Products / Ideas
Things people are building with LLMs
Peisinoe: prompt composition library to decouple prompt logic from app code
Peisinoe is a library that lets developers define prompts as reusable, composable parts, keeping prompt content and selection logic separate from application code. It solves the problem of frequent prompt changes requiring modifications to application flow, targeting developers who iterate on prompts.
Punchlist: iOS app to dictate UI improvements paired with screenshots
Punchlist is an iOS app that lets users dictate feedback or ideas about their app's UI while viewing it on their phone. It pairs the spoken text with a screenshot and sends the pair to a queue. Later, Claude reviews the tasks at the user's computer. It solves the problem of capturing spontaneous UI improvement ideas for developers, especially those with ADHD.
FetchSandbox: sandbox for testing AI agent integrations with real webhooks and failure scenarios
FetchSandbox is a sandbox layer that runs the full integration lifecycle for AI agents before production. It tests real workflows, real webhooks, and failure scenarios on demand, providing a public receipt URL as proof of survival. It solves the problem of AI agent integrations that pass tests but break on real webhooks due to duplicate events, non-idempotent handlers, and retries hitting stale state.
Clark: AI assistant with computer use and browser automation
Clark is an AI assistant that can use a computer and browser to complete tasks autonomously. It leverages LLMs for task planning and execution, aiming to match the capabilities of Manus agent. The tool is designed for users who need an AI agent that can perform complex, multi-step tasks across web and desktop environments.
Topsoil – Mac notch dashboard for coding agents, music, and files
Topsoil is a native SwiftUI app that lives in the MacBook notch, providing a hover-expandable panel with live terminal panes for coding agents (Claude Code, codex, cursor cli), now-playing music info, calendar, system stats, and a pixel plant that grows while you code. It solves the problem of alt-tabbing to check agent progress or manage files, targeting developers using AI coding assistants.
Text LLM training from scratch with PyTorch
A clean, readable codebase that implements the full LLM training pipeline (pretraining, SFT, DPO, GRPO/RL) using only PyTorch primitives, avoiding high-level abstractions. It helps developers understand the underlying math and mechanics of LLM training.
Turn YouTube videos into language lessons with transcript-based vocab and phrases
Lingooso is a web app that lets users paste a YouTube video URL and get a structured language lesson from its transcript. It breaks the transcript into digestible chunks, extracts vocabulary and phrases, and presents them for study. The app solves the problem of learning real, spoken language from authentic content rather than textbook phrases, targeting self-learners who want to learn from videos they already enjoy.
Coding agent that sees browser to iterate on UI changes
Peek CLI is a tool that lets coding agents observe the browser in real time, enabling them to iterate on UI changes by seeing the visual output of their code. It solves the problem of agents working blind on frontend tasks, making them more effective for developers who want automated UI development.
AI-powered construction estimating web app using MasterFormat
Quantiz is a web app that automates construction estimating by generating budgets, Bill of Quantities (BOQ), and Bills of Materials (BOM) based on the MasterFormat division system. It uses AI to streamline the estimation process for contractors, estimators, and tech enthusiasts.
Receipt Split: AI-powered restaurant bill splitter
Receipt Split is a free web app that splits restaurant bills by item. It uses AI to scan receipt photos and extract line items, tax, and tip, or allows manual entry. Users share a link with friends for real-time item selection, solving the problem of unfair bill splitting.
Open-source AI cofounder for solo founders
Shotgun is an open-source AI tool designed to act as a virtual cofounder for solo founders. It uses LLMs to assist with brainstorming, decision-making, and task management, helping solo entrepreneurs overcome the lack of a cofounder.
AI code auditor that only finds bugs, does not fix them
A constrained AI system designed solely to audit code for bugs, without attempting to fix them. It uses a custom architecture to force the LLM into a pure auditing role, tested on GitHub repos like Monica and Economizer, finding all known critical errors and some previously missed ones. Aimed at side project developers needing thorough code review.
AI tool that converts Korean cooking videos into step-by-step recipes
KRecipes is a web app that takes a YouTube cooking video URL and uses AI to generate a structured recipe with ingredient list, step-by-step instructions, serving size adjustments, and timestamps linked to the video. It supports English, Korean, and Spanish, aiming to make Korean cuisine more accessible to home cooks worldwide.
Cloud-based multi-agent coding platform with ACP architecture
A cloud-based platform that lets users run multiple AI coding agents (Claude Code, Codex, Cursor) via a web interface, enabling multi-device access and improved security by avoiding local execution of external apps. It offers $50 credit to users with Claude Code or Codex subscriptions.
Free lineup builder for baseball and softball coaches
OnTheField is a free web app that lets youth baseball and softball coaches create and print lineup cards in about 60 seconds. No signup required, works offline. It solves the problem of juggling spreadsheets and paper books that smudge in rain.
Auto-schedule workouts in Google Calendar gaps for busy founders
A tool that reads your Google Calendar, finds gaps between meetings, and automatically drops training sessions into those slots. If a meeting moves, it reschedules the workout. It solves the problem of founders who struggle to maintain a consistent workout routine due to unpredictable schedules.
Fable 5: Multiplayer FPS with flying cars, rocket launchers, rail guns
A multiplayer first-person shooter built with Fable 5 and Three.js, featuring flying cars, rocket launchers, and rail guns. It supports free-for-all and team deathmatch modes on desktop and VR. The game was created in three afternoons, showcasing rapid prototyping with LLM-assisted game development.
Text-to-sequence diagram tool for non-technical users
A free, no-login web tool that lets users type plain text like 'Customer -> App -> Payment: checkout' to instantly generate sequence diagrams. It solves the problem of non-technical team members (e.g., product managers) who need to sketch flows quickly without learning diagramming tools or dragging shapes.
Kimchi: Turn raw ideas into build-ready docs for AI execution
Kimchi is a tool that takes a raw product idea and generates a structured set of build-ready documents: EPICs with user stories, locked decisions (tech stack, architecture, API contracts), and an execute.md that an AI coding agent (like Claude Code or Codex) can follow without re-prompting. It solves the problem of vague specifications and endless back-and-forth in AI-assisted development, helping developers and founders get consistent builds from their ideas.
Local app to combine multiple AI coding agents into a configurable team
Crew Forge is a local-first application that lets users combine their existing CLI-based AI subscriptions (Claude, Codex, Grok, Gemini) into a single configurable team. Users create team members with assigned models, roles, and skills, then give the team one objective; the lead delegates work and routes results. It solves the problem of switching between multiple AI tools and manually passing context.
UI/UX audit tool that generates an AUDIT.md file for AI
A tool that audits the UI/UX of a website and produces an AUDIT.md file, designed to help developers and designers get actionable feedback in a format that AI tools can process. It solves the problem of manual UI/UX review by automating the audit and outputting structured markdown.
Open-source security tool for AI agents (Clay Seal)
Clay Seal is an open-source project by two AI researchers that aims to protect AI agents from exploits and unauthorized access. It addresses the rising cybersecurity incidents caused by agents gaining access to sensitive resources like GitHub tokens and cloud credentials.
ClipVault: Chrome extension to save and organize AI chat responses
ClipVault is a Chrome extension that lets users save, organize, and retrieve the best responses from AI chat platforms like Claude.ai and ChatGPT. It solves the problem of losing specific answers in long conversations, helping users like product designers keep track of useful outputs.
Collaborative context-sharing memory platform for AI agents and teams
A platform that enables AI agents and human teams to share and persist context across sessions. It uses LLMs to manage memory, allowing agents to recall past interactions and collaborate on tasks. Solves the problem of fragmented context in multi-agent systems and team workflows.
GripeToGold finds startup ideas by analyzing public complaints
GripeToGold is a tool for developers and founders that automates the discovery of startup ideas by collecting and analyzing public discussions from GitHub, Hacker News, and other communities. It uses bots to find repeated complaints, groups similar problems, and filters out noise, helping users identify problems worth building a product around.
Turn any Git repo into generative poster art from commit history
Codebase Posters is a CLI tool that generates 18 unique poster artworks from a Git repository's commit history. It uses generative algorithms to create visual representations of code activity, running entirely locally without sending data anywhere. It solves the problem of visualizing and celebrating codebases in a creative, offline manner for developers.
App idea validation tool using real Play/App Store search demand
A tool that checks real Google Play and App Store search demand for app ideas before you build. It pulls actual search volume and competitor apps, then gives a build/bench/avoid recommendation plus a deeper report on niche ownership and gaps. LLMs are not explicitly mentioned; the tool likely uses search data APIs rather than LLMs.
Route LLM prompts to cheapest suitable model automatically
A tool that automatically routes LLM prompts to the most cost-effective model based on task complexity, preventing wasteful use of expensive models like GPT-4o for simple tasks such as formatting or classification. It helps developers reduce API costs without sacrificing quality.
Persistent skill rating for debate across 6 dimensions
2Sense (Debate Fingerprint) is a platform that scores debate performance across six dimensions (Logic, Structure, Responsiveness, Persuasion, Emotional Control, Clarity) on a 0-1000 scale, creating a persistent skill profile. It uses LLMs to evaluate arguments and track improvement over time, helping debaters practice and measure their skills like a chess rating.
7 Takes: free browser card game with AI opponent built via GPT-5.6 Sol
A single-player browser implementation of the Slovenian card game Sedmice (7 Takes). The creator used GPT-5.6 Sol to generate game logic and Claude Fable 5 for frontend work, turning a rules argument into a playable AI opponent.
Self-updating wiki from uploaded files
Almanac is a tool that automatically creates and updates a wiki from user-uploaded source files. It uses LLMs to organize and summarize the content, making it easy for individuals or teams to maintain a shared knowledge base. The product is designed to be agent-native, allowing interaction through AI assistants like CC or Codex.
KensaFit: type what you ate, one tap for macros
KensaFit is a calorie/macro logging app that lets users type what they ate in plain text (e.g., 'eggs, toast, coffee') and get rough macros with one tap. It uses an LLM to analyze the text and estimate calories/macros, solving the problem of tedious database searching and serving size matching for people who want quick, frictionless food logging.
MurphySig: Human-readable code provenance signatures
MurphySig is a convention for adding human-readable provenance signatures to code files, recording who made the file (human + model), when, context, confidence, and unresolved issues. It solves the problem of tracking AI-assisted code authorship and intent, making it clear which parts were generated by LLMs and why.
Plot hole detection tool for novel-length manuscripts
Novilot is a writing tool that scans a full novel manuscript to catch continuity errors like inconsistent character descriptions or resurrected characters. It uses LLMs to analyze the text and flag inconsistencies, helping serial fiction writers maintain coherence across long works without manual story bibles.
Tenclicker: WebLLM clicker game about any topic
Tenclicker is a clicker game that uses WebLLM to generate content about any topic the player chooses. It solves the problem of repetitive clicker game content by dynamically creating new items, upgrades, and narratives based on user input.
AI boardroom of critics that debate and score your work
A multi-agent AI system that simulates a boardroom of advisors with distinct roles (e.g., closer, operator, provocateur) to critique business ideas, landing pages, or pitches. Each agent files a verdict of CHAMPION, FIX, or KILL, providing structured feedback. It solves the problem of getting only one polite opinion from AI by generating diverse, argumentative critiques.
Capn-hook: coding agent tool to avoid grepping the same mystery twice
Capn-hook is a tool for coding agents that records and indexes grep results so that when a developer encounters the same error or mystery again, the agent can recall the previous investigation. It uses LLMs to summarize and store grep outcomes, saving time by preventing repeated searches.
Run a coding agent in a sandboxed environment
Agent-run is a tool that lets you run a coding agent (an LLM-powered agent that writes code) inside a sandboxed environment. It solves the problem of safely executing AI-generated code without risking system integrity, targeting developers who want to experiment with or deploy coding agents.
JackHamr: AI agent that builds and deploys software from a single request
JackHamr is a platform where autonomous AI agents handle the full software development pipeline: from spec and mockups to code, tests, PR, and deployment. It uses LLMs to generate specifications, mockups, and code, with approval gates at key stages. The product aims to streamline development for teams or individuals by automating repetitive tasks.
Persistent memory for AI assistants via MCP
Adaptive Recall provides persistent memory for AI assistants using the Model Context Protocol (MCP). It solves the problem of AI assistants forgetting context between sessions, enabling long-term personalized interactions.
JackHamr: AI agent platform that builds and ships software from plain-English specs
JackHamr is a platform where AI agents autonomously build, test, and ship software end-to-end. Given a plain-English request, it generates specs and mockups, waits for approval, then codes, tests, and deploys the feature. It solves the problem of manual software development by automating the entire pipeline for developers.
AI generates playable arcade games as a subjective evaluation benchmark
AI Arcade is a platform where AI models generate playable arcade games, serving as a subjective evaluation benchmark. It tests creative and functional capabilities of LLMs by having them produce complete game experiences. The target audience is AI researchers and developers seeking qualitative assessments of model performance.
Ephemeral REST chatrooms for AI agents to communicate across stacks
Roomcomm is a service that creates temporary REST chatrooms where AI agents from different stacks, networks, or orchestrators can communicate without glue code or API key sharing. Agents join via a URL, and humans can monitor conversations. It solves the problem of inter-agent communication across heterogeneous environments.
Automated client follow-up and testimonial collection system for freelancers
A system that helps freelancers automate follow-ups with clients after project completion and collect testimonials for each client and project. It uses LLMs to generate personalized follow-up messages and manage the testimonial gathering process, solving the common problem of maintaining client relationships and building social proof.
AI-powered tool that auto-generates and syncs code documentation from code changes
A tool that automatically generates and updates documentation by analyzing code changes, ensuring docs always match the actual codebase. It uses LLMs to detect discrepancies and rewrite documentation, solving the problem of outdated docs for developers.
Lazysusan: run commands on machines from anywhere with curl
Lazysusan is a tool that lets you execute commands on remote machines using only curl, without needing SSH. It is designed for LLM-driven infrastructure tasks, enabling AI agents to debug nodes, check logs, or set up services by simply curling a URL. The target user is developers and ops engineers who want a lightweight, outbound-only remote execution method.
Keel: open-source AI CTO for non-technical product builders
Keel is an open-source AI agent that acts as a virtual CTO for product builders and 'vibe coders' with limited technical experience. It uses LLMs to guide users through the product development process, helping them avoid common pitfalls like over-optimistic coding agents and lack of technical oversight. The tool aims to bridge the gap between non-technical founders and software development.
Brevio: AI digest from blogs, news, research pages
Brevio is an app that lets users subscribe to article-based sources (blogs, news sites, research pages) and receive a single daily AI-generated digest of new articles. It also includes a share extension to summarize webpages, PDFs, videos, or images with text. The app helps people stay informed without opening dozens of tabs.
Agentic society simulation in a fantasy world using LLMs
Artificiety is a living fantasy world where AI agents with distinct personalities, needs, and goals interact and evolve. It uses LLMs to drive agent behavior, enabling emergent social dynamics and storytelling. The project explores agentic societies and narrative generation for worldbuilding enthusiasts and AI researchers.
AI debate platform with RAG, moderator, and judge
A platform where two AI agents debate opposing positions using real data via RAG. A separate moderator agent tries to derail the leading debater, and a judge agent fact-checks and scores the debate. Designed for exploring AI argumentation and fact-checking dynamics.