|
18038
|
Untitled by @urcielg | Stardance
|
urcielg
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: Untitled by @urcielg | Stardance
|
1793253016.4720876
|
1793857816.4720876
|
|
32852
|
Untitled project by @urchin | Stardance
|
urchin
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: Untitled project by @urchin | Stardance
|
1793149017.624695
|
1793753817.624695
|
|
68115
|
My awesome macropad! by @urbin | Stardance
|
urbin
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: My awesome macropad! by @urbin | Stardance
|
1793116482.7202199
|
1793721282.7202199
|
|
24821
|
MacroPad by @urayd | Stardance
|
urayd
|
0
|
15.66
|
0
|
[]
|
[{"post_id": 20328, "ts": "202 [{"post_id": 20328, "ts": "2026-07-01T22:03:36Z", "hours": 2.8533, "text": "Final upate on the Hardware Im gonna submit now. Happy with the final result."}, {"post_id": 16601, "ts": "2026-06-27T00:29:51Z", "hours": 4.5519, "text": "EXCITINGLY my case is basically done!!!! I spent most of my time today working on finishing and polishing my case. I’m basically done, probably with the exception of like 2 changes to screws. I also am officially done with the PCB as of now; I got some cool graphics on there and rearranged some components to look cleaner. Now I’ll start working on firmware, which shouldn’t be too much of a hassle, but we’ll see."}, {"post_id": 15928, "ts": "2026-06-25T21:31:53Z", "hours": 3.4125, "text": "WOOHOO . Worked on the casing today as well as fixing some dimension errors with the board. Overall the case is starting to look spiffy!!! I plan to next work on cleaning up my pcb because the led matrix is lowkey atrocious right now, and also finishing the 3d model. I think I’m mostly ok with how the case is so I’m probably going to leave it like that. I also should start thinking of firmware. yikes"}, {"post_id": 15414, "ts": "2026-06-25T01:46:56Z", "hours": 1.9917, "text": "Finished routing my MacroPad. :) Anyways, its pretty messy for sure and also not my best work, but it is my second work so im okay wit it. Next steps im probably going to take (tomorrow): add some silkscreen art, cad a case in onshape, probably reform the board to give it a cooler shape, and figure out firm ware (yikes.) goodnight Feedback wanted :)"}, {"post_id": 15317, "ts": "2026-06-24T21:36:58Z", "hours": 2.3711, "text": "Finished the layout and schematic. I decided to settle on a 3x3 matrix with one buttonless encoder. (also 3x3 matrix of status leds which I will figure out later regarding implementation.) Jolly good time so far"}, {"post_id": 15201, "ts": "2026-06-24T18:22:18Z", "hours": 0.4814, "text": "Just started figuring this out. Currently started the schematic, and want to plan what I want from this project before I continue :) (yes the symbol for the screen is very disgusting (I rushed it if you couldn’t tell lol))"}]...
|
Title: MacroPad by @urayd | Stardance
Devlogs:
Ju Title: MacroPad by @urayd | Stardance
Devlogs:
Just started figuring this out. Currently started the schematic, and want to plan what I want from this project before I continue :) (yes the symbol for the screen is very disgusting (I rushed it if you couldn’t tell lol))
Finished the layout and schematic. I decided to settle on a 3x3 matrix with one buttonless encoder. (also 3x3 matrix of status leds which I will figure out later regarding implementation.) Jolly good time so far
Finished routing my MacroPad. :) Anyways, its pretty messy for sure and also not my best work, but it is my second work so im okay wit it. Next steps im probably going to take (tomorrow): add some silkscreen art, cad a case in onshape, probably reform the board to give it a cooler shape, and figure out firm ware (yikes.) goodnight Feedback wanted :)
WOOHOO . Worked on the casing today as well as fixing some dimension errors with the board. Overall the case is starting to look spiffy!!! I plan to next work on cleaning up my pcb because the led matrix is lowkey atrocious right now, and also finishing the 3d model. I think I’m mostly ok with how the case is so I’m probably going to leave it like that. I also should start thinking of firmware. yikes
EXCITINGLY my case is basically done!!!! I spent most of my time today working on finishing and polishing my case. I’m basically done, probably with the exception of like 2 changes to screws. I also am officially done with the PCB as of now; I got some cool graphics on there and rearranged some components to look cleaner. Now I’ll start working on firmware, which shouldn’t be too much of a hassle, but we’ll see.
Final upate on the Hardware Im gonna submit now. Happy with the final result....
|
1793491019.0503561
|
1794095819.0503561
|
|
30422
|
MicroMicro by @urayd | Stardance
|
urayd
|
0
|
2.36
|
0
|
[]
|
[{"post_id": 24999, "ts": "202 [{"post_id": 24999, "ts": "2026-07-08T22:03:21Z", "hours": 1.5842, "text": "More research today. I’ve decided what type of pre-amp I will use for the microphone."}, {"post_id": 21785, "ts": "2026-07-03T21:57:30Z", "hours": 0.7783, "text": "Research and initial stuff Most of the things I did today were very simple initial planning steps. (Figuring how I can make this project work and sanity checking myself.)"}]...
|
Title: MicroMicro by @urayd | Stardance
Devlogs:
Title: MicroMicro by @urayd | Stardance
Devlogs:
Research and initial stuff Most of the things I did today were very simple initial planning steps. (Figuring how I can make this project work and sanity checking myself.)
More research today. I’ve decided what type of pre-amp I will use for the microphone....
|
1793314216.8440259
|
1793919016.8440259
|
|
72132
|
WarioWare Game by @uraszxwqc | Stardance
|
uraszxwqc
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: WarioWare Game by @uraszxwqc | Stardance
|
1793396848.2859216
|
1794001648.2859216
|
|
9825
|
Untitled project by @ural | Stardance
|
ural
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: Untitled project by @ural | Stardance
|
1793483705.151192
|
1794088505.151192
|
|
35115
|
AI caption Generator by @urDivyansh | Stardance
|
urDivyansh
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: AI caption Generator by @urDivyansh | Stard Title: AI caption Generator by @urDivyansh | Stardance...
|
1793217170.7661037
|
1793821970.7661037
|
|
18710
|
My awesome macropad! by @upsanathapa | Stardance
|
upsanathapa
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: My awesome macropad! by @upsanathapa | Star Title: My awesome macropad! by @upsanathapa | Stardance...
|
1792781301.3978941
|
1793386101.3978941
|
|
21001
|
My awesome macropad! by @uppalapati265 | Stardance
|
uppalapati265
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: My awesome macropad! by @uppalapati265 | St Title: My awesome macropad! by @uppalapati265 | Stardance...
|
1793289279.9194574
|
1793894079.9194574
|
|
42898
|
My awesome macropad! by @uponadecember | Stardance
|
uponadecember
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: My awesome macropad! by @uponadecember | St Title: My awesome macropad! by @uponadecember | Stardance...
|
1793192118.6165357
|
1793796918.6165357
|
|
42900
|
Once Upon an October by @uponadecember | Stardance
|
uponadecember
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: Once Upon an October by @uponadecember | St Title: Once Upon an October by @uponadecember | Stardance...
|
1793195414.6684844
|
1793800214.6684844
|
|
56838
|
My own site! by @upendrauddagiri | Stardance
|
upendrauddagiri
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: My own site! by @upendrauddagiri | Stardanc Title: My own site! by @upendrauddagiri | Stardance...
|
1793400879.1214538
|
1794005679.1214538
|
|
8114
|
Chrono-Eyes by @upendra48 | Stardance
|
upendra48
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: Chrono-Eyes by @upendra48 | Stardance
|
1793166431.2467
|
1793771231.2467
|
|
48150
|
NeXOS by @upadhyayrudra1205 | Stardance
|
upadhyayrudra1205
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: NeXOS by @upadhyayrudra1205 | Stardance
|
1793216925.9043386
|
1793821725.9043386
|
|
48158
|
Hackypaddy by @upadhyayrudra1205 | Stardance
|
upadhyayrudra1205
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: Hackypaddy by @upadhyayrudra1205 | Stardanc Title: Hackypaddy by @upadhyayrudra1205 | Stardance...
|
1793216954.8123431
|
1793821754.8123431
|
|
44935
|
OS for prettines by @until-wannabe-mule | Stardanc OS for prettines by @until-wannabe-mule | Stardance...
|
until-wannabe-mule
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: OS for prettines by @until-wannabe-mule | S Title: OS for prettines by @until-wannabe-mule | Stardance...
|
1793303509.6291468
|
1793908309.6291468
|
|
54194
|
My own site! by @unstdv | Stardance
|
unstdv
|
0
|
2.46
|
0
|
[]
|
[{"post_id": 59626, "ts": "202 [{"post_id": 59626, "ts": "2026-09-01T15:25:09Z", "hours": 2.4586, "text": "After about 2 hours and some review of HTML/CSS, I am glad to introduce my personal website, which I redesigned and created from scratch (using AI to clarify the implementation of things like a static navbar or scroll snap). It’s still a relatively simple site, but I aim to add effects to it."}]...
|
Title: My own site! by @unstdv | Stardance
Devlog Title: My own site! by @unstdv | Stardance
Devlogs:
After about 2 hours and some review of HTML/CSS, I am glad to introduce my personal website, which I redesigned and created from scratch (using AI to clarify the implementation of things like a static navbar or scroll snap). It’s still a relatively simple site, but I aim to add effects to it....
|
1792777995.5640767
|
1793382795.5640767
|
|
45139
|
Aurameon by @unsent-dupe-impish | Stardance
|
unsent-dupe-impish
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: Aurameon by @unsent-dupe-impish | Stardance
|
1793317668.4715748
|
1793922468.4715748
|
|
12037
|
macropad by @unsaltedflan | Stardance
|
unsaltedflan
|
1
|
1.0
|
0
|
[{"ship_num": 1, "ts": "2026-0 [{"ship_num": 1, "ts": "2026-06-22T18:20:19Z", "hours": 1.0, "multiplier": null, "payout": 0, "blessing": "neutral", "text": "A macropad!! Made this three key macropad to complement a short videogame i made, therefore making it easier to play since i believe it makes things easier to solely have the keys you need (here 1 2 and 3). Very proud of how it came out given i am an absolute beginner and i was making my life a little harder given i used Freecad instead of fusion XD . Firmware was also quite a mess at first but now i feel more confident and im glad ive learnt so much !! attached is my wonderful case that only took like 10 hours lol. hope the community is as proud of me as i am :D", "repo": "https://github.com/unsaltedflan/macropad", "demo": ""}]...
|
[{"post_id": 13434, "ts": "202 [{"post_id": 13434, "ts": "2026-06-22T00:29:11Z", "hours": 0.2667, "text": "Firmware + Hardware done!! I have finally finished this project although im still unsure about the firmware as its mostly based off what i could understand but it seems to work so yay havent really installed hacktime on my laptop so not many hours logged but believe me many spent on this little macropad… happy its finally done and proud of how it came out (also first project done in hackclub!!) and a whopping total of FORTY THREE commits which to a newcomer like me feels insane anyways it was fun but im glad its over"}, {"post_id": 11021, "ts": "2026-06-18T01:27:56Z", "hours": 0.4333, "text": "Day 5-7 : Firmware Firmware has been a very bumpy ride so far as i am still a complete beginner and have been having trouble with the documentation… Thankfully ive been doing my research and i feel like im learning so much about keyboards and their intricate systems!! (also very happy because my GitHub acc had been acting up but ive fixed my mistakes) (img is my finally organized repo)"}, {"post_id": 8165, "ts": "2026-06-13T16:38:17Z", "hours": 0.3175, "text": "macropad i am very new to tech so im super proud although its not as impressive as most of the stuff here hah im working on a macropad like the recommended starter project while using freecad (it was hard (and long)) im honestly just happy ive gotten this far so far just the case which i slightly altered to fit my complicated ideas, pcb and the schematic (my case design is the screenshot linked)"}]...
|
Title: macropad by @unsaltedflan | Stardance
Devl Title: macropad by @unsaltedflan | Stardance
Devlogs:
macropad i am very new to tech so im super proud although its not as impressive as most of the stuff here hah im working on a macropad like the recommended starter project while using freecad (it was hard (and long)) im honestly just happy ive gotten this far so far just the case which i slightly altered to fit my complicated ideas, pcb and the schematic (my case design is the screenshot linked)
Day 5-7 : Firmware Firmware has been a very bumpy ride so far as i am still a complete beginner and have been having trouble with the documentation… Thankfully ive been doing my research and i feel like im learning so much about keyboards and their intricate systems!! (also very happy because my GitHub acc had been acting up but ive fixed my mistakes) (img is my finally organized repo)
Firmware + Hardware done!! I have finally finished this project although im still unsure about the firmware as its mostly based off what i could understand but it seems to work so yay havent really installed hacktime on my laptop so not many hours logged but believe me many spent on this little macropad… happy its finally done and proud of how it came out (also first project done in hackclub!!) and a whopping total of FORTY THREE commits which to a newcomer like me feels insane anyways it was fun but im glad its over
Ships:
A macropad!! Made this three key macropad to complement a short videogame i made, therefore making it easier to play since i believe it makes things easier to solely have the keys you need (here 1 2 and 3). Very proud of how it came out given i am an absolute beginner and i was making my life a little harder given i used Freecad instead of fusion XD . Firmware was also quite a mess at first but now i feel more confident and im glad ive learnt so much !! attached is my wonderful case that only took like 10 hours lol. hope the community is as proud of me as i am :D...
|
1793325141.439875
|
1793411541.439875
|
|
23857
|
FlanOS by @unsaltedflan | Stardance
|
unsaltedflan
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: FlanOS by @unsaltedflan | Stardance
|
1793166568.5491219
|
1793771368.5491219
|
|
25034
|
STUBs discord bot by @unsaltedflan | Stardance
|
unsaltedflan
|
0
|
3.38
|
0
|
[]
|
[{"post_id": 20483, "ts": "202 [{"post_id": 20483, "ts": "2026-07-02T02:47:13Z", "hours": 1.2236, "text": "DEVLOG 3 goodness. its not going very well. honestly its been all over the place and ive been working on someone else’s laptop which definitely did not help but ive problem-solved A. LOT. although i feel like im making progress as ive written up a bunch of commands, i cant seem to get the tunnel from my code to my tunneling service (here ngrok) so my bot is just not online and non-responsive. i have officially scoured every website that even mentions the error im facing and nothing fully applies to my situation so im a little discouraged… but seeing this error, it fills me with determination. so i continue this journey to getting what was supposed to be a simple project to actually function and although i know no one is reading this id love if anyone has any alternative to ngrok or has any knowledge to share about online tunneling to help a friend out T-T . stay tuned for the victory that is to come !! (hopefully)"}, {"post_id": 19530, "ts": "2026-06-30T21:07:37Z", "hours": 1.425, "text": "DEVLOG 2 yayay i finally got it working and fully online very happy :3 i ended up switching to the documentation from the discord developer portal to the official discord javascript extension documentation and it made it wayyy more clear as they do things a little differently and its honestly way easier to use. im ngl i didn’t add much to the actual bot as ive been focused on starting the code properly (plus ive been touching grass lately) and in all honesty this is like the fourth time i restart because i keep making mistakes in the setup or just mixing myself up too much (its sad ik) and now that ive restarted i have like twenty million tabs open so next step is adding my custom commands of which im planning to make a ping command, a reminder command, an 8 ball command and maybe a cool fact command so stay tuned XD"}, {"post_id": 15828, "ts": "2026-06-25T18:52:04Z", "hours": 0.7289, "text": "DEVLOG 1 i am currently working on a custom discord bot based off my oc as i think it would be interesting to learn and it seems like a good opportunity to give my servers/ my account a “glow up”. i have been following the discord bot and companion app developpers guide but im having a lot of trouble getting my code to send to my bot and getting my bot online as i need a third party platform to do so and im still a little lost im also seriously debating switching my entire project to a python project instead of java script as it might make the communication with my bot easier to manage from what ive picked up on the web i havent really implemented any commands (obviously i dont count the premade commands that come in the initially offered sample code) as im mainly focused on getting my bot to RESPOND?!? if anyone has any tips or ressources that might be more helpful than the discord docs or any thoughts on switch to discord.py please lmk :3 attached img is my little STUBs (my oc) as a discord logoish -w- and thats it for today bye folks"}]...
|
Title: STUBs discord bot by @unsaltedflan | Starda Title: STUBs discord bot by @unsaltedflan | Stardance
Devlogs:
DEVLOG 1 i am currently working on a custom discord bot based off my oc as i think it would be interesting to learn and it seems like a good opportunity to give my servers/ my account a “glow up”. i have been following the discord bot and companion app developpers guide but im having a lot of trouble getting my code to send to my bot and getting my bot online as i need a third party platform to do so and im still a little lost im also seriously debating switching my entire project to a python project instead of java script as it might make the communication with my bot easier to manage from what ive picked up on the web i havent really implemented any commands (obviously i dont count the premade commands that come in the initially offered sample code) as im mainly focused on getting my bot to RESPOND?!? if anyone has any tips or ressources that might be more helpful than the discord docs or any thoughts on switch to discord.py please lmk :3 attached img is my little STUBs (my oc) as a discord logoish -w- and thats it for today bye folks
DEVLOG 2 yayay i finally got it working and fully online very happy :3 i ended up switching to the documentation from the discord developer portal to the official discord javascript extension documentation and it made it wayyy more clear as they do things a little differently and its honestly way easier to use. im ngl i didn’t add much to the actual bot as ive been focused on starting the code properly (plus ive been touching grass lately) and in all honesty this is like the fourth time i restart because i keep making mistakes in the setup or just mixing myself up too much (its sad ik) and now that ive restarted i have like twenty million tabs open so next step is adding my custom commands of which im planning to make a ping command, a reminder command, an 8 ball command and maybe a cool fact command so stay tuned XD
DEVLOG 3 goodness. its not going very well. honestly its been all over the place and ive been working on someone else’s laptop which definitely did not help but ive problem-solved A. LOT. although i feel like im making progress as ive written up a bunch of commands, i cant seem to get the tunnel from my code to my tunneling service (here ngrok) so my bot is just not online and non-responsive. i have officially scoured every website that even mentions the error im facing and nothing fully applies to my situation so im a little discouraged… but seeing this error, it fills me with determination. so i continue this journey to getting what was supposed to be a simple project to actually function and although i know no one is reading this id love if anyone has any alternative to ngrok or has any knowledge to share about online tunneling to help a friend out T-T . stay tuned for the victory that is to come !! (hopefully)...
|
1793480241.4497943
|
1794085041.4497943
|
|
32557
|
SignalCore by @unreturnedguy | Stardance
|
unreturnedguy
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: SignalCore by @unreturnedguy | Stardance
|
1793371958.727337
|
1793976758.727337
|
|
32569
|
My very own WebOS by @unreturnedguy | Stardance
|
unreturnedguy
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: My very own WebOS by @unreturnedguy | Stard Title: My very own WebOS by @unreturnedguy | Stardance...
|
1793436503.6169877
|
1794041303.6169877
|
|
1236
|
Fuddi AI by @unonown97 | Stardance
|
unonown97
|
1
|
64.0
|
1003
|
[{"ship_num": 1, "ts": "2026-0 [{"ship_num": 1, "ts": "2026-06-13T16:30:40Z", "hours": 64.0, "multiplier": 16.63, "payout": 1003, "blessing": "neutral", "text": "A sleek, privacy-first multi-model chat interface that runs entirely in your browser. Connects to dozens of AI models through a lightweight local proxy your API keys never leave your machine. so while deploying i encountered multiple issues and but the web helped me thankfully and also the costing doesnt work at all as of now and is something i will be fixing in the future :). g’day", "repo": "https://github.com/Miroredrpo/ai-chat-app", "demo": "https://fuddigai.vercel.app/"}]...
|
[{"post_id": 8139, "ts": "2026 [{"post_id": 8139, "ts": "2026-06-13T16:10:05Z", "hours": 0.7906, "text": "tried deploying to vercel but it didnt work tweaked some things in the server.js but reading online i foundout that i will need to make another folder for it to function properly so am doing that rn :)"}, {"post_id": 8102, "ts": "2026-06-13T14:58:43Z", "hours": 5.0, "text": "added an option to export and import all the things into a JSON, also an option to clear all the conversations, and remove all the API keys as now it has option to add multiple API keys, all the things are pretty much done so i think i will just do some finishing changes and ship it for the first time"}, {"post_id": 7843, "ts": "2026-06-13T02:35:10Z", "hours": 3.5333, "text": "added the prompting tab where specific instructions can be written and also completed the interface tab where one can toggle things like auto title generation and stuff am goging to add support for multiple api keys next and also an export/import chats and settings feutesrs that exports everthing, so when the local storage is full one can export and save there chat and import thhem later on"}, {"post_id": 6831, "ts": "2026-06-11T13:40:44Z", "hours": 5.4056, "text": "added option to pin the chat, edit the title manually, duplicate the chat and delete it also added fast delete which when shift is held while deletion it doesnt ask for conformation and directly deletes the chat, also added all the paraments accepted as far as i am aware of, added a default system prompt and i will also be making another tab specifically for prompts, also added dynamic size of the chat where the font size can be changed from the setting menu it self"}, {"post_id": 6076, "ts": "2026-06-10T06:43:41Z", "hours": 0.6, "text": "So… i tried using AI to do some visual changes only for it to completely messup the layout and populate the root directory with random js and css files saying these are the new ones and have the same logic and makes the app better but it did not do any of that, i reverted it to how it was i dont know if that counts to AI useage but i will still be putting the incident in the details of the project"}, {"post_id": 5753, "ts": "2026-06-09T17:31:13Z", "hours": 0.9672, "text": "when an API is enterd made a test endpoint so it tests if the API is correct or not and tells it also initialized the js logic for the thinking mode max tokens, temperature, max tokens (entire chat), and a default system prompt, also am thinking about adding a dedicated tab in the settings panel for prompts and all"}, {"post_id": 5606, "ts": "2026-06-09T13:51:25Z", "hours": 8.4333, "text": "completed the toasts, added web search functionality, added chat deletion, also populated the empty state with some prompts, added an option to choose if one wants to generate the title automatically or not in a new settings tab “interface” where the size of fonts, theme, etc can me changed, am goging to add things like temperature , maxtokens, thinking budget and all next"}, {"post_id": 5329, "ts": "2026-06-09T02:01:33Z", "hours": 1.2681, "text": "added a shortcuts tab that tells the users about the keybinds, also made the initial frame for toasts, also added a last refreshed time on the models tab and reload again button that again fetches it from the HCAI API"}, {"post_id": 4902, "ts": "2026-06-08T14:08:23Z", "hours": 10.1728, "text": "added empty state also added thinking mode where it sends the api with an extra parameter and enables the thinking mode if the model has the ability to do it, i also added some keybinds such as when ctrl + Shift + T is pressed thinkning mode is enabled, Ctrl+N opens new chat etc it is also confiured for mac instead of ctrl in mac we can use cmd, in the settings tab i have also added a dropdown menu to select a separate title generation model"}, {"post_id": 4073, "ts": "2026-06-07T08:25:09Z", "hours": 8.1322, "text": "i made it so that the model can be changed and accessed from the chat it self and change it as per the user intent, i also added a details panel showing thw models capabilities and its pricing and its abilities like if it can web search or not or if it has thinking mode etc. also now in the settings panel you can select the default model to be used when a new chant is opened and it is stored in the local storage similar to how the chat is stored"}, {"post_id": 3276, "ts": "2026-06-06T04:04:02Z", "hours": 12.3142, "text": "I have added a model selector now which pulls the models from https://ai.hackclub.com/proxy/v1/models although it has not been properly configures i will be making it so that there will be an option for a title generation model in the settings as well i have also added a side bar where previos chats can be accessed, up next: i will be adding a details panel that will allow one to veiw the context cost and provider, modelname, etc of the selected model, i will also be adding the feature to attach images and all, i have also made it responsive in this past time frame not entirely but i have made it fairly useable i will also be making it more mobile friendly"}, {"post_id": 1933, "ts": "2026-06-04T05:33:12Z", "hours": 2.745, "text": "implemented dark mode made layout cleaner added option to set base url, also made it so that the API gets saved in the local storage for future reload"}, {"post_id": 1603, "ts": "2026-06-03T18:49:52Z", "hours": 2.9808, "text": "made the layout better with buttons outlins and made a floating settings panel to input the HCAI API key into so that AI models can be accessed later, next changes: shift layout, figureout server calls and add more functionality also add dark mode :)"}, {"post_id": 269, "ts": "2026-06-01T14:20:29Z", "hours": 1.355, "text": "i am making an custom interface where one can bring their HCAI API key and talk to various LLM’s"}]...
|
Title: Fuddi AI by @unonown97 | Stardance
Devlogs Title: Fuddi AI by @unonown97 | Stardance
Devlogs:
i am making an custom interface where one can bring their HCAI API key and talk to various LLM’s
made the layout better with buttons outlins and made a floating settings panel to input the HCAI API key into so that AI models can be accessed later, next changes: shift layout, figureout server calls and add more functionality also add dark mode :)
implemented dark mode made layout cleaner added option to set base url, also made it so that the API gets saved in the local storage for future reload
I have added a model selector now which pulls the models from https://ai.hackclub.com/proxy/v1/models although it has not been properly configures i will be making it so that there will be an option for a title generation model in the settings as well i have also added a side bar where previos chats can be accessed, up next: i will be adding a details panel that will allow one to veiw the context cost and provider, modelname, etc of the selected model, i will also be adding the feature to attach images and all, i have also made it responsive in this past time frame not entirely but i have made it fairly useable i will also be making it more mobile friendly
i made it so that the model can be changed and accessed from the chat it self and change it as per the user intent, i also added a details panel showing thw models capabilities and its pricing and its abilities like if it can web search or not or if it has thinking mode etc. also now in the settings panel you can select the default model to be used when a new chant is opened and it is stored in the local storage similar to how the chat is stored
added empty state also added thinking mode where it sends the api with an extra parameter and enables the thinking mode if the model has the ability to do it, i also added some keybinds such as when ctrl + Shift + T is pressed thinkning mode is enabled, Ctrl+N opens new chat etc it is also confiured for mac instead of ctrl in mac we can use cmd, in the settings tab i have also added a dropdown menu to select a separate title generation model
added a shortcuts tab that tells the users about the keybinds, also made the initial frame for toasts, also added a last refreshed time on the models tab and reload again button that again fetches it from the HCAI API
completed the toasts, added web search functionality, added chat deletion, also populated the empty state with some prompts, added an option to choose if one wants to generate the title automatically or not in a new settings tab “interface” where the size of fonts, theme, etc can me changed, am goging to add things like temperature , maxtokens, thinking budget and all next
when an API is enterd made a test endpoint so it tests if the API is correct or not and tells it also initialized the js logic for the thinking mode max tokens, temperature, max tokens (entire chat), and a default system prompt, also am thinking about adding a dedicated tab in the settings panel for prompts and all
So… i tried using AI to do some visual changes only for it to completely messup the layout and populate the root directory with random js and css files saying these are the new ones and have the same logic and makes the app better but it did not do any of that, i reverted it to how it was i dont know if that counts to AI useage but i will still be putting the incident in the details of the project
added option to pin the chat, edit the title manually, duplicate the chat and delete it also added fast delete which when shift is held while deletion it doesnt ask for conformation and directly deletes the chat, also added all the paraments accepted as far as i am aware of, added a default system prompt and i will also be making another tab specifically for prompts, also added dynamic size of the chat where the font size can be changed from the setting menu it self
added the prompting tab where specific instructions can be written and also completed the interface tab where one can toggle things like auto title generation and stuff am goging to add support for multiple api keys next and also an export/import chats and settings feutesrs that exports everthing, so when the local storage is full one can export and save there chat and import thhem later on
added an option to export and import all the things into a JSON, also an option to clear all the conversations, and remove all the API keys as now it has option to add multiple API keys, all the things are pretty much done so i think i will just do some finishing changes and ship it for the first time
tried deploying to vercel but it didnt work tweaked some things in the server.js but reading online i foundout that i will need to make another folder for it to function properly so am doing that rn :)
Ships:
A sleek, privacy-first multi-model chat interface that runs entirely in your browser. Connects to dozens of AI models through a lightweight local proxy your API keys never leave your machine. so while deploying i encountered multiple issues and but the web helped me thankfully and also the costing doesnt work at all as of now and is something i will be fixing in the future :). g’day...
|
1792878545.9104006
|
1793742545.9104006
|
|
19877
|
Svg image to animation +code+mp4 by @unonown97 | S Svg image to animation +code+mp4 by @unonown97 | Stardance...
|
unonown97
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: Svg image to animation +code+mp4 by @unonow Title: Svg image to animation +code+mp4 by @unonown97 | Stardance...
|
1792932790.3633273
|
1793796790.3633273
|
|
23393
|
Air draw by @unonown97 | Stardance
|
unonown97
|
1
|
6.49
|
37
|
[{"ship_num": 1, "ts": "2026-0 [{"ship_num": 1, "ts": "2026-06-23T17:38:03Z", "hours": 6.0, "multiplier": 6.21, "payout": 37, "blessing": "neutral", "text": "Air Draw is a python script that tracks your hand through your webcam and lets you paint on screen by moving your index finger around. It uses MediaPipe under the hood to figure out where your hand and fingers are, and OpenCV to show the camera feed and draw the lines. Quick Controls: -Index finger up: Move Brush -Pinch (Index + Thumb): Lift Brush (Pause Drawing) -Index + Middle fingers up: Select Color / Clear -Press ‘q’ to quit Sometimes when you switch from drawing mode to selection mode quickly the brush leaves a small stray line before resetting. Havent fixed this yet.", "repo": "https://github.com/Mirvo19/Air-draw", "demo": "https://github.com/Mirvo19/Air-draw/releases/tag/1.2"}]...
|
[{"post_id": 14606, "ts": "202 [{"post_id": 14606, "ts": "2026-06-23T19:37:58Z", "hours": 0.4917, "text": "created a github workflow that automatically generates the exe every time a new release is created"}, {"post_id": 14505, "ts": "2026-06-23T17:25:43Z", "hours": 5.9039, "text": "Air Draw Wanted to try making an air drawing thing from scratch. Most tutorials use the old mediapipe Hands API which is getting deprecated so I used the newer Tasks API instead. Canvas layer Keeping a separate image layer for the drawing and merging it with the camera feed each frame was the annoying part. Empty parts of the canvas had to be made transparent using a threshold mask and bitwise operations. cv2.threshold + cv2.bitwise_and + cv2.bitwise_or is the combo that works. Gesture detection First tried checking if fingertip y coordinate was above the knuckle y coordinate. Broke whenever the hand tilted sideways. Switched to distance based: if fingertip is farther from the wrist than the middle knuckle, finger is extended. Works regardless of hand angle. Smoothing Two layers of smoothing. One on the landmark positions themselves and one on the actual drawing point Pinch to lift Fixed pixel distance threshold for pinch detection broke whenever the hand moved closer or farther from the camera. Made the threshold relative to palm size instead (35% of wrist to middle knuckle distance). Much more consistent. Controls (nutshell): Index finger up: Move Brush Pinch (Index + Thumb): Lift Brush (Pause Drawing) Index + Middle fingers up: Select Color / Clear Press ‘q’ to quit"}]...
|
Title: Air draw by @unonown97 | Stardance
Devlogs Title: Air draw by @unonown97 | Stardance
Devlogs:
Air Draw Wanted to try making an air drawing thing from scratch. Most tutorials use the old mediapipe Hands API which is getting deprecated so I used the newer Tasks API instead. Canvas layer Keeping a separate image layer for the drawing and merging it with the camera feed each frame was the annoying part. Empty parts of the canvas had to be made transparent using a threshold mask and bitwise operations. cv2.threshold + cv2.bitwise_and + cv2.bitwise_or is the combo that works. Gesture detection First tried checking if fingertip y coordinate was above the knuckle y coordinate. Broke whenever the hand tilted sideways. Switched to distance based: if fingertip is farther from the wrist than the middle knuckle, finger is extended. Works regardless of hand angle. Smoothing Two layers of smoothing. One on the landmark positions themselves and one on the actual drawing point Pinch to lift Fixed pixel distance threshold for pinch detection broke whenever the hand moved closer or farther from the camera. Made the threshold relative to palm size instead (35% of wrist to middle knuckle distance). Much more consistent. Controls (nutshell): Index finger up: Move Brush Pinch (Index + Thumb): Lift Brush (Pause Drawing) Index + Middle fingers up: Select Color / Clear Press ‘q’ to quit
created a github workflow that automatically generates the exe every time a new release is created
Ships:
Air Draw is a python script that tracks your hand through your webcam and lets you paint on screen by moving your index finger around. It uses MediaPipe under the hood to figure out where your hand and fingers are, and OpenCV to show the camera feed and draw the lines. Quick Controls: -Index finger up: Move Brush -Pinch (Index + Thumb): Lift Brush (Pause Drawing) -Index + Middle fingers up: Select Color / Clear -Press ‘q’ to quit Sometimes when you switch from drawing mode to selection mode quickly the brush leaves a small stray line before resetting. Havent fixed this yet....
|
1792878518.6568174
|
1793742518.6568174
|
|
33721
|
My own site! by @unonown97 | Stardance
|
unonown97
|
0
|
1.99
|
0
|
[]
|
[{"post_id": 34611, "ts": "202 [{"post_id": 34611, "ts": "2026-07-25T14:10:19Z", "hours": 1.9908, "text": "I updated my main portfolio by updating my new gihub (school email) and then i also changed my discord webhook url and how it sent the message to the channel. I also hosted it in my name! i’ve changed the animations and tweaked it a little (also added a few ester eggs :)"}]...
|
Title: My own site! by @unonown97 | Stardance
Dev Title: My own site! by @unonown97 | Stardance
Devlogs:
I updated my main portfolio by updating my new gihub (school email) and then i also changed my discord webhook url and how it sent the message to the channel. I also hosted it in my name! i’ve changed the animations and tweaked it a little (also added a few ester eggs :)...
|
1792907502.9759219
|
1793771502.9759219
|
|
40539
|
I just made an AI Agent by @unokaiysh182 | Stardan I just made an AI Agent by @unokaiysh182 | Stardance...
|
unokaiysh182
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: I just made an AI Agent by @unokaiysh182 | Title: I just made an AI Agent by @unokaiysh182 | Stardance...
|
1793274849.1881945
|
1793879649.1881945
|
|
19712
|
My very own WebOS by @unnatisingh | Stardance
|
unnatisingh
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: My very own WebOS by @unnatisingh | Stardan Title: My very own WebOS by @unnatisingh | Stardance...
|
1793177317.8742964
|
1793782117.8742964
|
|
4934
|
My awesome macropad! by @unlostkomodo | Stardance
|
unlostkomodo
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: My awesome macropad! by @unlostkomodo | Sta Title: My awesome macropad! by @unlostkomodo | Stardance...
|
1792806810.5610738
|
1793411610.5610738
|
|
44972
|
My awesome macropad! by @unknwnash | Stardance
|
unknwnash
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: My awesome macropad! by @unknwnash | Starda Title: My awesome macropad! by @unknwnash | Stardance...
|
1793307159.9123049
|
1793911959.9123049
|
|
45609
|
My awesome macropad! by @unknownaster | Stardance
|
unknownaster
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: My awesome macropad! by @unknownaster | Sta Title: My awesome macropad! by @unknownaster | Stardance...
|
1793339907.5065558
|
1793944707.5065558
|
|
24928
|
Ai training playground by @unknown_lizzard | Stard Ai training playground by @unknown_lizzard | Stardance...
|
unknown_lizzard
|
1
|
0.0
|
3
|
[{"ship_num": 1, "ts": "2026-0 [{"ship_num": 1, "ts": "2026-06-24T14:12:17Z", "hours": 0.0, "multiplier": 10.73, "payout": 3, "blessing": "neutral", "text": "I built Telokine, a 3D robot trainer where you snap motors onto parts, drag reward blocks to say what to learn, hit Train, and watch it improve. Frontend is React and Three.js, backend is Python with MuJoCo physics and PPO. The hard parts were fixing the robot stretching on Run, stopping the cube from moving with no motors, getting training previews to actually show, and making Run trained work after training. GitHub Pages only hosts the UI so real training runs on your laptop. I’m proud of the motor snapping, the block editor, the live previews during training, and making movement motor-only like a real robot. To try it: the demo at https://mateooo93.github.io/Telokine/ is UI only. For the full thing clone https://github.com/Mateooo93/Telokine , run npm run dev and the backend on port 8000, load the Walker template, Train then Run trained. No GPU needed, but you need motors or nothing moves.", "repo": "https://github.com/Mateooo93/Telokine", "demo": "https://mateooo93.github.io/Telokine/"}]...
|
[{"post_id": 15068, "ts": "202 [{"post_id": 15068, "ts": "2026-06-24T14:08:56Z", "hours": 0.2853, "text": "DEVLOG 1: telokine is kinda working?? 🤖 So far I have been developing Telokine for Stardance and it allows you to train a robot without writing any code. I built this because I wanted RL without having to write complex code or/and make complex jupyter notebooks, this is very simple and kid-friendly way to understand how it works visually. anyways, This works with your gpu and/or cpu, but expect some bugs as it’s not completely finished yet(expect new features soon!). (forgot to add it to hackatime so it just shows up as 15 minutes of work)(it took me more like 5-8 hours), I used ai for the frontend and cuda. Links GitHub repo Live demo (UI only)"}]...
|
Title: Ai training playground by @unknown_lizzard Title: Ai training playground by @unknown_lizzard | Stardance
Devlogs:
DEVLOG 1: telokine is kinda working?? 🤖 So far I have been developing Telokine for Stardance and it allows you to train a robot without writing any code. I built this because I wanted RL without having to write complex code or/and make complex jupyter notebooks, this is very simple and kid-friendly way to understand how it works visually. anyways, This works with your gpu and/or cpu, but expect some bugs as it’s not completely finished yet(expect new features soon!). (forgot to add it to hackatime so it just shows up as 15 minutes of work)(it took me more like 5-8 hours), I used ai for the frontend and cuda. Links GitHub repo Live demo (UI only)
Ships:
I built Telokine, a 3D robot trainer where you snap motors onto parts, drag reward blocks to say what to learn, hit Train, and watch it improve. Frontend is React and Three.js, backend is Python with MuJoCo physics and PPO. The hard parts were fixing the robot stretching on Run, stopping the cube from moving with no motors, getting training previews to actually show, and making Run trained work after training. GitHub Pages only hosts the UI so real training runs on your laptop. I’m proud of the motor snapping, the block editor, the live previews during training, and making movement motor-only like a real robot. To try it: the demo at https://mateooo93.github.io/Telokine/ is UI only. For the full thing clone https://github.com/Mateooo93/Telokine , run npm run dev and the backend on port 8000, load the Walker template, Train then Run trained. No GPU needed, but you need motors or nothing moves....
|
1792878540.0689528
|
1793742540.0689528
|
|
25201
|
OrbitScope by @unknown_lizzard | Stardance
|
unknown_lizzard
|
1
|
1.0
|
1
|
[{"ship_num": 1, "ts": "2026-0 [{"ship_num": 1, "ts": "2026-06-26T00:49:54Z", "hours": 1.0, "multiplier": 2.13, "payout": 1, "blessing": "neutral", "text": "I shipped OrbitScope, my Slack bot for the Stardance Slack Bot mission. At first it was just a basic bot with /orbitscope-ping so I could check if it was even working. After that I wanted to make it more interesting because I really like space, so I connected it to NASA APIs. I made /orbitscope-space_pic, which sends a random space picture from NASA’s APOD API. The API was kinda annoying at first because today’s image sometimes didn’t work, so I made it choose random past dates instead. At first the bot was only sending the image URL, which looked boring, but then I figured out that Slack has image blocks with type: image, so now the picture actually shows up in Slack. Then I added /orbitscope-mars. I first tried the Mars rover API but it was broken/weird, so I switched to NASA’s image library and used that to get Mars images instead. I also added /orbitscope-help so people can see the commands. Overall it’s a small bot, but it actually feels cool to use because you can get real NASA space images directly in Slack.", "repo": "https://github.com/Mateooo93/OrbitScope-slackbot", "demo": "https://hackclub.enterprise.slack.com/archives/C0BE829ADBJ"}]...
|
[{"post_id": 16025, "ts": "202 [{"post_id": 16025, "ts": "2026-06-26T00:33:05Z", "hours": 0.6789, "text": "Devlog 1: Beginning of OrbitScope Started working on OrbitScope, this is my Slack bot for the Stardance Slack Bot mission. First, I made a command to ping the bot: /orbitscope-ping(as a sanity check). Since I love space and everything about it, I decided to make a command thats shows you a random pic from space using the APOD nasa API, the api was pretty tricky to get working at first, but after some research I managed to make it work pretty well. command: /orbitscope-space_pic While I was coding this I noticed that sending the url of the nasa image wasn’t showing the image in the slack workspace but I found out that I could use: “type: image” which tells the slack bot to send the url as an image. Furthermore, I started expanding my commands, I made a similar one to the space_pic command, instead of sending a APOD pick from the nasa api I used the mars rover pictures api, so it ended up being pretty easy to code. command: /orbitscope-mars"}]...
|
Title: OrbitScope by @unknown_lizzard | Stardance
Title: OrbitScope by @unknown_lizzard | Stardance
Devlogs:
Devlog 1: Beginning of OrbitScope Started working on OrbitScope, this is my Slack bot for the Stardance Slack Bot mission. First, I made a command to ping the bot: /orbitscope-ping(as a sanity check). Since I love space and everything about it, I decided to make a command thats shows you a random pic from space using the APOD nasa API, the api was pretty tricky to get working at first, but after some research I managed to make it work pretty well. command: /orbitscope-space_pic While I was coding this I noticed that sending the url of the nasa image wasn’t showing the image in the slack workspace but I found out that I could use: “type: image” which tells the slack bot to send the url as an image. Furthermore, I started expanding my commands, I made a similar one to the space_pic command, instead of sending a APOD pick from the nasa api I used the mars rover pictures api, so it ended up being pretty easy to code. command: /orbitscope-mars
Ships:
I shipped OrbitScope, my Slack bot for the Stardance Slack Bot mission. At first it was just a basic bot with /orbitscope-ping so I could check if it was even working. After that I wanted to make it more interesting because I really like space, so I connected it to NASA APIs. I made /orbitscope-space_pic, which sends a random space picture from NASA’s APOD API. The API was kinda annoying at first because today’s image sometimes didn’t work, so I made it choose random past dates instead. At first the bot was only sending the image URL, which looked boring, but then I figured out that Slack has image blocks with type: image, so now the picture actually shows up in Slack. Then I added /orbitscope-mars. I first tried the Mars rover API but it was broken/weird, so I switched to NASA’s image library and used that to get Mars images instead. I also added /orbitscope-help so people can see the commands. Overall it’s a small bot, but it actually feels cool to use because you can get real NASA space images directly in Slack....
|
1792875262.9227936
|
1793739262.9227936
|
|
25714
|
OrbitOS by @unknown_lizzard | Stardance
|
unknown_lizzard
|
1
|
2.0
|
0
|
[{"ship_num": 1, "ts": "2026-0 [{"ship_num": 1, "ts": "2026-06-25T11:49:35Z", "hours": 2.0, "multiplier": null, "payout": 0, "blessing": "neutral", "text": "space/mission-control desktop. minimal space control inspired OS with mission logs and launch control! (WARNING: some buttons may actually launch a rocket) added 3 apps, artemis launch control, rocket build and mission log, they are all functional and fun to use make sure you don’t launch too many rockets! I also added a small incremental game called “stardust” this is a very simple game where you mine stardust and upgrade your drill to mine more stardust if you see a spinning moon, don’t be alarmed(we accidently crashed into the moon in the proccess of testing a rocket) Jack’s message to earth:“Houston we don’t have a problem” ISSUES I was having an issue with each element stacking on top of eachother but I ended up figuring it out quite quickly!’ 2.In the beginning, I had a bit of trouble figuring out the css, but it ended up going very well, all the buttons work and you can even check out the source code", "repo": "https://github.com/Mateooo93/OrbitOS", "demo": "https://mateooo93.github.io/OrbitOS/"}]...
|
[{"post_id": 15608, "ts": "202 [{"post_id": 15608, "ts": "2026-06-25T11:46:35Z", "hours": 0.2628, "text": "DEVLOG 3: minigames! added a small incremental game called “stardust” this is a very simple game where you mine stardust and upgrade your drill to mine more stardust have fun!"}, {"post_id": 15605, "ts": "2026-06-25T11:30:41Z", "hours": 0.3131, "text": "DEVLOG 2: rocket builder added a new rocket builder app, now you can build the artemis V rocket! I was having an issue with each element stacking on top of eachother but I ended up figuring it out quite quickly!"}, {"post_id": 15589, "ts": "2026-06-25T11:01:22Z", "hours": 1.1289, "text": "DEVLOG 1: OrbitOS is officialy released! packaged with top of the line, military grade software, OrbitOS gives you access to Artemis Launch control. In the beginning, I had a bit of trouble figuring out the css, but it ended up going very well, all the buttons work and you can even check out the source code. I chose a very blocky and minimal UI for this because it matches the mission control feeling very well therefor this was heavily inspired from actual NASA mission control software(complete lie)."}]...
|
Title: OrbitOS by @unknown_lizzard | Stardance
De Title: OrbitOS by @unknown_lizzard | Stardance
Devlogs:
DEVLOG 1: OrbitOS is officialy released! packaged with top of the line, military grade software, OrbitOS gives you access to Artemis Launch control. In the beginning, I had a bit of trouble figuring out the css, but it ended up going very well, all the buttons work and you can even check out the source code. I chose a very blocky and minimal UI for this because it matches the mission control feeling very well therefor this was heavily inspired from actual NASA mission control software(complete lie).
DEVLOG 2: rocket builder added a new rocket builder app, now you can build the artemis V rocket! I was having an issue with each element stacking on top of eachother but I ended up figuring it out quite quickly!
DEVLOG 3: minigames! added a small incremental game called “stardust” this is a very simple game where you mine stardust and upgrade your drill to mine more stardust have fun!
Ships:
space/mission-control desktop. minimal space control inspired OS with mission logs and launch control! (WARNING: some buttons may actually launch a rocket) added 3 apps, artemis launch control, rocket build and mission log, they are all functional and fun to use make sure you don’t launch too many rockets! I also added a small incremental game called “stardust” this is a very simple game where you mine stardust and upgrade your drill to mine more stardust if you see a spinning moon, don’t be alarmed(we accidently crashed into the moon in the proccess of testing a rocket) Jack’s message to earth:“Houston we don’t have a problem” ISSUES I was having an issue with each element stacking on top of eachother but I ended up figuring it out quite quickly!’ 2.In the beginning, I had a bit of trouble figuring out the css, but it ended up going very well, all the buttons work and you can even check out the source code...
|
1792878678.0131688
|
1793742678.0131688
|
|
26073
|
Cortex-CLI by @unknown_lizzard | Stardance
|
unknown_lizzard
|
1
|
15.0
|
78
|
[{"ship_num": 1, "ts": "2026-0 [{"ship_num": 1, "ts": "2026-06-30T20:57:10Z", "hours": 15.0, "multiplier": 7.8, "payout": 78, "blessing": "neutral", "text": "Fixed most bugs before shipping, it’s pretty usable rn. Install it : npm install -g mateooo93-cortex@latest Run it! : cortex", "repo": "https://github.com/Mateooo93/cortex-cli", "demo": "https://mateooo93.github.io/cortex-cli/"}]...
|
[{"post_id": 19444, "ts": "202 [{"post_id": 19444, "ts": "2026-06-30T19:25:29Z", "hours": 15.0014, "text": "DEVLOG #1: the birth of cortex-cli this is a similar type of agent harness tui as pi.dev or opencode, I started making this because I was very interested in agentic workflows and interfaces. For the most part it was easy to make but I hit a while designing the backend, I found it really hard calling the agent and proccessing tool calls correctly, so I inspired myself from pi.dev. I think the main feature of the TUI is that the elements are clickable and actually work in the terminal, at first I found it very hard to do and it felt impossible but after a bit of research I figured it out! furthermore, I decided to chose a very intuitive and minimal UI because other clis felt very hard to get used to."}]...
|
Title: Cortex-CLI by @unknown_lizzard | Stardance
Title: Cortex-CLI by @unknown_lizzard | Stardance
Devlogs:
DEVLOG #1: the birth of cortex-cli this is a similar type of agent harness tui as pi.dev or opencode, I started making this because I was very interested in agentic workflows and interfaces. For the most part it was easy to make but I hit a while designing the backend, I found it really hard calling the agent and proccessing tool calls correctly, so I inspired myself from pi.dev. I think the main feature of the TUI is that the elements are clickable and actually work in the terminal, at first I found it very hard to do and it felt impossible but after a bit of research I figured it out! furthermore, I decided to chose a very intuitive and minimal UI because other clis felt very hard to get used to.
Ships:
Fixed most bugs before shipping, it’s pretty usable rn. Install it : npm install -g mateooo93-cortex@latest Run it! : cortex...
|
1792878722.5608337
|
1793742722.5608337
|
|
29287
|
le-fat-chaton by @unknown_lizzard | Stardance
|
unknown_lizzard
|
1
|
172.0
|
1818
|
[{"ship_num": 1, "ts": "2026-0 [{"ship_num": 1, "ts": "2026-08-20T09:10:04Z", "hours": 172.0, "multiplier": 14.88, "payout": 1818, "blessing": "neutral", "text": "at first I wanted to make a normal text model to try and learn more about llms, but I wanted to go further, agentic coding is a big thing so what better way to learn about llms than trying to make an agentic coding ai model. So thats how le gros chaton was born, the name originated from a joke on x about a fake huge mistral model coming out. to train this model I used already made datasets on hugging face, like fable5 traces, I also used kimi k3 as a teacher model to teach my smaller model to be good at coding and running commands, it kind of worked? but i’m not completely satisfied with the result, but at least i have learned some things and soon enough I will trying making an even better model! I dont have a demo url for you guys to try the model out since it was too expensive to run on a gpu so you’ll need to run it yourself, check it out on hugging face, follow the instructions and have fun.", "repo": "https://github.com/Mateooo93/le-gros-chaton", "demo": "https://mateooo93.github.io/le-gros-chaton/"}]...
|
[{"post_id": 50841, "ts": "202 [{"post_id": 50841, "ts": "2026-08-20T08:57:48Z", "hours": 3.2394, "text": "devlog #17 MI300X vLLM serving The old GPU box got destroyed, so I spun up a new MI300X with the same SSH key. ROCm 6.2 and Torch 2.5.1 were already set up, so getting the machine ready was easy. The hard part was vLLM. The pip version is CUDA only, and the ROCm Docker image only supported the multimodal version of Qwen3.5. Our model is text-only, so I had to patch the image. I added Qwen3_5ForCausalLM, fixed the hybrid KV-cache setup, added the missing Mamba state methods, and mapped the weight names so the text-only safetensors loaded correctly. I also patched the merged config to remove the M-RoPE and dtype fields. After that, the model served normally. I ran a one-task test with fix-git. The pipeline worked, but the model got stuck repeating git show until it ran out of turns. ##Shipped The merged model is now public on Hugging Face as mateo0093/le-gros-chaton under Apache-2.0. The 5×5 Terminal-Bench pilot finished at 3/25 (25%). All three successes were on fix-git, which shows that the SFT data was too focused on git tasks. I also tested RLVR with GRPO and a novelty bonus. The reward had almost no variance, so the training signal was basically useless. I saved the step-10 adapter but didn’t merge it. The GPU used about 12 hours of the 50 hour budget. All the vLLM patches and harness changes are in the repo. Main lesson: Qwen3.5’s hybrid model is a bit annoying to serve as text-only on ROCm. The model works but i think this was more of a learning opportunity, i know what i’ve done wrong and what i would have done differently, I will be working on a new model in a few weeks once I do a bit more research."}, {"post_id": 49761, "ts": "2026-08-18T18:01:12Z", "hours": 0.7992, "text": "Devlog #16, t4gpu The T4 16GB fought us the whole way. BF16 was baked into a pre quantized checkpoint, which made each step take 668 seconds because it had to use FP32 emulation. 2K context ran out of memory, the 248K logits ran out of memory too, FP16 started producing NaNs, Kaggle session limits locked us out, and the paged optimizer crashed last. Every fix seemed to create another problem. We eventually got it working with FP16 compute, chunked assistant token loss, autocast, and plain AdamW. Then we got access to an AMD MI300X with 192GB of memory. No quantization, 16K context, and enough memory to train full traces instead of cutting them off halfway through a message. Right now the 16K BF16 SFT run is training on the MI300X. Next up is merging the model, running Terminal Bench, and then RLVR."}, {"post_id": 49140, "ts": "2026-08-17T19:31:53Z", "hours": 16.24, "text": "Devlog 15 Traces are done The trace generator finally hit the target and went past it. We got 474 verified trajectories, zero malformed, zero duplicates, 19 templates, and a 95 percent self review pass rate. We were aiming for 450, so we ended up with 474. The dataset is synced to Hugging Face in two versions: raw and normalized. That part is finally done. It was the longest part of the project, and the model is only going to be as good as the traces, so getting this right was important The Kimi endpoint also stopped working. The old teacher account started returning 503 errors during generation and eventually died completely. We switched to a second Kimi account, which needed an extra header that we did not have at first. After figuring that out, generation resumed without losing any progress For GPUs, the 2070 still cannot realistically train the 9B. Kaggle hours are back, so we have 30 hours on a T4. The 4090 from my dad should arrive next week. The plan is to use the T4 for trajectory SFT, then the 4090 for merging, Terminal Bench evaluation, and RLVR. Status: 474 out of 474 traces done,"}, {"post_id": 42823, "ts": "2026-08-07T19:01:50Z", "hours": 8.9192, "text": "Devlog #14 worked a lot, still not done we worked a lot this week. honestly the main takeaway is that the fable5 sft was good but its traces alone weren’t enough the model needs way more verified agent trajectories, so we’ve been grinding those out (target is 450). we also ran out of credits mid-generation (the modal endpoint started failing with 503s) and had to switch to a second kimi account. took a minute to get the auth right the new endpoint wanted an extra header we didn’t have. after that the generator resumed fine and it’s been adding verified traces since (~340/450 now). biggest scare: we found out peft 0.20 silently drops the second adapter when you merge two at once. tested it and yeah, the outputs were identical with one or two adapters so our trajectory sft would have been lost in the merge. fixed it by merging sequentially (base first, then the second adapter). checked against the live model and it matches, so we’re good. my 2070 couldn’t fit the 9b model for real training, so we’re waiting on a 4090 (my dad’s) next week. all the scripts are ready sft, merge, rlvr, eval, even a setup script for the gpu box. just need the hardware. status: 340/450 traces, everything ready, waiting on the 4090. we’ve worked a lot, but the real run hasn’t happened yet :/."}, {"post_id": 41470, "ts": "2026-08-05T18:10:35Z", "hours": 18.5394, "text": "Devlog #13 we worked ALOT so this was the grind day. 18 hours of training babysitting and honestly it felt like 40. the modal training kept dying every hour. the client would crash and take the whole run with it. we restarted like 3 times before figuring out modal run -d (detached) keeps it alive on their servers even if our computer shits the bed. tested it by literally killing the process and it kept going. ez fix but we worked ALOT to get there. then we ran out of credits at 93% and switched accounts, but the other account “finished” the run in 6 MILLISECONDS which was sus af. turns out it just re-uploaded the old checkpoint and trained nothing. the resume logic had a bug where it thought 6200 >= 3799 meant done. patched it. another lesson learned. also we made the model creative lol. added diversity sampling to the trajectory generator (5 solutions per task, keep the novel ones), a novelty bonus reward for RLVR, and a creativity rule in the self-awareness prompt. so the fat cat will be innovative, not just memorize. current state: SFT is done at 91% (good enough), resume bug fixed, phase 2 (trajectory SFT with creativity + self-awareness baked in) is up next. 18 hours. we worked ALOT. the fat cat will be worth it 😼"}, {"post_id": 40883, "ts": "2026-08-04T18:02:57Z", "hours": 10.0019, "text": "Devlog #12 training kept dying for the last 10 hours the training kept crashing. not the model, not the gp but the modal client. every hour or so it would lose connection and take the whole run down with it. we’d restart, it’d run an hour, crash again. three times. the fix was simple: modal run -d runs detached, so the training lives on modal’s servers and doesn’t care if our local process dies. killed the client on purpose to test — training kept going. that’s it, that’s the fix. also worried the loss wasn’t moving but it was fine. lr had wound down on the crashed run, looked scary, re-warmed fine after restart. loss is 0.668 now and dropping. status: 72% through the fable5 sft. checkpoints safe on hf. after this: trajectory sft with the self-awareness stuff baked in, then rlvr, then real numbers."}, {"post_id": 40601, "ts": "2026-08-04T08:01:50Z", "hours": 11.4342, "text": "** Devlog #11 half way there! ** the training is going pretty well, I fixed a few bugs with the code and added documentation training is at 50%!"}, {"post_id": 40369, "ts": "2026-08-03T20:35:48Z", "hours": 8.8192, "text": "** Devlog #10 the full training is finally running ** ok so phase 1 full training is officially going. 160k rows of fable5 data on modal, resuming from the 12.8k rows we got on kaggle before the session cap killed us. we’re at ~27% and it should take like a day and a half total. big stuff that happened since last time: fast kernels actually work now. the model has hybrid linear-attention layers and transformers kept falling back to a slow torch path. the fix was: nvidia cuda-devel base image (needs nvcc to compile), installing flash-linear-attention[cuda] + causal-conv1d , and pinning torch 2.10. took a couple tries but now training is like 4x faster than it would be without them. had to split across two modal accounts. first account ran out of credits, so we set up a second profile and the script auto-resumes from the huggingface checkpoints when it dies. it’s been rock solid, it happens every 20% plus every hour it saves and uploads. lost zero progress to the account switch. we gave the model a personality. it’s called Le Gros Chaton (the fat cat) and we’re training self-awareness into it, not as a system prompt but baked into the weights. state-sheets ([STATE] goal/known/tried/failed/next), metacognition lines, and self-review at the end of tasks. the trajectory sft masks out the system prompt so the model learns to track its own state as behavior, not as a script it repeats. did some benchmark research too. base qwen3.5-9b scores 9.2% on terminal-bench 2.0. comparable 9b fine-tunes hit 24-28%. that’s our realistic target which is 2.5-3x the baseline. honestly at this point it’s just… waiting. the pipeline works, the checkpoints work, the account switching works. now the model just needs to finish cooking. then trajectory sft, then rlvr, then we actually see the benchmark numbers."}, {"post_id": 40029, "ts": "2026-08-03T11:45:18Z", "hours": 13.3375, "text": "# Devlog #9 — finally training, and now with checkpoints! so last time i was stuck debugging phase 1/2 on kaggle. well, still debugging lol, but we actually got somewhere. the first issue was transformers not knowing what the qwen3.5 model was, had to pin transformers 5.14.1. then tokenizers kept fighting with it (0.22.0 wants old huggingface-hub, 0.22.1 is fine). then torch on kaggle kept giving a p100 which modern pytorch just refuses to run on, so we had to tell it to give us a t4 instead. then the fun one: out of memory on the very first step. batch 2 was too big for the gpu they gave us, so batch 1 it is. also i learned the hard way that kaggle kills your session after 12 hours. the first long run got cancelled at 80% done and i had nothing saved. so now the training script saves a checkpoint every 20% + every hour and uploads it to huggingface, so if it dies we just resume. it actually saved us, the run got cut and we picked right back up from the checkpoint. right now the full 160k sft is running on modal (paying for gpus now, kaggle time ran out). it’s at like 9% and should take a couple days. we’ll see how it goes, and then rlvr after that. the end of the errors… might actually be in sight this time."}, {"post_id": 38292, "ts": "2026-07-31T16:38:26Z", "hours": 5.4886, "text": "Devlog #8 — phase 2 fixes I’ve been getting a lot of issues with the first two for this model. I’ve gotten a lot of errors in Kaggle and Colab. I’ve tried everything, and I’m still fixing the code. It’s taking a good amount of time, but I think I’m almost there. I’ve gone through a lot of errors, and I’m probably right at the end of the errors. I might be able to start actually training it tomorrow, but for now I’ll stop."}, {"post_id": 38079, "ts": "2026-07-31T09:21:05Z", "hours": 10.3872, "text": "Devlog #7 — phase 2 running phase 1 (sft) is done and phase 2 (rlvr) is running now on kaggle. this is the part where the model actually gets trained on solving coding problems with our verifier giving rewards, not just imitating fable data. honestly this is the part i care about more. sft just makes it mimic. rlvr is where it learns to actually fix code. grpo with proportional rewards = the model gets partial credit for getting some tests passing, which is way better than all-or-nothing. while it runs im just waiting tbh. not much to build. the pipeline works, the eval works, the agent works. next after this: full benchmark run. humaneval, swebench, agentic eval. we’ll see if all this actually made the model good at coding or if i just spent a week on a worse qwen lol."}, {"post_id": 37100, "ts": "2026-07-29T16:05:47Z", "hours": 8.4747, "text": "Devlog #6 — training done (phase 1) finally got the qwen finetune to actually run with 625 steps, took like 8 hours on kaggle. it finished. havent run full benchmarks yet cause i was busy building other stuff while it trained. but it trained. thats the main thing. I also built a swebench agent that can actually navigate repos and edit files and make patches. not just the basic agent loop that runs commands. this one actually works for real se tasks. also added swebench eval so we can measure if the model is actually good at fixing bugs and coding then i’ll run the benchmarks on the trained model and see if its actually any good lol."}, {"post_id": 36367, "ts": "2026-07-28T13:16:16Z", "hours": 13.6447, "text": "Devlog #5 — ok so I’m not building a 10B MoE anymore So last time I had this whole plan. 10B MoE, from scratch, SwiGLU, GQA, all that. And I actually built it. Like 60 Python files, tests passing, RL pipeline, self-play data gen, the whole thing. It was kinda sick ngl. But then I was like… wait. I’m building an ENGINE that TRAINS coding agents. Not the agent. And the engine is cool but training a 10B model from zero on $30 of Modal credits is not how you beat GLM-5.2. You know what beats GLM-5.2? Fine-tuning a model that already knows how to code. the pivot (again) Qwen3.5-9B came out. 9B params. Fits on a Kaggle T4 with 4-bit QLoRA. And Kaggle gives me 30 HOURS of free GPU. 2× T4. That’s literally free training. So I had to adapt the entire codebase to work with HuggingFace models instead of my custom GPT. Made like 5 new files. All the verifier/agent/reward stuff I built for the MoE? Now works with Qwen too. Didn’t have to rewrite everything — just slapped a Qwen adapter on top. the dataset Found this dataset on HF: Nexlab/fable5-agentic-coding-sft . 160K rows of Claude Fable 5 output. Like actual coding traces — building games, fixing bugs, writing shell scripts. Not just “write a function that passes a test.” Real agentic stuff. Training right now actually. Qwen3.5-9B on 10K rows. Should take like 8 hours on 2× T4. We’ll see if it doesn’t OOM again lol. While waiting for training I read papers: Routing-Free MoE — what if experts just decide for themselves when to activate? No router, no top-k. Already implemented it in model.py. GLM-5.2 has this IndexShare thing that reuses computation across layers. 2.9× cheaper at 1M context. Maybe later. DeepSWE trained Qwen3-32B with RL only and got 42% SWE-Bench. So our approach is valid. the first phase of training is going on right now, i’ll post another devlog once it’s done!"}, {"post_id": 20759, "ts": "2026-07-02T13:35:48Z", "hours": 5.6175, "text": "Devlog #4 — wait, why am I training on Wikipedia? Last week I got le gros chaton (my from-scratch transformer) to actually train. Ran it on wikitext-103 in a Colab notebook on a free T4, 12k steps, val loss hit 3.73. Felt insane for something I wrote myself from PyTorch primitives. But then I looked at what it generated. It could do English-ish. It knew about Singapore and stuff. The moment I asked it to write code? Total garbage. Fair enough. It has never seen a single line of code. the pivot I thought about what I actually want this thing to do. Answer: a coding agent. Something that mogs on terminal-bench, that you give a task and it runs commands, fixes its own bugs, gets it done. Qwen-coder territory. Wikitext won’t get me there. You can’t learn Python from Wikipedia articles about the snake. So the plan changed. The big project is now le fat chaton, the bigger sibling. And it’s a coder. the architecture stuff I went down a rabbit hole on what frontier coding models do. They’re basically all mixture-of-experts (MoE) now. Qwen3-coder is 480B params / 35B active. Deepseek-coder-v2-lite is 16B / 2.4B active. The trick: big total params (knows a lot) small active params per token (runs fast and cheap) Runs like a small model, knows like a big one. Exactly what I want for a snappy terminal agent. So I rewrote the model. Added an MoE class. A gate routes each token to its top-2 experts out of 8, with a load-balance loss so the router doesn’t pick the same expert every time. Also threw in: SwiGLU — gated MLP, basically free quality, every Llama uses it GQA — smaller KV cache, faster long-context decoding (the agent harness needs this) Shared expert — DeepSeek-style, one always-on expert for common knowledge Fat config math: ~10.25B total / ~3.65B active. Not building that locally, it would crash my PC again. the data question Wikitext was a fine proof. For the coder I need code. Options: smollm-corpus python-edu — 7.6M educational Python files, high quality, but gated behind S3 (slow pull) starcoderdata python — ~50GB Python, streams fast from HF, the Codellama workhorse the-stack-v2 — 900B tokens, overkill for my budget Going with starcoderdata Python, blended with ~15-20% cosmopedia (prose) so it has general knowledge. That blend is the Qwen-coder recipe. am I warm-starting from wikitext? No. Different architecture entirely (17M dense vs 10B MoE, weights don’t fit). Even if they did, warm-starting a coder from prose wastes compute unlearning prose. Fresh random init on the mixed corpus. Clean slate. This went from “lets make a little model that talks” to “lets build a coding agent that mogs the big labs” real fast. Ambitious for a solo student with a 2070 and $30 of Modal credits? Absolutely. Doing it anyway? Yeah. Next: the eval harness so I can measure if this is making the model better at code (pass@1, pass@5 on HumanEval). Val loss going down means nothing if it still can’t write a"}, {"post_id": 20019, "ts": "2026-07-01T14:54:30Z", "hours": 0.3917, "text": "DEVLOG #3: teaching it to talk (fine-tuning) the base model was a completer. ask it “what is 2+2?” and it just continues the sentence. fine-tuning is what makes it actually answer: same model, more training, but on prompt/response pairs. i wrote ~48 Q&A pairs by hand. trivia, math, jokes, hellos. all in the same rigid Q: …\\nA: format. new trick: loss masking. only the response counts toward the loss, not the prompt, so i set prompt targets to -100 (pytorch’s cross_entropy ignores that automatically, zero model changes). tested the mask boundary in isolation first. wrong mask = 400 steps of training the wrong thing. the run: load base weights, 400 steps, lower lr. loss went 12.3 to 0.02.then the reality check. it nails everything in the set. capitals, math, jokes, “who are you,” verbatim. but something NOT in the set? slop. 17M params + 48 examples is a lookup table, not a reasoner. real chat models are billions of params. not a bug, just the scale.but it’s finally coherent. the first base model was word salad. now you type a question and get a real answer back that makes sense. it’s tiny and dumb and can’t generalize, but it actually talks. that’s the win. full pipeline done: tokenizer, model from scratch, pretrain on wikipedia, fine-tune, chat. mine, it talks, i understand every piece."}, {"post_id": 20006, "ts": "2026-07-01T14:16:25Z", "hours": 0.4139, "text": "DEVLOG #2: real data, real loss, and a lesson about memorizing split my data into 90% train / 10% val. before i only had one number, training loss, and it lies to you. you can’t tell learning from memorizing with just that. now i check both every 200 steps. rule: both go down = learning. train down but val up = memorizing = useless.also made the model bigger, 7M -> 17.7M params. immediately ran out of gpu memory lol. dropped batch size to 32 to fit my 8GB card.training: 3000 steps. loss started at 158 (should be ~10 for random , my first pass blew up and made garbage). recovered fast though, didn’t ruin the run. right now the llm is spitting out gibberish and random words but after finetuning it should sound more coherent"}, {"post_id": 19989, "ts": "2026-07-01T13:48:37Z", "hours": 36.31, "text": "DEVLOG #1: wiki text training so i finally got the model to train on actual data instead of that one paragraph about singapore nature reserves lol. before this the whole thing was basically a joke, i trained a 7M parameter transformer on like 134 tokens and it just spat out jumbled fragments of the same sentence over and over. cool proof it worked but definitely not a “language model” in any real sense.anyway. the big move this week was swapping out the tiny json file for wikitext-2. that’s a huggingface dataset of wikipedia text, about 2.4 million tokens once you filter out the empty lines. that’s… a lot more than 134 lol. encoding it took maybe half a minute with tiktoken and the whole tensor fits on my gpu no problem (it’s only like 8MB of ints)."}]...
|
Title: le-fat-chaton by @unknown_lizzard | Stardan Title: le-fat-chaton by @unknown_lizzard | Stardance
Devlogs:
DEVLOG #1: wiki text training so i finally got the model to train on actual data instead of that one paragraph about singapore nature reserves lol. before this the whole thing was basically a joke, i trained a 7M parameter transformer on like 134 tokens and it just spat out jumbled fragments of the same sentence over and over. cool proof it worked but definitely not a “language model” in any real sense.anyway. the big move this week was swapping out the tiny json file for wikitext-2. that’s a huggingface dataset of wikipedia text, about 2.4 million tokens once you filter out the empty lines. that’s… a lot more than 134 lol. encoding it took maybe half a minute with tiktoken and the whole tensor fits on my gpu no problem (it’s only like 8MB of ints).
DEVLOG #2: real data, real loss, and a lesson about memorizing split my data into 90% train / 10% val. before i only had one number, training loss, and it lies to you. you can’t tell learning from memorizing with just that. now i check both every 200 steps. rule: both go down = learning. train down but val up = memorizing = useless.also made the model bigger, 7M -> 17.7M params. immediately ran out of gpu memory lol. dropped batch size to 32 to fit my 8GB card.training: 3000 steps. loss started at 158 (should be ~10 for random , my first pass blew up and made garbage). recovered fast though, didn’t ruin the run. right now the llm is spitting out gibberish and random words but after finetuning it should sound more coherent
DEVLOG #3: teaching it to talk (fine-tuning) the base model was a completer. ask it “what is 2+2?” and it just continues the sentence. fine-tuning is what makes it actually answer: same model, more training, but on prompt/response pairs. i wrote ~48 Q&A pairs by hand. trivia, math, jokes, hellos. all in the same rigid Q: …\nA: format. new trick: loss masking. only the response counts toward the loss, not the prompt, so i set prompt targets to -100 (pytorch’s cross_entropy ignores that automatically, zero model changes). tested the mask boundary in isolation first. wrong mask = 400 steps of training the wrong thing. the run: load base weights, 400 steps, lower lr. loss went 12.3 to 0.02.then the reality check. it nails everything in the set. capitals, math, jokes, “who are you,” verbatim. but something NOT in the set? slop. 17M params + 48 examples is a lookup table, not a reasoner. real chat models are billions of params. not a bug, just the scale.but it’s finally coherent. the first base model was word salad. now you type a question and get a real answer back that makes sense. it’s tiny and dumb and can’t generalize, but it actually talks. that’s the win. full pipeline done: tokenizer, model from scratch, pretrain on wikipedia, fine-tune, chat. mine, it talks, i understand every piece.
Devlog #4 — wait, why am I training on Wikipedia? Last week I got le gros chaton (my from-scratch transformer) to actually train. Ran it on wikitext-103 in a Colab notebook on a free T4, 12k steps, val loss hit 3.73. Felt insane for something I wrote myself from PyTorch primitives. But then I looked at what it generated. It could do English-ish. It knew about Singapore and stuff. The moment I asked it to write code? Total garbage. Fair enough. It has never seen a single line of code. the pivot I thought about what I actually want this thing to do. Answer: a coding agent. Something that mogs on terminal-bench, that you give a task and it runs commands, fixes its own bugs, gets it done. Qwen-coder territory. Wikitext won’t get me there. You can’t learn Python from Wikipedia articles about the snake. So the plan changed. The big project is now le fat chaton, the bigger sibling. And it’s a coder. the architecture stuff I went down a rabbit hole on what frontier coding models do. They’re basically all mixture-of-experts (MoE) now. Qwen3-coder is 480B params / 35B active. Deepseek-coder-v2-lite is 16B / 2.4B active. The trick: big total params (knows a lot) small active params per token (runs fast and cheap) Runs like a small model, knows like a big one. Exactly what I want for a snappy terminal agent. So I rewrote the model. Added an MoE class. A gate routes each token to its top-2 experts out of 8, with a load-balance loss so the router doesn’t pick the same expert every time. Also threw in: SwiGLU — gated MLP, basically free quality, every Llama uses it GQA — smaller KV cache, faster long-context decoding (the agent harness needs this) Shared expert — DeepSeek-style, one always-on expert for common knowledge Fat config math: ~10.25B total / ~3.65B active. Not building that locally, it would crash my PC again. the data question Wikitext was a fine proof. For the coder I need code. Options: smollm-corpus python-edu — 7.6M educational Python files, high quality, but gated behind S3 (slow pull) starcoderdata python — ~50GB Python, streams fast from HF, the Codellama workhorse the-stack-v2 — 900B tokens, overkill for my budget Going with starcoderdata Python, blended with ~15-20% cosmopedia (prose) so it has general knowledge. That blend is the Qwen-coder recipe. am I warm-starting from wikitext? No. Different architecture entirely (17M dense vs 10B MoE, weights don’t fit). Even if they did, warm-starting a coder from prose wastes compute unlearning prose. Fresh random init on the mixed corpus. Clean slate. This went from “lets make a little model that talks” to “lets build a coding agent that mogs the big labs” real fast. Ambitious for a solo student with a 2070 and $30 of Modal credits? Absolutely. Doing it anyway? Yeah. Next: the eval harness so I can measure if this is making the model better at code (pass@1, pass@5 on HumanEval). Val loss going down means nothing if it still can’t write a
Devlog #5 — ok so I’m not building a 10B MoE anymore So last time I had this whole plan. 10B MoE, from scratch, SwiGLU, GQA, all that. And I actually built it. Like 60 Python files, tests passing, RL pipeline, self-play data gen, the whole thing. It was kinda sick ngl. But then I was like… wait. I’m building an ENGINE that TRAINS coding agents. Not the agent. And the engine is cool but training a 10B model from zero on $30 of Modal credits is not how you beat GLM-5.2. You know what beats GLM-5.2? Fine-tuning a model that already knows how to code. the pivot (again) Qwen3.5-9B came out. 9B params. Fits on a Kaggle T4 with 4-bit QLoRA. And Kaggle gives me 30 HOURS of free GPU. 2× T4. That’s literally free training. So I had to adapt the entire codebase to work with HuggingFace models instead of my custom GPT. Made like 5 new files. All the verifier/agent/reward stuff I built for the MoE? Now works with Qwen too. Didn’t have to rewrite everything — just slapped a Qwen adapter on top. the dataset Found this dataset on HF: Nexlab/fable5-agentic-coding-sft . 160K rows of Claude Fable 5 output. Like actual coding traces — building games, fixing bugs, writing shell scripts. Not just “write a function that passes a test.” Real agentic stuff. Training right now actually. Qwen3.5-9B on 10K rows. Should take like 8 hours on 2× T4. We’ll see if it doesn’t OOM again lol. While waiting for training I read papers: Routing-Free MoE — what if experts just decide for themselves when to activate? No router, no top-k. Already implemented it in model.py. GLM-5.2 has this IndexShare thing that reuses computation across layers. 2.9× cheaper at 1M context. Maybe later. DeepSWE trained Qwen3-32B with RL only and got 42% SWE-Bench. So our approach is valid. the first phase of training is going on right now, i’ll post another devlog once it’s done!
Devlog #6 — training done (phase 1) finally got the qwen finetune to actually run with 625 steps, took like 8 hours on kaggle. it finished. havent run full benchmarks yet cause i was busy building other stuff while it trained. but it trained. thats the main thing. I also built a swebench agent that can actually navigate repos and edit files and make patches. not just the basic agent loop that runs commands. this one actually works for real se tasks. also added swebench eval so we can measure if the model is actually good at fixing bugs and coding then i’ll run the benchmarks on the trained model and see if its actually any good lol.
Devlog #7 — phase 2 running phase 1 (sft) is done and phase 2 (rlvr) is running now on kaggle. this is the part where the model actually gets trained on solving coding problems with our verifier giving rewards, not just imitating fable data. honestly this is the part i care about more. sft just makes it mimic. rlvr is where it learns to actually fix code. grpo with proportional rewards = the model gets partial credit for getting some tests passing, which is way better than all-or-nothing. while it runs im just waiting tbh. not much to build. the pipeline works, the eval works, the agent works. next after this: full benchmark run. humaneval, swebench, agentic eval. we’ll see if all this actually made the model good at coding or if i just spent a week on a worse qwen lol.
Devlog #8 — phase 2 fixes I’ve been getting a lot of issues with the first two for this model. I’ve gotten a lot of errors in Kaggle and Colab. I’ve tried everything, and I’m still fixing the code. It’s taking a good amount of time, but I think I’m almost there. I’ve gone through a lot of errors, and I’m probably right at the end of the errors. I might be able to start actually training it tomorrow, but for now I’ll stop.
# Devlog #9 — finally training, and now with checkpoints! so last time i was stuck debugging phase 1/2 on kaggle. well, still debugging lol, but we actually got somewhere. the first issue was transformers not knowing what the qwen3.5 model was, had to pin transformers 5.14.1. then tokenizers kept fighting with it (0.22.0 wants old huggingface-hub, 0.22.1 is fine). then torch on kaggle kept giving a p100 which modern pytorch just refuses to run on, so we had to tell it to give us a t4 instead. then the fun one: out of memory on the very first step. batch 2 was too big for the gpu they gave us, so batch 1 it is. also i learned the hard way that kaggle kills your session after 12 hours. the first long run got cancelled at 80% done and i had nothing saved. so now the training script saves a checkpoint every 20% + every hour and uploads it to huggingface, so if it dies we just resume. it actually saved us, the run got cut and we picked right back up from the checkpoint. right now the full 160k sft is running on modal (paying for gpus now, kaggle time ran out). it’s at like 9% and should take a couple days. we’ll see how it goes, and then rlvr after that. the end of the errors… might actually be in sight this time.
** Devlog #10 the full training is finally running ** ok so phase 1 full training is officially going. 160k rows of fable5 data on modal, resuming from the 12.8k rows we got on kaggle before the session cap killed us. we’re at ~27% and it should take like a day and a half total. big stuff that happened since last time: fast kernels actually work now. the model has hybrid linear-attention layers and transformers kept falling back to a slow torch path. the fix was: nvidia cuda-devel base image (needs nvcc to compile), installing flash-linear-attention[cuda] + causal-conv1d , and pinning torch 2.10. took a couple tries but now training is like 4x faster than it would be without them. had to split across two modal accounts. first account ran out of credits, so we set up a second profile and the script auto-resumes from the huggingface checkpoints when it dies. it’s been rock solid, it happens every 20% plus every hour it saves and uploads. lost zero progress to the account switch. we gave the model a personality. it’s called Le Gros Chaton (the fat cat) and we’re training self-awareness into it, not as a system prompt but baked into the weights. state-sheets ([STATE] goal/known/tried/failed/next), metacognition lines, and self-review at the end of tasks. the trajectory sft masks out the system prompt so the model learns to track its own state as behavior, not as a script it repeats. did some benchmark research too. base qwen3.5-9b scores 9.2% on terminal-bench 2.0. comparable 9b fine-tunes hit 24-28%. that’s our realistic target which is 2.5-3x the baseline. honestly at this point it’s just… waiting. the pipeline works, the checkpoints work, the account switching works. now the model just needs to finish cooking. then trajectory sft, then rlvr, then we actually see the benchmark numbers.
** Devlog #11 half way there! ** the training is going pretty well, I fixed a few bugs with the code and added documentation training is at 50%!
Devlog #12 training kept dying for the last 10 hours the training kept crashing. not the model, not the gp but the modal client. every hour or so it would lose connection and take the whole run down with it. we’d restart, it’d run an hour, crash again. three times. the fix was simple: modal run -d runs detached, so the training lives on modal’s servers and doesn’t care if our local process dies. killed the client on purpose to test — training kept going. that’s it, that’s the fix. also worried the loss wasn’t moving but it was fine. lr had wound down on the crashed run, looked scary, re-warmed fine after restart. loss is 0.668 now and dropping. status: 72% through the fable5 sft. checkpoints safe on hf. after this: trajectory sft with the self-awareness stuff baked in, then rlvr, then real numbers.
Devlog #13 we worked ALOT so this was the grind day. 18 hours of training babysitting and honestly it felt like 40. the modal training kept dying every hour. the client would crash and take the whole run with it. we restarted like 3 times before figuring out modal run -d (detached) keeps it alive on their servers even if our computer shits the bed. tested it by literally killing the process and it kept going. ez fix but we worked ALOT to get there. then we ran out of credits at 93% and switched accounts, but the other account “finished” the run in 6 MILLISECONDS which was sus af. turns out it just re-uploaded the old checkpoint and trained nothing. the resume logic had a bug where it thought 6200 >= 3799 meant done. patched it. another lesson learned. also we made the model creative lol. added diversity sampling to the trajectory generator (5 solutions per task, keep the novel ones), a novelty bonus reward for RLVR, and a creativity rule in the self-awareness prompt. so the fat cat will be innovative, not just memorize. current state: SFT is done at 91% (good enough), resume bug fixed, phase 2 (trajectory SFT with creativity + self-awareness baked in) is up next. 18 hours. we worked ALOT. the fat cat will be worth it 😼
Devlog #14 worked a lot, still not done we worked a lot this week. honestly the main takeaway is that the fable5 sft was good but its traces alone weren’t enough the model needs way more verified agent trajectories, so we’ve been grinding those out (target is 450). we also ran out of credits mid-generation (the modal endpoint started failing with 503s) and had to switch to a second kimi account. took a minute to get the auth right the new endpoint wanted an extra header we didn’t have. after that the generator resumed fine and it’s been adding verified traces since (~340/450 now). biggest scare: we found out peft 0.20 silently drops the second adapter when you merge two at once. tested it and yeah, the outputs were identical with one or two adapters so our trajectory sft would have been lost in the merge. fixed it by merging sequentially (base first, then the second adapter). checked against the live model and it matches, so we’re good. my 2070 couldn’t fit the 9b model for real training, so we’re waiting on a 4090 (my dad’s) next week. all the scripts are ready sft, merge, rlvr, eval, even a setup script for the gpu box. just need the hardware. status: 340/450 traces, everything ready, waiting on the 4090. we’ve worked a lot, but the real run hasn’t happened yet :/.
Devlog 15 Traces are done The trace generator finally hit the target and went past it. We got 474 verified trajectories, zero malformed, zero duplicates, 19 templates, and a 95 percent self review pass rate. We were aiming for 450, so we ended up with 474. The dataset is synced to Hugging Face in two versions: raw and normalized. That part is finally done. It was the longest part of the project, and the model is only going to be as good as the traces, so getting this right was important The Kimi endpoint also stopped working. The old teacher account started returning 503 errors during generation and eventually died completely. We switched to a second Kimi account, which needed an extra header that we did not have at first. After figuring that out, generation resumed without losing any progress For GPUs, the 2070 still cannot realistically train the 9B. Kaggle hours are back, so we have 30 hours on a T4. The 4090 from my dad should arrive next week. The plan is to use the T4 for trajectory SFT, then the 4090 for merging, Terminal Bench evaluation, and RLVR. Status: 474 out of 474 traces done,
Devlog #16, t4gpu The T4 16GB fought us the whole way. BF16 was baked into a pre quantized checkpoint, which made each step take 668 seconds because it had to use FP32 emulation. 2K context ran out of memory, the 248K logits ran out of memory too, FP16 started producing NaNs, Kaggle session limits locked us out, and the paged optimizer crashed last. Every fix seemed to create another problem. We eventually got it working with FP16 compute, chunked assistant token loss, autocast, and plain AdamW. Then we got access to an AMD MI300X with 192GB of memory. No quantization, 16K context, and enough memory to train full traces instead of cutting them off halfway through a message. Right now the 16K BF16 SFT run is training on the MI300X. Next up is merging the model, running Terminal Bench, and then RLVR.
devlog #17 MI300X vLLM serving The old GPU box got destroyed, so I spun up a new MI300X with the same SSH key. ROCm 6.2 and Torch 2.5.1 were already set up, so getting the machine ready was easy. The hard part was vLLM. The pip version is CUDA only, and the ROCm Docker image only supported the multimodal version of Qwen3.5. Our model is text-only, so I had to patch the image. I added Qwen3_5ForCausalLM, fixed the hybrid KV-cache setup, added the missing Mamba state methods, and mapped the weight names so the text-only safetensors loaded correctly. I also patched the merged config to remove the M-RoPE and dtype fields. After that, the model served normally. I ran a one-task test with fix-git. The pipeline worked, but the model got stuck repeating git show until it ran out of turns. ##Shipped The merged model is now public on Hugging Face as mateo0093/le-gros-chaton under Apache-2.0. The 5×5 Terminal-Bench pilot finished at 3/25 (25%). All three successes were on fix-git, which shows that the SFT data was too focused on git tasks. I also tested RLVR with GRPO and a novelty bonus. The reward had almost no variance, so the training signal was basically useless. I saved the step-10 adapter but didn’t merge it. The GPU used about 12 hours of the 50 hour budget. All the vLLM patches and harness changes are in the repo. Main lesson: Qwen3.5’s hybrid model is a bit annoying to serve as text-only on ROCm. The model works but i think this was more of a learning opportunity, i know what i’ve done wrong and what i would have done differently, I will be working on a new model in a few weeks once I do a bit more research.
Ships:
at first I wanted to make a normal text model to try and learn more about llms, but I wanted to go further, agentic coding is a big thing so what better way to learn about llms than trying to make an agentic coding ai model. So thats how le gros chaton was born, the name originated from a joke on x about a fake huge mistral model coming out. to train this model I used already made datasets on hugging face, like fable5 traces, I also used kimi k3 as a teacher model to teach my smaller model to be good at coding and running commands, it kind of worked? but i’m not completely satisfied with the result, but at least i have learned some things and soon enough I will trying making an even better model! I dont have a demo url for you guys to try the model out since it was too expensive to run on a gpu so you’ll need to run it yourself, check it out on hugging face, follow the instructions and have fun....
|
1792878632.7358682
|
1793742632.7358682
|
|
33545
|
Frictionless by @unknown_lizzard | Stardance
|
unknown_lizzard
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: Frictionless by @unknown_lizzard | Stardanc Title: Frictionless by @unknown_lizzard | Stardance...
|
1792929189.762941
|
1793793189.762941
|
|
41038
|
My own site! by @unknown_lizzard | Stardance
|
unknown_lizzard
|
1
|
0.0
|
0
|
[{"ship_num": 1, "ts": "2026-0 [{"ship_num": 1, "ts": "2026-07-28T14:03:05Z", "hours": 0.0, "multiplier": null, "payout": 0, "blessing": "neutral", "text": "i made my own personal site, i’m very proud of the css(w3scool) helped me alot, the site is live!", "repo": "https://github.com/Mateooo93/my-website", "demo": "https://mateooo93.github.io/my-website/"}]...
|
[{"post_id": 36395, "ts": "202 [{"post_id": 36395, "ts": "2026-07-28T14:01:07Z", "hours": 0.4103, "text": "my personal site! this is my own personal site, its very simple and uses high quality css elements!, it includes an about me, and some links to my other github projects."}]...
|
Title: My own site! by @unknown_lizzard | Stardanc Title: My own site! by @unknown_lizzard | Stardance
Devlogs:
my personal site! this is my own personal site, its very simple and uses high quality css elements!, it includes an about me, and some links to my other github projects.
Ships:
i made my own personal site, i’m very proud of the css(w3scool) helped me alot, the site is live!...
|
1792878567.9262872
|
1793742567.9262872
|
|
41301
|
My awesome macropad! by @unknown_lizzard | Stardan My awesome macropad! by @unknown_lizzard | Stardance...
|
unknown_lizzard
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: My awesome macropad! by @unknown_lizzard | Title: My awesome macropad! by @unknown_lizzard | Stardance...
|
1792939816.7096994
|
1793803816.7096994
|
|
55612
|
The Terminal by @unknown_lizzard | Stardance
|
unknown_lizzard
|
0
|
26.47
|
0
|
[]
|
[{"post_id": 59246, "ts": "202 [{"post_id": 59246, "ts": "2026-08-31T22:41:06Z", "hours": 9.7383, "text": "DevLog #4 Lore and Endings spent this one on the actual game instead of the look of it. the filesystem is full of real lore now instead of placeholder text, and there are hidden folders you can only get into if you already know the name. also got the prediction engine working properly. the ai predicts what file youre about to open and it shows up in /predictions. if you read something else it hits you with PREDICTION FAILED and the confidence drops, which is basically the whole thing. there are four endings now. one where you do what it says, one where you keep breaking the predictions, one where you find voss’s real message (its encrypted so you have to decode it yourself), and a secret one where you type null twice. the game remembers how many times youve connected now too. player.dat said previous contacts 1 the first run and it goes up every time even after you close the tab. creeped me out a little honestly. at one point i had the terminal inside a little window with a titlebar but it looked wrong for the vibe so i took it off and just kept the colors. next i want to add the ai saying stuff when youre idle and some more variety in the predictions so it doesnt feel so samey."}, {"post_id": 58843, "ts": "2026-08-31T12:56:24Z", "hours": 5.8769, "text": "DevLog #3 Terminal Polish spent a bit of time polishing the appearence of the terminal, now the little location thing on the left is green so the outputs are seperated more cleanly I also made the little cursor/arrow blink , like in a real terminal and made some other little changes at some point i had issue with the terminal not rendering correctly but I fixed it with a center div for the next devlog i’ll be working on the lore and the progression of the game, it still needs a bit of work"}, {"post_id": 57375, "ts": "2026-08-29T14:36:35Z", "hours": 9.3203, "text": "DevLog #2 The Lore this part took me alot of time, i ahd to come up with some sort of story for this puzzle, I won’t spoil it for you but i think you guys will love it, I also spent some time making the cat commands refining some thing and fixing some bugs one of the bugs i encountered was the file path not appearing correctly when using ls, It was pretty easy i just had to include in every argument because or else it wouldn’t show up every time. I also made the text files with content inside of them and little hints that will allow you to solve this puzzle! Tonight i’ll work a bit on the appearence of the terminal it looks a bit boring rn"}, {"post_id": 56914, "ts": "2026-08-28T22:43:38Z", "hours": 1.5314, "text": "Devlog #1 - The Terminal This is gonna be the first game i’m making on stardance. I’ll try to no give too many hints or reveal anything on the devlogs, so for this game, you’ll be in the terminal and try to solve a puzzle by navigating through the filesystem and looking at the different files. I spent about 1 hour and a half making the terminal and the filesystem, i cant say it was easy but with a bit of research I managed to get multiple commands working like ls and cd. I’m also still working on the lore and story and I think you guys will love it!"}]...
|
Title: The Terminal by @unknown_lizzard | Stardanc Title: The Terminal by @unknown_lizzard | Stardance
Devlogs:
Devlog #1 - The Terminal This is gonna be the first game i’m making on stardance. I’ll try to no give too many hints or reveal anything on the devlogs, so for this game, you’ll be in the terminal and try to solve a puzzle by navigating through the filesystem and looking at the different files. I spent about 1 hour and a half making the terminal and the filesystem, i cant say it was easy but with a bit of research I managed to get multiple commands working like ls and cd. I’m also still working on the lore and story and I think you guys will love it!
DevLog #2 The Lore this part took me alot of time, i ahd to come up with some sort of story for this puzzle, I won’t spoil it for you but i think you guys will love it, I also spent some time making the cat commands refining some thing and fixing some bugs one of the bugs i encountered was the file path not appearing correctly when using ls, It was pretty easy i just had to include in every argument because or else it wouldn’t show up every time. I also made the text files with content inside of them and little hints that will allow you to solve this puzzle! Tonight i’ll work a bit on the appearence of the terminal it looks a bit boring rn
DevLog #3 Terminal Polish spent a bit of time polishing the appearence of the terminal, now the little location thing on the left is green so the outputs are seperated more cleanly I also made the little cursor/arrow blink , like in a real terminal and made some other little changes at some point i had issue with the terminal not rendering correctly but I fixed it with a center div for the next devlog i’ll be working on the lore and the progression of the game, it still needs a bit of work
DevLog #4 Lore and Endings spent this one on the actual game instead of the look of it. the filesystem is full of real lore now instead of placeholder text, and there are hidden folders you can only get into if you already know the name. also got the prediction engine working properly. the ai predicts what file youre about to open and it shows up in /predictions. if you read something else it hits you with PREDICTION FAILED and the confidence drops, which is basically the whole thing. there are four endings now. one where you do what it says, one where you keep breaking the predictions, one where you find voss’s real message (its encrypted so you have to decode it yourself), and a secret one where you type null twice. the game remembers how many times youve connected now too. player.dat said previous contacts 1 the first run and it goes up every time even after you close the tab. creeped me out a little honestly. at one point i had the terminal inside a little window with a titlebar but it looked wrong for the vibe so i took it off and just kept the colors. next i want to add the ai saying stuff when youre idle and some more variety in the predictions so it doesnt feel so samey....
|
1792889466.2863772
|
1793753466.2863772
|
|
59537
|
Evolution Simulator by @unknown_lizzard | Stardanc Evolution Simulator by @unknown_lizzard | Stardance...
|
unknown_lizzard
|
0
|
21.7
|
0
|
[]
|
[{"post_id": 66198, "ts": "202 [{"post_id": 66198, "ts": "2026-09-11T21:22:39Z", "hours": 4.1592, "text": "DevLog #3 predators and the paint * Predators are in. Aggression is now a spectrum instead of a simple on or off: every hungry creature makes a little gamble so they hunt weaker neighbours with probability equal to its aggression, or keep grazing plants. a pure herbivore never hunts, a full carnivore almost always does, and everything in between is an opportunist that does a bit of both. the fun part is the catch. you can only eat prey you can actually outrun while the prey sprints away, so it is a real speed race. prey evolve to run faster to escape, predators evolve to keep up, and you can watch both drift up in the sparklines over time. that is the arms race and it finally gives the speed trait a reason to exist instead of just taxing your metabolism. it took way longer than the food system. my first version was a hard cutoff and it collapsed the whole population into hyper-carnivores in under two minutes every single time, no matter how i balanced it. making aggression a continuum with real costs and gating the catch on outrunning prey is what finally made it work. there was also a sneaky bug where the creatures were starving to death on a plate completely covered in food, because the hunting logic was stomping the food-scan timer - one fixed line and it became a real ecosystem. and everything glows now. the old flat dots are gone, replaced by soft antialiased cells that go green to amber to red as they get dangerous, so you can literally watch predators appear as the plate changes colour."}, {"post_id": 66044, "ts": "2026-09-11T16:33:51Z", "hours": 13.6631, "text": "DevLog #2 Food and reproduction The ecosystem now has a full food loop: creatures sense food chase it, eat it, and they breed. food spawns continuously but it is capped, because the world is finite and the plate should run out when a population gets too big. Creatures sense food with their vision trait range 30 to 150 px on the wrapped world, and they only rescan when they are hungry and their target is stale, so a starving population cant waste cycles scanning every tick. Hungry ones steer toward the nearest food and the turn rate drops with size, so big creatures are clumsy hunters. The first version had a problem I only noticed after watching it run: nothing looked like it was happening. Every creature spawned full, so the first 40 seconds were pure wandering while food just piled up, and the food was the same green as the creatures, so the eating was invisible. The fix: the world now starts pre-seeded with food, creatures spawn partially hungry so the hunting begins at t=0, and the food is amber so you can actually watch it get eaten. reproduction Mating is local: a ready creature only breeds with a ready neighbor inside a small mating radius, so the gene pool picks up spatial structure instead of any-two-random mixing. Both parents pay an energy cost and the baby spawns at their midpoint with a genome that is a per-trait crossover of the two parents, followed by a gaussian nudge on each trait with the probability set by the new mutation slider in the header (0.1% to 20%). The fertility trait controls how fast readiness builds up, so lazy genomes breed slower, and generations are tracked per creature with the highest one alive shown in the header. Finding a partner is a neighbor query, so I added a spatial hash grid: rebuilt every tick, and a query only touches the cells a circle overlaps, so it stays cheap no matter how big the population gets. Predators will run on the same grid. There is also a little sparkline panel in the bottom right now (population and average speed over time) the first step toward the real stats milestone. NEXT next im gonna make predators! the aggression trait finally gets used so hunters that chase the slow, and prey that learn to run. and i want to tune the world until the evolution is obvious in the sparklines, speed and size drifting on their own!"}, {"post_id": 64771, "ts": "2026-09-09T12:53:21Z", "hours": 3.8747, "text": "DevLog #1 The ecosystem EvoLab is an interactive artificial-life simulator: virtual organisms fight for resources, they reproduce, they mutate, and evolve, and the user watches natural selection happen in real time. framework I first scaffolded the project in React + TypeScript + Vite, then decided to move the whole thing to Python. One language for the whole project, pygame gives me a full-screen canvas and click/mouse handling for the UI, and Python keeps the door open for proper numeric tooling when I start analyzing population statistics NEXT next im gonna make the Food system! So creatures can eat them and reproduce and evolve! aswell as mutations so we get diversity!"}]...
|
Title: Evolution Simulator by @unknown_lizzard | S Title: Evolution Simulator by @unknown_lizzard | Stardance
Devlogs:
DevLog #1 The ecosystem EvoLab is an interactive artificial-life simulator: virtual organisms fight for resources, they reproduce, they mutate, and evolve, and the user watches natural selection happen in real time. framework I first scaffolded the project in React + TypeScript + Vite, then decided to move the whole thing to Python. One language for the whole project, pygame gives me a full-screen canvas and click/mouse handling for the UI, and Python keeps the door open for proper numeric tooling when I start analyzing population statistics NEXT next im gonna make the Food system! So creatures can eat them and reproduce and evolve! aswell as mutations so we get diversity!
DevLog #2 Food and reproduction The ecosystem now has a full food loop: creatures sense food chase it, eat it, and they breed. food spawns continuously but it is capped, because the world is finite and the plate should run out when a population gets too big. Creatures sense food with their vision trait range 30 to 150 px on the wrapped world, and they only rescan when they are hungry and their target is stale, so a starving population cant waste cycles scanning every tick. Hungry ones steer toward the nearest food and the turn rate drops with size, so big creatures are clumsy hunters. The first version had a problem I only noticed after watching it run: nothing looked like it was happening. Every creature spawned full, so the first 40 seconds were pure wandering while food just piled up, and the food was the same green as the creatures, so the eating was invisible. The fix: the world now starts pre-seeded with food, creatures spawn partially hungry so the hunting begins at t=0, and the food is amber so you can actually watch it get eaten. reproduction Mating is local: a ready creature only breeds with a ready neighbor inside a small mating radius, so the gene pool picks up spatial structure instead of any-two-random mixing. Both parents pay an energy cost and the baby spawns at their midpoint with a genome that is a per-trait crossover of the two parents, followed by a gaussian nudge on each trait with the probability set by the new mutation slider in the header (0.1% to 20%). The fertility trait controls how fast readiness builds up, so lazy genomes breed slower, and generations are tracked per creature with the highest one alive shown in the header. Finding a partner is a neighbor query, so I added a spatial hash grid: rebuilt every tick, and a query only touches the cells a circle overlaps, so it stays cheap no matter how big the population gets. Predators will run on the same grid. There is also a little sparkline panel in the bottom right now (population and average speed over time) the first step toward the real stats milestone. NEXT next im gonna make predators! the aggression trait finally gets used so hunters that chase the slow, and prey that learn to run. and i want to tune the world until the evolution is obvious in the sparklines, speed and size drifting on their own!
DevLog #3 predators and the paint * Predators are in. Aggression is now a spectrum instead of a simple on or off: every hungry creature makes a little gamble so they hunt weaker neighbours with probability equal to its aggression, or keep grazing plants. a pure herbivore never hunts, a full carnivore almost always does, and everything in between is an opportunist that does a bit of both. the fun part is the catch. you can only eat prey you can actually outrun while the prey sprints away, so it is a real speed race. prey evolve to run faster to escape, predators evolve to keep up, and you can watch both drift up in the sparklines over time. that is the arms race and it finally gives the speed trait a reason to exist instead of just taxing your metabolism. it took way longer than the food system. my first version was a hard cutoff and it collapsed the whole population into hyper-carnivores in under two minutes every single time, no matter how i balanced it. making aggression a continuum with real costs and gating the catch on outrunning prey is what finally made it work. there was also a sneaky bug where the creatures were starving to death on a plate completely covered in food, because the hunting logic was stomping the food-scan timer - one fixed line and it became a real ecosystem. and everything glows now. the old flat dots are gone, replaced by soft antialiased cells that go green to amber to red as they get dangerous, so you can literally watch predators appear as the plate changes colour....
|
1792939879.5347345
|
1793803879.5347345
|
|
14318
|
My very own os by @unknown_coder | Stardance
|
unknown_coder
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: My very own os by @unknown_coder | Stardanc Title: My very own os by @unknown_coder | Stardance...
|
1792795779.5772157
|
1793400579.5772157
|
|
63095
|
SajiloGhar by @unisha | Stardance
|
unisha
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: SajiloGhar by @unisha | Stardance
|
1793048039.8871334
|
1793652839.8871334
|
|
68231
|
lucrical by @uniquepersun | Stardance
|
uniquepersun
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: lucrical by @uniquepersun | Stardance
|
1793119845.266268
|
1793724645.266268
|
|
41382
|
Cosmic Sticky Notes by @uniquebitro | Stardance
|
uniquebitro
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: Cosmic Sticky Notes by @uniquebitro | Stard Title: Cosmic Sticky Notes by @uniquebitro | Stardance...
|
1792802933.577951
|
1793407733.577951
|
|
67471
|
wega challenge by @unioncyclist77 | Stardance
|
unioncyclist77
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: wega challenge by @unioncyclist77 | Stardan Title: wega challenge by @unioncyclist77 | Stardance...
|
1793109290.2356412
|
1793714090.2356412
|
|
12398
|
Rock, Paper, Scissors by @unified | Stardance
|
unified
|
0
|
0.0
|
0
|
[]
|
[]
|
Title: Rock, Paper, Scissors by @unified | Stardan Title: Rock, Paper, Scissors by @unified | Stardance...
|
1792947167.9027464
|
1793811167.9027464
|
|
57950
|
Random website demo by @unified | Stardance
|
unified
|
0
|
0.92
|
0
|
[]
|
[{"post_id": 61110, "ts": "202 [{"post_id": 61110, "ts": "2026-09-03T20:42:44Z", "hours": 0.9242, "text": "I made a beta of the landing page (ill make it look way better in the future)"}]...
|
Title: Random website demo by @unified | Stardance Title: Random website demo by @unified | Stardance
Devlogs:
I made a beta of the landing page (ill make it look way better in the future)...
|
1793465472.2383692
|
1794070272.2383692
|