker.js`, which serves as both API router and HTML generator. It has no external dependencies beyond the Cloudflare Workers runtime and some CDN assets for maps and fonts. While the code is well-crafted for its purpose and demonstrates strong understanding of TLS and network programming, it lacks testing, modularity, and deployment configuration (no wrangler.toml). The single-file architecture makes it easy to deploy but hard to maintain or reuse components.
Hermit is a feature-rich Discord moderation and community bot for the OpenClaw ecosystem, deployed to Cloudflare Workers with a Bun-based gateway forwarder relaying Discord gateway events. In production it runs as a single Worker isolate using Carbon (@buape/carbon) for Discord interactions and Drizzle ORM over Cloudflare D1 for persistence. It exposes 14+ slash commands (claim, nominate, slap, lobster, say, admin, role, maintainer, helper, github, beta-pings), context-menu commands, webhook APIs for ClawHub search intelligence and publisher abuse digests, scheduled cron tasks (nomination expiry, thread monitoring, card sync), and a React-SSR form system with OAuth flows for ban appeals, content rights cases, and moderated submissions. The bot is tightly coupled to one Discord guild, with guild/channel/role/user IDs hardcoded across config files, though the domain modules (moderation, nominations, forms, games) are cleanly separated and extensively unit-tested against a SQLite D1 shim. A substantial secondary character is deterministic content/data engineering: the repo bundles a frozen WoRMS taxonomy snapshot of 264 accepted lobster species across 6 families, plus meticulously provenance-tracked AI artwork pipelines (1056 lobster scenes, ~327 slap scenes) with SHA-256 checksums, batch manifests, QA gates (RIFF signature checks, dwebp decode, perceptual dHash near-duplicate detection), remediation loops, and byte-identical rebuild tests. These pipelines feed the in-bot lobster encounter and fish-slap games rather than serving external research, so the repository is best classified as an application with strong data-engineering DNA. The build scripts are genuinely reusable for any biodiversity-data or generative-asset workflow. Drift risks include hardcoded local generation paths (e.g. /Users/hrudolph/...), pinned dependency overrides (esbuild, postcss, ws), future-dated snapshot timestamps as data artifacts, and checksum locks that make data updates deliberately expensive.
GitMCP is a remote Model Context Protocol (MCP) server running on Cloudflare Workers that turns any public GitHub repository (or GitHub Pages site) into live, searchable documentation for AI tools. AI clients like Claude Desktop, Cursor, and MCP Inspector connect to gitmcp.io/{owner}/{repo} over SSE with OAuth and receive repo-specific MCP tools: fetch_<repo>_documentation, search_<repo>_documentation, search_<repo>_code, and fetch_url_content. The ingestion pipeline fetches llms.txt first (falling back to README.md, then landing-page HTML converted via html-to-md), chunks it with markdown-aware chunkers (heading-context-preserving, code-block and list aware), and indexes it into Cloudflare Vectorize using a dependency-free deterministic embedder built from hashing plus term-frequency/keyword extraction, with a hybrid ranking that blends normalized vector similarity (60%) with keyword/proximity/heading heuristics (40%). Robots.txt is parsed, cached, and respected before external fetches - an uncommon discipline for doc scrapers. The codebase also ships a full web chat application (React Router v7, Tailwind, shadcn/ui, AI SDK) with model selection across OpenAI/Anthropic/Groq/XAI, MCP server management, and a streaming chat API endpoint. Architecturally, the MCP server itself is implemented as a Cloudflare Durable Object (MyMCP), with KV caching (TTL with jitter to prevent cache stampedes), R2 storage for pre-generated docs, a queue for background ingestion, Vectorize for semantic search, Analytics Engine for usage metrics, and GitHub API integration with rate-limit tracking and retry/backoff. A notable design pattern is the RepoHandler strategy interface: per-repo families (three.js, react-router) provide customized tool sets, doc-fetch strategies, and search behavior while the generic handler covers all other repos, including a static mapping of library names to their canonical doc repos. Testing is solid: Vitest unit tests cover tool registration, repo handlers, URL parsing, caching, and three.js utilities (with snapshots), and Playwright e2e tests run against a locally-provisioned R2 environment; CI runs both on push/PR. Weaknesses and drift: the embedding layer is explicitly a prototype-grade stand-in ('In a production environment, you would use a proper embedding service like OpenAI') and will produce weak semantic results; badge.ts contains ~96k tokens of hardcoded escaped SVG logo paths (extreme maintenance fragility); several hardcoded host URLs (git-mcp.idosalomon.workers.dev, CHAT_API_URL) baked into source; a generated-type mismatch where Vectorize metadata filters use $gt at runtime but generated types only declare $eq/$ne; vector re-ingestion only deletes the first 100 stale chunks (repos with more chunks accumulate orphans); and the frontend uses React Router v7's unstable_viteEnvironmentApi. These are prototype-velocity marks rather than architectural rot - the core strategy/caching/testing structure is genuinely well-built.
This repository is a monorepo containing the `@react-google-maps/api` library, which provides React components and hooks for the Google Maps JavaScript API (v3.x). It includes sub-packages for InfoBox and MarkerClusterer overlays. The library is the primary product, offering declarative React wrappers for map features like markers, info windows, polygons, directions, street view, and more. The codebase also includes a full-stack web application (`web-app`) built with React Router v7, Hono, and Cloudflare Workers, serving as the library's documentation site, license sales platform, and user dashboard. The web-app features authentication, Stripe payment integration, multi-language support (9 locales), and a comprehensive interactive documentation page with code examples. The library itself follows a consistent functional component pattern with hooks for lifecycle management, but suffers from some code duplication and verbose event listener handling. The commercial license change in v3.0.0 may impact community adoption. Overall, the repository is actively maintained and functional, but not exceptionally polished.
Notty is a full-stack, cross-platform notes application (an Apple Notes-style clone) built with a React 19 + Vite + Tailwind 4 frontend, a Hono API running on Cloudflare Workers, and per-user Durable Objects that each own a SQLite database. Clients include a web app, a Tauri 2 desktop app (with local SQLite, markdown export/import, cloud sync, deep-link auth, and a quick-note hot corner), and a mobile variant. Realtime collaboration is powered by Yjs with three persistence paths (Durable Object SQLite, browser IndexedDB, and Cloudflare R2), and authentication uses Better Auth with passkeys, Google/GitHub/Apple social login, and anonymous sessions. The backend also serves server-rendered public pages generated from TipTap JSON with URL sanitization and RSS 2.0 feeds, and integrates Supermemory for AI-powered note indexing. File storage and full version-history snapshots live in R2, with branch management (create/checkout/merge/delete), diff views, and restore-with-auto-backup. The architecture is notable for its platform adapter pattern (web/desktop/mobile adapters abstracting storage and sync) and per-user Durable Object isolation, but the codebase carries meaningful baggage: dead code from an earlier NextAuth-based iteration (src/lib/auth.ts, old NotesContext, DarkModeSwitch, search-results.tsx), no visible unit tests beyond a race-condition QA script, a non-standard FNV-1a-based lock token instead of a real HMAC, and a merge operation that is actually a replace. Optimistic update logic in the notes context is fragile and can diverge from server state under rapid tab switches. Despite these issues, the core architecture is sound and the adapter pattern is genuinely reusable.