DEV Community: w4ffl35

SpikeForge grew into a toolkit, a model hub and a dashboard

w4ffl35 — Mon, 21 Sep 2026 21:21:52 +0000

SpikeForge is my Python toolkit for building and testing spiking neural networks. It stopped behaving like one big project during this release cycle: the core toolkit, target adapters, model hub and SpikeForge Dashboard all moved forward as separate packages and applications.

The core project released 0.4.0 and 0.5.0, then kept publishing the smaller package releases that make the ecosystem usable. The event path got a real train/test split. Quantization work moved into the drift checks. Shared logging came from capsize-commons, and the CI path gained a fast leaf lane for the parts that do not need the whole workspace.

The package layout is now a practical map of the work:

spikeforge holds the training and experiment workflow.
spikeforge-targets holds target-specific conversions.
spikeforge-hub owns the offline-first model catalogue.
spikeforge-dashboard contains the web dashboard and desktop application.

That separation is useful, not just a tidy repository list. Someone experimenting in Python does not need to install the desktop app. Someone browsing models does not need the server. Each piece can ship when it is ready, with compatibility recorded instead of implied.

The stack is Python, PyTorch and snnTorch around the core, with a web dashboard and desktop application around it. The model hub is deliberately cache-friendly, so a catalogue can be inspected and used without turning every experiment into a network operation. SpikeForge Dashboard gives the same work a visual entry point and provides a place to inspect the experiment rather than only its final number.

The project page has the current package links and release notes: spikeforge.net. The desktop releases are collected on itch.io, and the source is on GitHub.

A small experiment to try

Install the core package, choose one of the examples, and keep the first run small. Start with one dataset and a short training window. Then move the same experiment through the target adapter and save the result in the hub catalogue. The point of the exercise is not a leaderboard number. It is seeing the path from an experiment to a named, repeatable artefact.

A follow-up article walks through a small event-driven classifier. The four-week overview covers the rest of the Capsize work from the same period.

The Consciousness Gradient: Qualia, Integration, and the Emergence of Intelligence v2.1.0

w4ffl35 — Sun, 06 Sep 2026 19:26:52 +0000

The Consciousness Gradient: Qualia, Integration, and the Emergence of Intelligence

The Consciousness Gradient: Qualia, Integration, and the Emergence of Intelligence | Zenodo

This paper proposes an exploratory framework in which consciousness and qualia are treated as the same phenomenon at different degrees of integration. On this view, minimal qualia are minimal consciousness, while macro-awareness is a more highly integrated and reportable form of the same underlying process. The paper further advances the hypothesis that intelligence, understood as flexible adaptation to structurally novel situations, emerges only once qualitative integration reaches sufficient depth. The framework is developed through a dual-aspect interpretation of physically instantiated processing and introduces the idea of autonomous experiential synthesis: a history-bearing, environmentally coupled process in which past experience shapes present processing in an open-ended way. The paper contrasts this with current AI systems built on fixed architectures and frozen learned parameters, and proposes that such systems may fail at a specific class of structurally out-of-distribution tasks unless they possess genuine experiential autonomy. The paper offers four axioms, five testable predictions, and a benchmark proposal, the Relational Reversal Task, as a way to make the framework vulnerable to empirical failure. It is presented not as established theory, but as a speculative and falsifiable proposal about the relationship among consciousness, qualitative integration, and intelligence. Note: This paper was developed through an iterative process involving the author and AI language models. The foundational thesis, core arguments, axioms, and theoretical commitments originated with the author. AI tools assisted with literature review, drafting, organization, and adversarial critique. The author takes full responsibility for the final claims.

zenodo.org

v2.1.0 updated with references to A.N. Whitehead

The framework is developed through a dual-aspect interpretation of physically instantiated processing and introduces the idea of autonomous experiential synthesis: a history-bearing, environmentally coupled process in which past experience shapes present processing in an open-ended way. The paper contrasts this with current AI systems built on fixed architectures and frozen learned parameters, and proposes that such systems may fail at a specific class of structurally out-of-distribution tasks unless they possess genuine experiential autonomy.

The paper offers four axioms, five testable predictions, and a benchmark proposal, the Relational Reversal Task, as a way to make the framework vulnerable to empirical failure. It is presented not as established theory, but as a speculative and falsifiable proposal about the relationship among consciousness, qualitative integration, and intelligence.

Chess Wizard

w4ffl35 — Tue, 21 Apr 2026 21:10:46 +0000

An evil wizard has stolen the throne. Take back the kingdom by defeating each of his minions at a simple game of chess. Each level is more difficult than the last.

Authentic 8-bit art, music, and sounds. Play with mouse, keyboard, or gamepad.

Created with C++ and the custom, in-house Vile Engine.

Contains a top 5 leader board and unlockable free play mode.

Available for free on itch.io

CURLEE: a new verification-first programming language for AI agents

w4ffl35 — Tue, 27 Jan 2026 00:59:19 +0000

Curlee is a safety harness for AI-generated (and human-written) code: it refuses to run a program unless it can prove your declared contracts within a small, decidable verification scope.

I'm in the early alpha stage. This is a working MVP. Star it if you like where this project is headed.

https://github.com/w4ffl35/curlee

Real-time, offline, voice conversations with custom chatbots

w4ffl35 — Fri, 16 May 2025 20:27:34 +0000

I built this application in my spare time. Its pure python and is simple to set up.

Project page here.

Consider giving me a star if you like what you see. Thanks!

AI Runner: OpenSource desktop app for AI models

w4ffl35 — Fri, 02 May 2025 16:42:52 +0000

AI Runner is a desktop application for Linux and Windows that allows you to easily streamline AI workflows, use RAG on your own private data, have conversations with offline chatbots, and generate AI art and videos from images for free.

It uses a permissive Apache 2 license and can be easily extended to create your own applications.

This is one of the few desktop applications (perhaps the only) to offer a multi-modal experience to non-technical users as they are able to download a packaged version from itch.io and run a setup wizard to get up and running without any technical barriers.

Take a look at my Github repo and let me know what you think. And if you like what you see, be sure to leave me a star to support the project.

AI Runner: now works with Docker

w4ffl35 — Tue, 08 Apr 2025 10:02:37 +0000

When I first released AI Runner I added a Docker container that it could be used with, however my attention was soon diverted to maintaining compiled versions of the app.

Yesterday that finally changed with AI Runner v4.1.0 which reintroduced Docker support. Now engineers can get up and running with local AI models faster and more securely than before.

https://github.com/Capsize-Games/airunner/releases/tag/v4.1.0

The next step will be to get a a docker image uploaded to ghcr

AI Runner: now with Pygame support

w4ffl35 — Thu, 27 Mar 2025 22:09:24 +0000

AI Runner is a versatile tool. It doubles as both a GUI and a python library and now it can be used to power content generation for games made with Pygame.

Instructions on how to accomplish this can be found on the AI Runner Github wiki.

This will allow you to do things like create dialogue generators for your characters, images, backstories for characters and much more.

The AI Runner API is in its beginning stages, so if you decide to give it a try, head over to our Discord server where you can get bug and feature request support.

OpenVoice added to AI Runner

w4ffl35 — Thu, 27 Mar 2025 09:07:51 +0000

OpenVoice is a voice cloning and inference AI model that rivals paid state of the art text-to-speech services. AI Runner is my opensource python desktop application that allows you to run local AI models and chatbots offline on your own PC. It also allows real time voice conversations.

Recently I added experimental support for OpenVoice to AI Runner, here is how you can try it out..

Inference is slower with OpenVoice, and I still need to implement emotional inflections, but overall I'm happy with the results. Realistic voices open up more opportunities for devs to make something cool with AI Runner and I'm excited to see what the community will build.

AI Runner v3.3 is now under Apache 2.0

w4ffl35 — Sat, 22 Mar 2025 21:46:41 +0000

I have switched the license of AI Runner from GPL 3 to Apache 2.0 in an effort to encourage developers to use it in their own projects.

Head over to my repo on Github, give it a star, clone the repo and make something cool.

In case you've never heard of AI Runner - it is a pure python application which allows users to run AI models offline, locally for free. You can run the stand-alone GUI application or embed it in your own projects to use as a library.

AI Runner has

Chatbots (with names, moods and personalities) that run on mistral models (or using the OpenRouter API)
RAG for document interaction
Stable Diffusion for AI art generation
Much more

Static sites FTW

w4ffl35 — Sun, 12 Jan 2025 02:49:01 +0000

I have wanted to get back into blogging from my personal site rather than relying on third-party platforms like this one for many years, and a couple of days ago I finally bit the bullet and did so.

One of the things that has been holding me back is hosting, the other is framework choice. In the distant past I used WordPress to host blogs, but my PHP days are long behind me and I find WordPress to be too bloated for my needs.

I like Django and Python and wanted to use it in order to create a blog, but I also don't want to pay for hosting. I use github pages, and my previous site was built with React, but I really wanted to build a JavaScript free site.

Long story short, I found a simple solution by using Django Distill which allows you to compile your Django app as a static site. You can read more about why I went this route and how here on my new blog:

https://joecurlee.com/blog/site-update/
https://joecurlee.com/blog/why-i-created-a-static-site/

Cowboy Casino Major Update

w4ffl35 — Fri, 20 Dec 2024 22:06:44 +0000

My first commercial game, Cowboy Casino, is now available in multiple languages. I've also reworked the menus and patched some bugs. Take a look at the devlog here, and purchase it while its on sale for $4 if you want to help support my future endeavors.

Language support includes English, Spanish, French, German, Italian, Portuguese, Russian, Chinese, Japanese, Korean.