1689 projects published
A guide website about living in Berlin that the author relies on as his livelihood. AI systems train on its content and divert its audience, and traffic is down 80-85% compared with the same month pre-LLMs. Work in the comment focuses on cutting operating costs to the bare minimum: shutting down one service, replacing a VPS with Cloudflare Workers, trimming SaaS subscriptions, and self-hosting what he can.
A Win/Mac/Linux desktop app for lawyers that points to a filesystem-based project catalog and lets them quickly spin up AI agent sessions by selecting a project via autocomplete search and starting an agent (Claude Code, Codex, OpenCode) in an isolated VM, providing strict context management and agent isolation on their own machine.
This site lets couples create wedding invitation websites where guests can RSVP directly on the invitation. Users get a personalized link for their invitation with different domain options.
Deepreed is an AI-assisted writing tool that constrains LLMs to make targeted edits on a document instead of generating large blocks of text. This keeps most of the writing in the author's own style and easy to understand, while still leveraging the strengths of LLMs, and avoids wading through long LLM output where mistakes slip through.
A cloud development environment for running coding agents, built around a task management system that lets agents split large jobs into subtasks and run each one on a different agent, model, or machine. Progress is visible at a glance, and tasks can run autonomously for long stretches (weeks at a time) with only occasional steering input to the coordinator.
A bookmarking and read-later app where bookmarks are private by default but can be shared publicly. Recent features include tags, collections, and importing browser bookmarks, and there is a companion Chrome extension.
A Unix/Linux environment that runs in the browser, letting users work in a Linux-like shell from a web page.
A work-in-progress relational and logic programming language inspired by Datalog and Bruce MacLennan's work on relational programming. It adopts the open-world assumption to allow missing values and heterogeneous data while enforcing data integrity, and is motivated by the idea that logic languages can provide strong correctness guarantees and a source of reinforcement learning in the LLM era. It favors relational data models over the object-oriented view.
Papillon is a real-time collaborative workspace for product teams that brings chat, docs, files, links, tasks, whiteboards, and calls into one place, organized around project channels as an alternative to jumping between Slack, Notion, and Linear. It emphasizes speed, a hacker/IDE feel, and offline support, and lets teams build small internal apps with AI directly inside the workspace — such as shared support rotas, spreadsheets, dashboards, and custom workflows — running in an isolated Cloudflare OS sandbox. It is available as a web app and a Tauri-based macOS app, with security/trust info and uptime history published on the site.
This service manages the polling layer for the WhatsApp Agent API: it handles polling, cursors, webhooks, and retries so serverless apps that scale down to zero can send and receive messages via plain HTTP requests and webhooks. Agent tokens are stored encrypted.
A tool for quickly eyeballing the stylistic choices of various LLMs side by side, focused on models at 4-bit quantization under 100GB of weights, as a qualitative alternative to benchmark-driven model comparison.
Micro is a personal assistant and home server application, originally started as a Go framework and evolved toward its original mission of being a Google alternative or superapp. It currently ships as a PWA, with an Android app in development.
An open-source, fully customizable AI workspace in which the author does most of their day-to-day work.
This writing app features a revamped writing screen with better themes plus text checks and features brought in from other parts of the app, and a new Electron-based desktop build integrating Harper.js for grammar checks. It aims to be useful for writing overall rather than only surfacing possible text improvements or a readability score.
This support workflow app adds text expansion alongside its existing support tooling, now covering everything the author needs for a day job in support; the latest improvements are pending release.
An open-source personal website and blogging platform made in Edinburgh, offering all the features wanted for personal sites and blogs.
AI detection tool that analyzes the writing process rather than the final draft. It tracks unusual changes, including pastes, throughout the full revision history to give teachers a clearer picture of how much AI-generated text is present at any time, so they can discuss it with a student before jumping to accusations.
AI agent that learns how a developer writes code by reading their repository and automatically extracting DSL recipes from how they structure endpoints, handle data access, write workers, use queues, and organize tests. Recipes combine across stacks and patterns into millions of variations, aiming to make code generation deterministic, cut token usage and agent wall-time, and follow existing code conventions.
An AI mock interview tool for system design interview preparation. Users practice system design interviews with AI to learn how to assemble studied material into a working design under interview conditions; it is currently in beta with free access offered in exchange for feedback.
Tutor is an open-source local AI study tool that extracts concepts from study materials (books, online documentation, and YouTube courses), generates lessons in simple language, and tests the learner. Users plug in whichever API provider they prefer with their own key; the author uses it with GPT Luna and Haiku.
This Apple Watch app is a voice-first second brain that captures thoughts instantly from the wrist using the Ultra's Action button or Siri keywords, without needing an iPhone nearby (using cellular or Wi-Fi). Audio is transcribed using EU-based speech-to-text with zero data retention and turned into searchable notes and summaries.
A fully open-source voice-to-text transcription pipeline for Linux, billed as the fastest and most accurate Linux transcriber. The author patched upstream libraries like ONNX to make them work with latest models such as Parakeet.
A directory website listing books that have been banned or pulled from Amazon for being too 'spicy', built as a way to track a topic of interest.
A directory website tracking the fan game decompilation and recompilation scene, which is rapidly growing with the rise of AI reverse engineering tools.
This product provides modern, self-custodial money movement, letting users move and control their own funds without relying on custodial banks. The author notes its relevance in light of the recent Ueno Bank collapse in Paraguay.
This is an IoT automation platform built on top of AtomVM that lets AI agents control a fleet of devices, providing device management, telemetry, and OTA updates. It exposes device capabilities (sensors, actuators, cameras) as tools via MCP, giving AI agents the ability to interact with the physical world. It supports all ESP32 variants and has partial support for the RPi Pico 2 W; the backend is Elixir/Phoenix and the firmware is written in Gleam with occasional Erlang.
A privacy-respecting, fully open-source personal task manager with a weekly planning session so tasks don't get forgotten. Tasks can be added easily by forwarding an email or sending a Telegram message. Free of ads, trackers, and subscriptions, and self-hostable for those who want to run it themselves.
A multi-player LLM interface built on an underlying graph data structure, combined with a publishing platform, that lets people work together to explore deep questions. Built with Elixir, Phoenix, LiveView, and Cytoscape.
A local-first archive for AI conversations that imports chats from providers lacking proper export options, such as Google AI Studio, OpenRouter, and LM Studio. It supports semantic search, MCP integration, and continuing old conversations with llama.cpp.
Open source Android app for splitting group expenses on trips, in shared flats and at events. Requires no account and has no ads or tracking, keeping all data on the phone. Groups are shared directly between friends' phones over Wi-Fi or Bluetooth using QR-code pairing with end-to-end encryption and no server involved. Expenses can be added by voice or by reading a receipt with the camera. Version 2.4 in development adds a simpler way to invite people into a group and settling up with the fewest payments.