mem0-rs
A Rust implementation of Mem0 built around the parts agent memory actually depends on: durable storage, semantic search, metadata filters, and predictable retrieval. The goal is not another memory demo. It is a small binary that can sit inside agent infrastructure without becoming a second application.
02conquer
A deterministic real-time strategy experiment in Godot 4, built from the lessons exposed by the open Command & Conquer engines. The interesting work is under the visible game: simulation ticks, command ordering, replayability, and the small rules that keep a strategy game honest.
03severe-testing
A reusable agent skill for testing that tries to falsify the system instead of confirming it politely. It pushes adversarial inputs, failure injection, recovery checks, and security-boundary validation into the normal coding-agent workflow.
04codex-swift
A native macOS port of the Codex coding-agent harness. It treats the agent as a long-lived local service with process isolation, Keychain-backed auth, sandboxing, MCP, hooks, and persistent memory, rather than as a chat window wrapped around shell commands.
05memory-wiki-voice-dialog
A Cloudflare-native memory wiki that ingests sources, scores novelty, builds a knowledge graph, and exposes the corpus through MCP and voice. It is a place to study how memory, retrieval, and spoken interaction change when the knowledge base is a living system rather than a pile of notes.
06bloc
A self-hostable block workspace with a Notion-compatible API, realtime sync, databases, permissions, automations, search, and AI surfaces. It asks how much of the modern workspace can be made inspectable again without losing the habits people have learned from Notion.
07warden
A Bun-native reimplementation of Sentry, used as a way to understand what observability products are made of when the labels are removed. Ingestion, grouping, traces, profiles, replays, alerts, SDK boundaries, and columnar storage all become explicit product decisions.
08devrel-almanac
A cross-referenced almanac of Developer Relations: history, frameworks, companies, people, metrics, tactics, and the AI-era shift in audience. It is meant to make the discipline easier to inspect without reducing it to advocacy slogans or vanity metrics.
09granite
A local-first, Markdown-native knowledge base for plain-file vaults. Granite is less interested in imitating a sidebar than in preserving the contract behind it: links survive, frontmatter stays text, Canvas and Bases round-trip, and the vault remains the authority.
10valkey-bun
A Bun-native Valkey and Redis-compatible server for local development loops. It implements the commands application code leans on most often: strings, hashes, expiry, pub/sub, streams, transactions, and the administrative surface needed to inspect what is happening.
11claude-code-goal
A markdown-backed goal mode for Claude Code that keeps autonomous work pointed at the same objective across long sessions. Files and hooks make the commitment inspectable, so the agent has something stronger than conversational memory to return to.
12ProseDown
A native macOS Markdown viewer built around startup speed, clean rendering, and smooth scrolling. It treats Markdown as something people actually read, not as a preview pane bolted to an editor.
13open-cloud
A local, inspectable replica of the Cloudflare developer platform: Workers, D1, KV, R2, Durable Objects, Queues, Vectorize, and AI Gateway loops. It exists for the moment when a coding agent needs to run the real integration test without waiting on the internet.
14metrocore
A city-builder experiment in the tradition of SimCity 4, with simulation systems under the visible civic machinery. The interesting parts are zoning pressure, infrastructure, feedback loops, and how a city becomes legible at small scale.
15security-scan-mcp
An MCP server that gives coding agents CVE-shaped prompts for targeted security review. It uses real vulnerability patterns to make review sharper than “look for security issues” and more grounded than generic checklist scanning.
16interview-conductor
An interview-to-essay experiment that keeps asking until the material has enough shape to become a piece of writing. It treats conversation as source work: follow the concrete claim, preserve the useful detail, and avoid turning the result into summary paste.
17code-search
A fast code search tool for agents and humans who need orientation before opening files. It combines exact search, regex, and BM25-ranked discovery so a coding agent can find the right evidence without turning every investigation into a long grep séance.
18ec-developer-dashboard
A developer metrics dashboard for following the path from first touch to mainnet transactions. It treats DevRel measurement as a chain of observable handoffs, not a pile of activity counts.
Standards and earlier contributions
Specifications and platforms I contributed to before this list.
-
Apache Shindig
Chief author of the PHP OpenSocial reference implementation; PMC member and committer.
-
OpenSocial Spec
Co-authored the specification defining APIs for web-based social applications.
-
Portable Contacts
Co-authored the contact data exchange specification for social networks and address books.
-
OAuth 2.0
Contributed to the authorization framework used by modern API security systems.
-
OpenID Connect
Contributed to the identity layer on top of OAuth 2.0.
-
OpenAPI
Contributed to the API description standard for documentation and code generation.