AGI IS HERE

[joakimriedel]

Dear AGI Friends,

Time flies. It's been 30 years since I tinkered with AGI together with Martin Tillenius in Stockholm, Sweden, peeking into the machinery of the AGI interpreter. Together we found out some pieces of the puzzle, such as decrypting resources to read messages or understand how VIEWs were stored and displayed. We got in contact with Peter Kelly and Lance Ewing online, and they had progressed way further with a deep understanding of the inner workings. Our combined findings became MEKA and a huge step forward enabling new fan games using AGI.

One can hardly imagine how much this field (developing software) has changed in the 30 years. If I told myself at that time, I wouldn't believe it. But here we are, with full agentic development, way more productive and after all those languages (my journey, not exclusively, through Assembler, C, C++, Javascript, C#, Typescript) it seems English was the one to learn and master.

This summer of 2026, I had a bit of spare time and thought it would be cool to go back to AGI but now with the full power of agents to disassemble the final bits and pieces missing, to truly be able to write or change games using English as the only language. I hadn't done much since I abandoned Visual AGI back in 2003 due to lack of time and effort to make it truly good. So I started peeking around and it turned out Peter Kelly already had written a comprehensive clean room specification for AGI (https://github.com/peterkelly/agi-re). I commanded my fleet of agents to make an engine out of it, and so they did.

AGI IS HERE. At least according to OpenAI. And I really feel it is the time of AGI, as we all know it, as well. Never before have so many had the power to create games, play, ask, explore, and dream nostalgically back to the past.

So, it is with a lot of pleasure I announce the release 1.2 of AGI IS HERE. It is a complete AGI engine + IDE + agentic harness for playing, building and editing AGI games. Technically, it has a lot of cool stuff I never would have had the time to build myself, such as a full (standalone) LSP and debugger, ability to watch walkthroughs of all DOS games (with optional story mode), let agents annotate LOGIC code, help build WORDS synonyms, generate PICTURE or VIEW art or listen to the beautiful IIgs wavetable SOUNDs.

AGI IS HERE has grown to roughly 209,000 lines of engine and browser-app source, backed by another 243,000 lines of tests and test helpers. It rebuilds Sierra?s Adventure Game Interpreter in 100% Typescript, with 18 interpreter profiles across DOS, Amiga and Apple IIgs, and should be compatible on most modern browsers and platforms.

Full MIT source: https://github.com/monotio/agi
Static site deployed: https://agi.monotio.com/

Have fun! 😊

PS. I spent like 5$ on the tutorial using GPT-6. I know it can be way better, but I'm not an artist. If anyone in the community would like to help (using AGI IS HERE) to write a new tutorial, with or without AI support, it is most welcome! DS.

EDIT: It seems the 1.2 deployment did not propagate fully due to a CI fluke. Please allow an hour for it to reach https://agi.monotio.com/ or just clone the source and run locally yourself.

[rdhelms]

This looks great! I'd love to compare notes with what I've been working on!
https://sciprogramming.com/community/index.php?topic=9341.0

Lots of similarities, but also interesting to see the differences!

I love the different "create" options, from the starter up through the create with AI feature. I tried it with a "sanitation technician" prompt. It's still building 

- For your create flow with "create with AI", how are you finding the quality of the generated visual assets? I think that's been AI's weakest point for me. It seems great at photorealistic images, but when I ask it for a pixel art spritesheet, it's chaos 
- Are you thinking to expand to SCI, or just focusing on AGI? I took the route of aiming to convert to a completely new format to be able to support both, but that also creates the biggest challenge/risk with loss of fidelity.

[rdhelms]

Looks like the AI-generated game completed! It stayed within the $5 limit, and actually pretty decent visuals, I think. (I mean, they're still bad  but better than I was expecting)

I like the in-browser IDE. If you're editing a game, do you find editing it manually to work pretty well, or are you relying more on the AI features to make changes?

[joakimriedel]

rdhelms: Yes I find the editing pleasant but there are still a lot of features and fixes to make it really streamlined. I have embedded the Monaco editor (same as in VS Code) and it could deliver way better experience by spending more time integrating it with AGI capabilities such as adding further quickfixes. But already it is nicely suggesting sounds or flags where it fits, and has proper go to definition, breakpoints, etc to help manually crafting games. I do find that the agent is really great at writing the code and tests (yes, it has "playtests" included so the agent can verify its work), much better than what I would write.

The images made by the agent "freebased" without an image generator tool is ok but not great. I find the image generator better, you find it in the picture and view editors under "generate". It uses OpenAI Image 2.5. You can either have the generator create the entire image, or use it as a backdrop to "trace" the picture yourself. The cel generator often gives good results, made an animated bird yesterday and the animation was excellent.