Open source
Everything here is public, licensed and versioned. The libraries install with one command; the references are free to read and to reuse.
Libraries
Small, single-purpose tools, published on PyPI and installable today.
politeclient
When a program of yours hits someone else's API in a loop, the usual outcome is getting rate-limited or banned for asking too fast, or falling over on the first slow response. politeclient waits when it should, retries what is worth retrying, and doesn't ask twice for what it already has.
It does not speed anything up — it deliberately goes slower so the other server doesn't shut the door on you. And it doesn't replace requests: it builds on it.
pip install politeclientwebhook-replay
A webhook arrives once and is gone. If your code fails while handling it, you can't ask Stripe or GitHub to send it again so you can debug. This captures it once and lets you re-fire it at your local app as many times as you need.
It is not a tunnel and not a cloud service: it runs on your machine and sends nothing anywhere. It doesn't verify signatures for you — it keeps them intact so your code can.
pip install webhook-replayscaffld
Starting a Python project properly costs half a morning — tests, formatting, CI, licence, structure. This leaves it set up and working in seconds, from a terminal assistant.
It is not a framework and ties you to nothing. It writes ordinary files and gets out of the way; afterwards the project is yours and doesn't depend on scaffld at all.
pip install scaffldframesig
Find specific moments inside a video — when a marker appears on screen, when an indicator changes — without training any model. You describe what to look for in a config file and it finds it by pixel signature.
It doesn't recognise objects or understand the scene — this is not AI vision. It looks for colour patterns in regions you define, which is why it's fast, reproducible, and needs no GPU and no training data.
pip install framesigReferences
Open documents on method, written to be checked rather than believed.
The GEO Handbook
The open reference on Generative Engine Optimization: the method, with a primary source behind every claim and its limitations stated.
It is a reference, not a service: it explains the method and states its limits, it doesn't do the work for you.
The GEO Cookbook
Six reproducible experiments, measured offline with a deterministic script — and, for each one, what the number does not prove. Published as a citable dataset.
The six measurements are offline and deterministic: they measure machine-readability, NOT whether an engine cites you more.
prompt-engineering-evidence
A prompt-engineering reference where every technique is graded: solid, mixed, or folklore. The point is separating what holds up from what gets repeated.
It grades techniques by the evidence behind them; it is not a list of tricks that promises results.
If any of these is useful to you, take it — that is what the licence is for. And if you find something wrong, the issue tracker is open.