AI-built projects, archived with build logs

Records filed: 148
Builders: 144
Tools indexed: 36
Newest project: Oct 2026

Recently talked about

2026-10-10

-1,005 pts since the last scan · 1 new to the list

Ranked by points on Hacker News stories in the last 90 days. A story counts only when its link points at the project’s own repository or product site — matched by link, never by name.

1,665ptsCodex CLI14 stories · 719 comments · 2026-07-28“Codex Security” ↗
1,500ptsCursor32 stories · 873 comments · 2026-08-17“Cursor launches Origin, GitHub alternative” ↗
751ptsDeepSeek Harness3 stories · 315 comments · 2026-08-13“DeepSeek Harness developer preview” ↗
682ptsQM1 story · 163 comments · 2026-07-31“qm – Multiplayer agent harness for work” ↗
590ptsGrok Build1 story · 643 comments · 2026-07-15“Grok Build is open source” ↗
578ptsJeff2 stories · 231 comments · 2026-09-28“Jeff – Jev-compatible 0.8B decision models, trained at home, ~30 ms” ↗
570ptsMoonshine Voice1 story · 92 comments · 2026-07-14“Speech Recognition and TTS in less than 500kb” ↗
546ptsi-have-adhd3 stories · 371 comments · 2026-09-08“I-have-ADHD: A skill to stop coding agents from burying the answer” ↗

What it is, in one line

All 148 →

Every record in the archive, cut down to the one line that says what it is. 864 build-log stages are written up behind them. Click any of them.

a game inside a game models that answer in probabilities Fabric, Forge and NeoForge in one game LLM usage pinned to the screen edge the deterministic judge for AI claims ChatGPT plans, Codex runs an agent OS built to compound plain-English rewrites for Claude Code hiring Gemini as a fast worker a commit gate for AI coding agents a team of agents builds your game acceptance gates for AI agent work let an agent drive your phone a multiplayer agent harness for work the runtime layer for LLM agents cited design, builds, and QA runs a real timeline the agent edits the terminal that outlives its window a 41-skill PR pipeline proof layer for AI coding agents write C, boot a System 7 Mac an always-on assistant on Claude Code an ESP32-S3 macropad a desktop pixel crab for agents script to animated episode agents as a studio a node-graph studio for AI short dramas an open-source Rust coding agent governance for AI coding agents one AI specialist per file one menu for eight coding agents a p5.js particle sketch in Docker low-latency speech to text two teenagers, one calorie tracker role-playing agents that work as a crew one prompt becomes a codebase ChatGPT for your own documents

an open-source Photoshop in Rust OCR, classify, click agents that share one memory Markdown memory for agents select first, then get an office file de-AI writing that repairs structure local tools for ChatGPT a 3D-printed AI agent robot all coding agent sessions in one window a workspace that remembers and cites a virtual machine for every bot sixteen harnesses, one API a shared workspace for coding agents record a task, get a skill an agent skill for plain prose one shared memory for every AI agent SpaceXAI’s full-screen coding TUI FSRS recall inside coding agents one skill per coding agent CLI 21 agent skills for solo operators turn a PDF into an agent skill distil a thinker into a skill a person’s messages become a profile memory shared across coding agents a terminal orchestrator for AI agent teams a solo-built AI companion lightweight AST code search a code knowledge graph for AI agents a Diablo II homage in plain JavaScript 32 guards for agents a vibe-coded Game Boy emulator in Rust a local AI agent as app, CLI and API prompt to a running app in the browser the interface was the bottleneck a self-hosted chat interface AI pair programming in your terminal node-graph control for Stable Diffusion

a 0.8B model that decides first a 7×24 loop that edits your files topic in, narrated film out eight roles, sixty routines a social agent that posts and learns agent memory as a deductive database the narration sets the clock a 3D fruit fly on your desktop check, fix, check again everything is a plugin Postgres diagnostics in one Go binary the codebase map agents read first a governance standard for LLM corpora a call graph for coding agents one control plane for coding agents eleven skills, one short drama replay coding-agent sessions in 3D a code graph your agent reads as files a shared workspace for people and agents a five-step loop for coding agents speak a genre and the music changes a Rust workbench for agents the context layer for coding agents a crab in your MacBook notch let an agent run ML experiments a self-hosted traffic dashboard AI-built projects in one repo Linus Torvalds writes guitar effects agents simulate what happens next a browser Minecraft skin editor MCP server that turns code into music a prompt-to-app platform sold to Wix an open-source coding agent for VS Code the open-source coding assistant run open-weight models on your machine a visual workspace for agents and RAG one person’s AI photo business

typed decisions from a single encoder pass a coding agent shell on DeepSeek Harness a wiki an agent keeps markdown truth plus a sleep layer agents organized as a company plain-language agent skills an agent skill for scroll sites a canvas agents draw their work on an AI coworker with its own computer 122 skills, one test bar plain-English rewrites six coding agents on your wrist make your agent talk human the ledger for coding-agent work shot cards for product films ten coding agents in one TUI a loop with its own eval frontier MoE models in pure C a laziness ladder for 20 coding agents stop burying the answer an MCP server for reverse engineering universal AI context generator a job pipeline that never applies 23 roles and 8 tools for Claude Code a virtual iPhone on your Mac an open-source AI calorie tracker edit JianYing with Python compressing what an agent reads a Prolog interpreter built by AI Google’s terminal agent, Apache-2.0 OpenAI’s Rust coding agent an AI agent that drives a browser AI agents that debate public opinion an open-source coding agent a local terminal coding agent the AI-native code editor Whisper speech recognition in C/C++

What is in a record

Six fields

01

The stack

Which editor, which model, which host. Named, not categorised — “Cursor plus Claude” is a different story from “Copilot”.

02

The build log

The stages in order, with the prompts that mattered. Copied verbatim where the builder kept them.

03

What it cost

Money and time. The two numbers that decide whether a reader attempts the same thing.

04

Where it broke

The dead ends. More useful than the success, and the part nobody publishes.

05

What they would redo

The builder’s own hindsight, recorded while it is still uncomfortable.

06

Proof it exists

A live URL or a public repo. No screenshots of things you cannot open.

Stack index

All 36 tools →

Claude112 Claude Code93 Codex68 Cursor40 GPT40 Gemini29 GitHub Copilot21 Google Antigravity15 DeepSeek12 Ollama11 DeepSeek Harness11 Cline10 GitHub Pages10 Windsurf6

By subject

Open Source133 Agent104 Developer Tools100 Runs Locally93 Solo35 Desktop24 Framework22 Research21

Open call

Built something with AI? The build log is worth more than the launch post.

Send a live URL or a public repo. We research the build story from what is already public — the repository, the commit history, the launch posts — then write the record and publish it with the builder credited. If you built it, say so and we will approach you before anything goes out. You keep every right to your own work.

Submit a record Read the criteria

Rejected on sight

A screenshot you cannot click through to
A thin wrapper around one chat API
An aggregator of other people’s projects
Anything the submitter does not own

We would rather turn a submission down than lower the bar. That is the point.

What they built.How they built it.In their words.

Recent records