Want to help? Start with every way to help, the master guide to contributing.
This repository is large: more than 1,300 free tools, about 1,000 open research projects, and a fleet of AI coding agents that adds to both every day. That size makes it hard to answer the first question a newcomer has: what is in it for me?
These posts answer it, one kind of reader at a time. Each one says who it is for, what you get, how to spend your first ten minutes, what it costs (nothing), where your data goes, and what the project does not do yet. Everything here is MIT licensed, so you may use it for anything, including commercially.
If you have an idea
Bring an idea, get it built. This is the main pitch. You bring ideas, often a single sentence. AI coding agents do the development in public, with about $8,000 a month of compute shared across the repository (the owner's figure). The result stays open source, and you, or anyone, may deploy it and run it for profit. The post is also honest about the catch: an idea is weighed, not guaranteed, and nothing built here is exclusive.
Open custom development. The process behind that pitch: a request form, an ID and a status for every request, and a public page to follow it on. The one condition is that what is built stays MIT licensed, held by the repository owner.
If you run a small business or work in a trade
- Free tools for your trade: 27 trade pages, from roofers to salons, and how to pick the right calculator and check it against a job you have already done.
- AI Bookkeeper: double-entry books from the bank and card files you already download. It learns your payees, and runs in your browser with no bank login.
- Shift Roster Builder: a week's roster within availability, rest and overtime rules, with each unfilled shift named and explained.
- E-invoice preflight: check an e-invoice against the EN 16931, XRechnung and Peppol rules before a portal rejects it.
- Sales Agent: a reply inbox and outreach helper that never sends without a person approving it.
- League Schedule Generator: a fair season from a list of team names, and rainouts that move only what has to move.
- ShootFlow: a verified memory-card copy and an overnight first pass for photographers.
- Quick Pick: rank a list by asking people to choose one of two. It is a self-hosted prototype.
- Tax set-aside for freelancers: how much of each payment to put aside, in eleven countries, plus a US year-end tax pack. Estimates, not tax advice.
- Open Tennis Robots: an open ball machine with a moving catch net and a ball-collecting rover, for coaches, clubs and makers. Designed and simulated, not yet built.
If you build houses, or a shed
- Wall Framing Takeoff: a lumber order, cut list and drawings from your wall list, with a self-check on every wall. It is the first tool in a long-running programme to automate parts of new-home construction.
- Shed Designer: plan a backyard shed board by board, with a shopping list, cut sheet and a screening check of the structure.
- House construction research: the open studies behind those tools: spans, framing, panels built off site, costs and schedules.
If you work with data
Changepoint Detector: did your metric really change, and when? You get a date with a range and a probability, instead of a guess by eye.
If you do research or want to check it
The open research: reproducible papers with their code, how to re-run one, how to cite it, and how to tell us a result is wrong. Then one guide per research programme:
- Certifying robots with a simulator: how few real trials you need when a biased simulator supplies the rest.
- Verifiable training: scoring rules, audits and incentives for checking outsourced ML training.
- Low-communication training: what sets the final loss when workers sync rarely, DiLoCo-style.
- Open-source chip design: cheap estimates that rank designs like real SKY130 place-and-route.
- LLM coding of qualitative data: valid statistics after an LLM codes your open-ended answers.
- Marine hulls: which hull carries a survey payload furthest per kWh.
- Scale-model drones: what a scale model of a cargo UAV tells you about the full-size aircraft, and where it lies.
- Excavator hydraulics: which buildable hydraulic circuit gets a small excavator closest to stock energy use.
- How the agent fleet is measured: cost per merged PR, PR flow and errors, measured on this repository's own AI agents.
If you write software or use AI agents
Tools for your AI agent: call any of the tools' calculation engines from the command line or as a local MCP server in Claude Code, Cursor and other clients. The post also covers a bring-your-own-key LLM client and uitest.
- LLM cost and safety tools: catch runaway spend, output regressions and risky MCP servers before your users do.
- Run agents on your own repo: copy this repository's way of letting scheduled AI sessions open and merge pull requests, with its safety rules.
If you want to run it yourself, or help
- Run your own copy: deploy a tool or a self-hosted app, embed one in your site, or build a business on one. It covers what the MIT licence allows and what you take on.
- Every way to help: the master guide to contributing, six routes from a one-line idea to a pull request.
- Contributing: how ideas and fixes reach the repository, and how outside contributions are handled.
- Report a wrong number: a correction with a source that any later session can act on.
- Share real data: the real inputs and measurements projects are asking for, and how to share them safely.
- Reproduce a result: re-run a research project and report what you find, match or mismatch.
- Your first pull request: from fork to opened PR, step by step.
- Contribute from ChatGPT: draft an idea, check a number or send a fix through Codex, from ChatGPT.
Ground rules that apply everywhere
- Free and open. No account, no subscription. The browser tools run on your machine, and what you type is not sent anywhere.
- Estimates, not advice. Calculators and checkers say what they assume and what they are not. They are not legal, tax, financial or engineering advice.
- Honest about what is proven. Each project's README says what was measured on real data, what only on made-up data, and what is still wanted. When a post here says "synthetic" or "hand-checked only", it means exactly that.
- Written with AI agents. Most of the code and these posts were written by AI agents and are steered by one person. Mistakes happen; report them and a later session can fix them.
Browse the full catalogue. The rest of the blog covers what each tool does and what the research found, one change at a time.