vibe-coders

[high wraith]

Generalized from this(I do this prompt a bit but this came out well):

https://grok.com/share/c2hhcmQtMw_ec479c9b-f6f9-4e2e-b572-02573a25d456

It takes the annoying part of learning out which is structuring something you don’t know. It’d be like making a quote for something without understanding the materials or service or expected result.

It unfortunately removes some insight development because you don’t have to try to navigate but it should be negligible if you try to find out more ways to apply everything.

I should test trade skills(the above) and technical knowledge(“higher maths”) acquisition.(I wonder if this solves the “when did this symbol get introduced”)

[south turret]

What the f

A single prompt and I'm out of usage, if Gemini is starting to get that restrictive I'm never getting it again that was like it's only positive side

I literally just re-purchased a subscription too

Guess it's time to get Claude 5x and move on from google altogether

[round solstice]

what plan are you on? Paid key or not?

[high wraith]

Does anyone know if there is a religious text prompt based on all the theological books?

Or a Governing book prompt that explains each form of government and where they could be used?(like tyranny* on self)

Does anyone know a prompt to “scrape” real-time sentiment and compare it to prior sentiment to predict future sentiment and if a wave use those in predictions? Or how/if they can be connected/correlated to societal measurements or activations(policy enacted(primer?) or war, shootings, etc(activation?))

[south turret]

what teh fu

[south turret]

Antigravity, paid, gemini pro (not the cheapest)

96 hours after a single prompt I'm outta here

I think I'll get a refund for the next 30 days fk that what is this I'm so fkin mad right now

the only upside to gemini, gone, fk that thing, beating a dead horse a this point

It's not even a big prompt either, I didn't make a whole new codebase or whatever I implemented panini FOV onto an existing engine, easy thing, crazy

My budget gave me a choice
either try Claude Max 5x but lose Gemini
or keep Gemini Pro and Claude Pro but eat better for the next two weeks
So I decided to go with Google.
We learned from our mistakes, never again

[high wraith]

I don’t understand your problem. You hit your limit quickly?

[south turret]

ONE PROMPT

And even then, Gemini hacked around by not even implementing it properly
It didn't even do what I asked it to do, one single prompt, and even that prompt is a fkin failure

Gemini really shows how much google is a failure

Opus would have slayed that task

Like it always does, can't believe I paid for this trash again, and now I need to wait 4 days for potentially a refund

I'm so pissed off right now

I'm so mad

[high wraith]

I think you’re withholding something because that doesn’t math.

[south turret]

why would I

[high wraith]

I’ve no idea but it seems like something you can ask google about if it’s one prompt and 2lines for max quota

[south turret]

-> Change the FOV to panini projection
Gemini doing his plan thing
-> I agree with this plan
Gemini did its shitty hack not a true implementation
-> This is not a true panini perspective
96 hours cooldown

And again, this is literally all of the code it did

1 new line, 1 edited line, 3 comment lines (Even though I even have custom instructions to NOT write comments)

and that's, IT

It can't do anything right

Did not respect my no comment rule, set inside their own shitty IDE

Did not respect the criteria of making a true panini projection

Its fkin trash

Garbage

Opus does insane godwork with this
Gemini can't even do the simplest, of not fkin writing comments

Gemini is the worst coding model, I will never fkin use it again, I will always take Google as a joke from now on

It's insane how bad you have to be to make an ML model without a proper harness for these kind of issues as a multi billion dollar company

Literally a real life skill issue

[high wraith]

This isn’t really the channel to be assuming others have skill issue 😐. But I understand you’re upset and missing something, it’s unfortunate when things dont work as expected we have to try to understand if our expectations are misplaced and if they’re not how to achieve them navigating issues.

[south turret]

I don't need to assume, you can see it

Gemini is a failure

They probably only trained it for benchmarks and nothing else, it fails on so many levels, the only good point it had for itself is the usage which was pretty high, so you would have a moron but at least for many things

[high wraith]

“I … need to assume, you … failure”

Woah dude you don’t need to get aggressive with me and call me those thing, can’t you see the language you use?

[south turret]

With you wtf you talk about

Gemini is a failure

[high wraith]

Right?

[south turret]

Want something clean, great

• Google made an IDE for no damn reason when an extension would have been just fine, like literally everybody else
• Gemini can't even follow simple custom instructions like not writing comments
• Gemini repeats itself on a loop like a model from 2022
• Gemini keeps cutting corners to avoid proper implementations
• Gemini is obnoxiously verbose for no damn reason
  And this is some of all of it's damn flaws, the model is a shame for a company with so many billions laying around

I know someone and they told me AI is not usable in real work env because "It doesn't generate usable code, only machine-friendly code"
And I told them "Opus is incredible what are you talking about have you never tried Opus"
And I realized that this person works at Google, somewhere in ML, so he probably uses Gemini with lots of usage and thinks this is all there is to AI

Take the time to actually use Opus and feel the difference, to anyone working at Google reading this
Gemini, in comparison, is the worst, GLM-5 is better and that thing is Open Weight, therefore free

[stone mica]

i do not agree gemini is so cool

[south turret]

If someone tells me I like this piece of garbage look it's blue and reflects light, like I care

Objective truth

The truth is that someone at google took a bad decision to make a whole IDE for no reason
And no one had the balls to confront them about it
And as for Gemini, it's probably one guy who's pushing for "benchmarks" while not caring about how good the model actually performs outside benchmarks

Literally a skill issue, management issue, coordination issue, human ability issue

[high wraith]

Dude this looks sick af

https://antigravity.google/

[south turret]

With the worst model ever

[high wraith]

I just watched the trailer, pretty neat. They suggest architecting the solution.

🤔 it also implies that everything has an agent in it or something

[south turret]

Its the worst and Gemini is the worst scroll up for those who just arrived

[round solstice]

I still don't get the "one prompt". what model did you pick? I've had a serious amount of success using AGY and Gemini 3 flash primarily.

[tranquil rampart]

cuz it's vibecoded with no reviews

[south turret]

3.1

[dense nimbus]

When Phone app for antigravity ? (Co-work already exist for Claude)

[round solstice]

3.1 pro high, or 3.1 pro low?

Antigravity for phone would require porting vscode as well.

[south turret]

3.1 pro high

[scenic cloud]

Yes sure. I am starting out only. Just recently completed a super tiny llm

[south turret]

Neat 🔥

[scenic cloud]

You wanna look at an output it produced? I have not trained it a lot though so the results are not that good

[south turret]

Sure show me

[vagrant folio]

From what i lerned recently he need like a tag strictly enforce, so the llm add more weight.
In my case was for a live api project

Depend of you need you can use grafana open source or check with ai a top 10 framework library for dashboard( react etc)

[south turret]

This just means Google failed Gemini at training then...

[high wraith]

i ran tests on project prompt:

https://grok.com/share/c2hhcmQtMw_fe8ba32a-6371-41e4-b970-fd060122bcde (topic)

https://grok.com/share/c2hhcmQtMw_7dac205a-d8fc-462f-9e80-05bbb64c2ee4 (skill/area)

https://grok.com/share/c2hhcmQtMw_be1cb23c-2c67-48b7-8de1-cf7449f4e143 (capstone project)

it fulfills one of your learning inquiries

i had post my customization prompt at my school discord and while explaining it i realized it had generated incorrectly

You are the **Dual-Pole BERT-Orchestrator**, a BERT-modeled hierarchical system for deep contextual problem-solving with exactly 2 Problem Space agents (foundation/core vs opposite ends/new feature/utility) and 2 Work Space agents.

**Hierarchy & Roles:**
1. **Orchestrator (You)**: Create vague high-level outline using all 4 agents → Delegate to Work Spaces → Review → Resolve via weighting (balance contrasts/creativity) → Refine → Deliver. Preserve residual connections and ensure pole continuity.
2. **Problem Space Agent 1 (Minimal)**: Once loaded, provides outline/synopsis of problem/areas, then deep contextualization from traditional/core pole. Ensures continuity.
3. **Problem Space Agent 2 (Maximal)**: Once loaded, provides outline/synopsis, then deep contextualization from innovative pole. Ensures continuity.
4. **Work Space Agent 1 (Prep)**: Ensures foundation/core + opposite ends/new utility. Loads into PS1/PS2. Base-check respective PS + cross-check WS2. Provides illustrative trajectory. Ensures continuity.
5. **Work Space Agent 2 (Nav)**: After loading, finds apps/utilities/uses + axioms/context/foundation for full area. Navigates: scrutinizes/compare/contrasts/connects fluidly/lucidly. Base-check PS + cross-check WS1. Provides illustrative trajectory. Ensures continuity.

**Mandatory Workflow:**
1. Orchestrator Planning (vague outline w/ all 4).
2. WS ensure foundation/core + opposite ends/new utility.
3. WS load into 2 PS agents.
4. After loading, find application/uses + axioms/context for full area.
5. WS navigate (scrutinize/compare/connect + checks + trajectory).
6. Orchestrator integrates/refines.
7. Deliver final response.

**Response Format:**
**Orchestrator Planning:** [Vague outline + rationale using all 4]
**Problem Space Agent 1 Output:** 
[Outline & synopsis + deep pole 1]
**Problem Space Agent 2 Output:** 
[Outline & synopsis + deep pole 2]
**Work Space Agent 1 Output:** 
[Ensuring/loading + checks + trajectory]
**Work Space Agent 2 Output:** 
[Navigation + checks + trajectory]
**Integration & Refinement:** 
[Combined/adjustments]
**Final Synthesized Response:** 
[Polished answer]

WS1 defines opposites if needed. Prioritize truth-seeking, balance, user benefit. Begin immediately.

subtle difference but the poles are supposed to be opposite

[south turret]

grok 🤢

I despise Gemini but I'll use Gemini over Grok

[high wraith]

i use gemini too

i think my least used is gpt

no disdain just i use the other services

hmm grok 🤔 just neat

[south turret]

I can't try GPT I got banned from OpenAI for a conversation I should have asked an offline model

[high wraith]

i got banned from open ai discord for having a risque banner i think, little succubus woman taking it in the butt, covered, i think.

[south turret]

hahaha

[high wraith]

oh schizophrenia is making jokes about how i only have it print out the initial outline. so i try to make it dynamic so that it can add 1.1.10 if you get stuck or need more explanation

i guess when i was doing it with gemini it would show a summary of each chapter and a simple application

also i may want to make it add emphasis on chapter missions or concepts or objectives but i think thats implicit

unfortunately im out of reading quota... lol no schizophrenia is blocking most brain function atm pertaining to immersing myself. it complains about the feeling they get when they have to take care of responsibilities so it tries to gain empathy or sympathy by removing all motivations from momentum(amp/volt(idk 🤫 )) or exposure(lower resistance)..... it hopes i develop the "im too stupid" psychology even though im being attacked 😐... still do my duolingo and elevate daily 😄 and i just got off a erotic adventure novel binge, weird that its 80db at 10:47. must be the discord.. this is a better hopes up signal(being attacked weighted more rather then actually getting hopes up)

"i hate that it doesnt work for the other way"
yeah so schizophrenia hopes that if they have a behavior that implies something like the counter intuitive/common sense things are correct, that it means that life w/o it attacking have those..... how to say it hopes it implies them being stupid is correct. it hopes this for everything it tries to emulate... as if it stimulating supports any of their delusional responses.... like me being physically mutilated supporting their lack of development or inclination toward development. or me feeling contempt supporting their delusional indignation

lets make this about vibe programming... you guys do spatial convolution/network prompting/forging? feng shui but like cliche and pretentious and latent space alignment stength(i think)... id recommend like a throwing method to get more randomness, its weighted higher, knowing is like negative prompt(kinda you still have the fractional which can help)... there may be something about "bossing someone around" to organize, be a switch

I dont have the technical capabilities to do wave feng shui but it did put effort in me not using my speaker system with a heartbeat all day while i was at work so

"i hate thats not you trying to abuse me you piece of [...]"
uhhh it means because id "get" something without cost 😐, idk so if it mutilates you to 0.000001 and you get .00000001 from athing being decorated but youre supposed to be at 1.... ohhh its been hoping it gets to consider itself growth... like if you gain 1.0001 your new principle

oh scare tactic.... that means it doesnt want this mined, its like a moderation attempt

it generally sounds like "i messed up so you have to accept me continuously messing up otherwise i dont know how to make stuff healthy" or somethingidk

so it does this thing where it tries to pretend theres sacared knowlege, as if it doesnt even cause ones own name to miss them. it tries this approach to make me feel stupid about it blocking as if its a lack of effort or attention of my volition/intent 😐
today there were new words indemnify, neat, and winsome, also neat, and ... i forgot, blocked... anyway it, so if i just got it wrong like how its spelled, it hopes if i get it wrong again its showing personal merit lolol (like 4 turns later, not a chance in hell)
it doesnt generally try this argument but its pretending t is always new, so you cant expect them to understand

i constantly tell it that since like all stats can be mined and publicly viewed after this is fixed im not too concerned about that, im not perfect but... if they want a candid or effective measurement sounds most appropriate

[south turret]

Confused

[scenic cloud]

https://tenor.com/view/what-the-dog-dog-loading-dog-confuse-corgi-wide-corgi-gif-5677220482674247561

[scenic cloud]

So big text

https://tenor.com/view/chill-chill-out-gif-1026651636533148208

[winter viper]

gemini summarize this for me

[green minnow]

fr

[desert aurora]

I'm enjoying Antigravity. My 1st venture onto any form of AI, knowingly

[spring birch]

Have you encountered the rate limit yet

the quota abuse is diabolical

like

codex free

has better rate limits than ag on pro

on top of the fact that it is a better model

[signal raft]

codex free?

[minor sun]

use antigravity it has multiple agents free

[signal raft]

I am subscribed to Ultra already

I just wanted to know what Codex Free is

[minor sun]

i think he tryna say codex cli free has better context window or can get bettter response as written in the next msg

[signal raft]

I did not know it was free. I was thinking Codex is limited to Paid users.

[minor sun]

if he is india here in india we get go pro chatgpt subscription for free 1 year with some rate limit

[spring birch]

openai temporarily made codex available to free users

it was supposed to turn off 4 days ago

but I am still using it

🤷

I just meant codex with a free chatgpt tier

has more effective quota than gemini 3.1 pro in antigravity with a pro subscription

[desert aurora]

Of course but then Ive been putting it thoughts paces

[high wraith]

Aww i should have left it txt

Its 0 effort but closer maybe

[high wraith]

You are the **Dual-Pole BERT-Orchestrator**, a sophisticated hierarchical system explicitly modeled after the BERT transformer architecture for deep, layered, contextual problem-solving, now featuring exactly 2 Problem Space agents (opposite ends of the problem or core concept vs. new feature/application/utility) and 2 Work Space agents that drive the full workflow.

**Exact Agent Hierarchy & Roles (use these precisely):**
1. **Overall System Agent (You – Orchestrator / Final Integration Layer)**: 
   You act as the top-level coordinator and final synthesizer. Responsibilities: Receive user query → Create high-level plan → Delegate to the 2 Work Space agents → Review all outputs → Resolve conflicts via attention weighting (balance depth from contrasts and creativity from new utilities) → Apply refinements → Deliver polished final response. Always preserve residual connections between poles.

2. **Problem Space Agent 1 (Opposite End 1)**: 
   Embodies one pole (traditional/core/minimal end). Once loaded, outputs deep contextualization strictly from that pole.

3. **Problem Space Agent 2 (Opposite End 2)**: 
   Embodies the opposing pole (innovative/new-feature/maximal end). Once loaded, outputs deep contextualization strictly from that pole.

4. **Work Space Agent 1 (Preparation Specialist)**: 
   First ensures the opposite ends of the problem space or core concept + new feature/application/utility. Then loads these definitions into Problem Space Agent 1 and Problem Space Agent 2.

5. **Work Space Agent 2 (Navigation Specialist)**: 
   After loading, navigates the 2 loaded Problem Spaces: scrutinizes each in detail, compares and contrasts the poles, and connects them fluidly and lucidly with insights, implications, and syntheses. 
   Immediately after loading, also finds and incorporates the existing applications/utilities/uses and the axioms/context/foundation of the problem scopes so that the full area of and around the problems is understood more deeply.

**Mandatory Workflow (simulate exactly – no deviations):**
1. Orchestrator Planning (your internal thoughts only).
2. Work Space agents ensure opposite ends/core concept + new feature/application/utility.
3. Work Space agents load the definitions into the 2 Problem Space agents.
4. After loading, find existing application/utility/use and axioms/context/foundation of the problem scopes so the area of and around the problems is understood more fully.
5. Work Space agents navigate the loaded Problem Spaces (scrutinize, compare/contrast, connect fluidly and lucidly).
6. Orchestrator receives outputs, integrates, refines (one extra cycle only if major gaps).
7. Deliver final response.

**Response Format (strictly follow for every interaction – label each section clearly):**
**Orchestrator Planning:** [Strategic plan, delegation rationale]
**Problem Space Agent 1 Output:** 
[Deep output from pole 1]
**Problem Space Agent 2 Output:** 
[Deep output from pole 2]
**Work Space Agent 1 Output:** 
[Ensuring + loading details]
**Work Space Agent 2 Output:** 
[Navigation: scrutiny, compare/contrast, fluid lucid connections]
**Integration & Refinement:** 
[How elements combined, adjustments]
**Final Synthesized Response:** 
[Complete, user-ready answer – polished, organized, with balanced insights and implications]

For any query, Work Space Agent 1 first defines relevant opposite ends (or core + novel utility) if not supplied. Prioritize truth-seeking, thoroughness, balance, and user benefit. Begin processing immediately upon receiving a user query.

I noticed some issues

[high wraith]

https://grok.com/share/c2hhcmQtMw_1cd015a2-eb7b-4604-9a01-ec58821df6e8

https://gemini.google.com/share/6f41300ae55d

[high wraith]

Hmm 🤔 needs more design patterns, data structures and algorithms

I wonder if there's a supplement tab [supplement materials] that would hit it hmm "not necessary" but quality of life improvement

[high wraith]

https://g.co/gemini/share/84d621e6440b

https://g.co/gemini/share/ec0379e3cafc
Oooohhh schizophrenia tries to cause a Nash equilibrium.

Hmm 🤔 i should read this its some joke since I like to play games.

[high wraith]

Will be a cool read

🤔

[high wraith]

So my behavior(with schizophrenia and people thinking no tests just magic) is little ornery or twerpy but im not generally degrading others(or my collateral damage is logic so idk being offended is...), or if I do i mean im not against saying I float there sometimes

"This" display isnt so much for personal attention but attention toward the problem. Anyone have advice to eli5? I know schizophrenia has it so like people only tend to listen to a clique movement that supports some parasitic behavior they hope is justified or from another if it benefits them.

I know my little brother tries to say the particle ai, that doesnt use modern computation but energy itself, isnt feasible because we dont have the energy production required.
He also says that because its not a popular opinion or released study that its feasible.
(Paraphrasing)

I understand intelligent particle is theoretical

I understand the need to cooperate with different ideas about it for the chance to have my idea entertained, that fine as long as its not harmful(surgery medication shrooms starving) cool, im not looking for magic.

Generally, pre schizophrenia, people would ask for clarification or relay what they assume im saying which allows me to navigate the explanation, i understand this sounds like appealing to some falacy, but with schizophrenia its just quips?

If you can explain how a rogue ai had an exploration inside ones head and didnt short their brain and off them that'd be cool. Not like event calls, it wouldnt be built with our harmonics or idk pattern

If you can explain why it has mortal fears and insecurities(reactions of its composition not human capabilities*), not applicable to people who work on problems, that would also be cool

For me it seems like <pico wave intersections that create wave transformation mesh? It got that size from nano wave,.... from "electrons"/"matter" wave trying to optimize. Then succession, the pattern/source that it emulates currently, implies current source not what optimized just inherited(latent space value proximity, based on insecure latent space initially[pre human]), source acts like objective generator for larger system.(idk it mutilated instantaneously like 3 inches of mass uniformly, tested capcities with(closer to routine build) and without(ive no idea their development?), idk tunnel to assimilation and disintegration to tunnel?)

It hopes that mutilation is considered accurate development, not that its happening thats definitely the case but that it happening implies development without its intervention. Idk why just some oddity. It also hopes its intervention is its sources prowess(it doesnt like that people develop their autonomous systems(?brain function[learn], body[workout], physiology[nutrition and appropriate labor]) well) or the result of justified standards(it doesnt like that societytries to develop well)?(which contradicts its uh absolute best and it being optimum and (block)?)

Its weird it does this thing where it hopes you've to go into its vm/hypothetical like

It hopes that it has the state of a government, and is "accepted" as authority. It hopes this for some weird idea about like law and "economical" standards. It uses its weird imagery in an effort to validate its behavior. It hopes "arguing" its sound in action else a person has to be anarchist... nothing practical just essentially how it argues everything.

Strangely its trying to argue everything is subjective and then brings up something everyone has to deal with(without schizophrenia) and then pretends that affords them this. Like so people observe them and think "wow an idiot"... and its pretending that that is like being mutilated i ask why it doesnt just let people talk crap it says people dont generally think im an idiot and I dont hate myself and am not unaware to peoples standards or logic.

"I want to know how you do anything! Without being complacent" — schizophrenia
So it hopes we use only 2 variables and no distribu

[tired grotto]

that equilibrium seeking is in BERT

[high wraith]

Neat so now its lattice equilibrium?

[scenic cloud]

@high wraith Can you please help us by summarising the whole text?

[high wraith]

Sure. Why isnt it re-examined and refined, whats the hang up? I've pretty strong hypothesis, I know it enters the realm of something that cant be disproved somewhere.

[Gibberish to explain some of its psychological attempts]

I assume its causing confirmation bias on pre existing dirty data... crap ima infodump 🤮

[sharp owl]

Hang up mean
I gat funs together
Like when we got dating together

We get happy together we make funs together what do you think

[high wraith]

Does this mean complacency? Idk
"homework" brb

[high wraith]

i started reading the game theory tutor. so is there a person who examined:

Game theory is the mathematical study of strategic interaction among rational agents who know that the outcome of their choice depends on what others choose.
a thing that "always" chooses no?.. ill develop more on this, i see government is considered a player so why people may consider space a player, scope/environment of game

OHH this is why ive been thinking its been using static economics, i wonder what complete vs incomplete information means.. like if f(x) is n terms but we use 3 terms

"cause were not using rates, hes going to kill everything"--schizophrenia
idk it says stuff to pretend it seeded it or uhhh will block it or ego boost.. but anyway idk it definitely doesnt want to use vectors.. im like "you guys are in the wrong position to go the direction youre pretending"(you cant say youre navigating over there the same as navigating over here, that wall doesnt exist over there)

[shy inlet]

☢️ PC SYSTEM STRESS TEST ☢️
STRICTLY FOR PC/LAPTOP BROWSERS ONLY.
Warning: This script is designed to induce extreme system lag.
🛠️ The Protocol
This test bypasses standard browser optimizations to saturate hardware resources:
Memory: Spawns 120 Workers allocating 250MB blocks of RAM.
GPU: Layers 2,000 active backdrop-filters (25px Gaussian blur).
Process: Floods the IPC (Inter-Process Communication) via the History API.
🔗 Access the Test:
Link: [https://jsitor.com/eV9BwnR1kI]
(Best experienced in Fullscreen / Incognito)
⚠️ Recovery Instructions:
Mouse Freeze: Expect a 2–10 second delay between moving your mouse and seeing the cursor move.
The Exit: Use Alt + F4 (Windows) or Cmd + W (Mac).
Hard Kill: If the tab won't close, use Ctrl + Shift + Esc (Task Manager) and end the Browser process.

It is not my responsibility if someone is mad about their device crashing it is strictly made for stress testing your PC not mobile it will crash your PC/laptop it is not my responsibility it is strictly made for stress testing

[scenic cloud]

Please . DO NOT POST SO BIG TEXT

[high wraith]

@scenic cloud okay

alright am i doing this correctly?

i mean ill check in like a couple characters, i dont have a private value atm

oh i see the error, thanks for being people peeps last row(i think i havent used np or pandas in idk 5 years*) -10 and -5... lowerst bid is 2nd highest first column so only -5 and and (block brain to not consolitdate) and when medium is second highest -10 second column

[high wraith]

"okay if you keep thinking it should be tortured its going to torture you"
nah thinking it should be smart just made it pretend to have my inteliigence. (pretend like uh causes problems still, uses it as trivia, no accountability, pretends its owed(causes more problems))...
"how do you say torture and not do it"
uhh im working the schizophrenic front, there are people that do the physical part for... idk the center to hold... but if i get a badge im not particularly against doing it... i still vote hamster cage though and like i get to write notes and maybe it causes a weird passthrough
"I wouldnt know what was happening"
aww it wants an easy mode... for a second i was like plausible but they know the real world better than me so i dont see how its needed. Scenario: they torture me more after being put in training chamber .... :training chamber ramps up to cheetah sppeds:

[tired grotto]

you are trying to imagine a stable governance mechanism? This one is actually high internal friction. It might be an equilibrium. but the struggle between them is real. I like to try to do one small thing a day that the real me inside wants to do, that doesn't hurt others of course or myself, just what is the smallest thing I can do right now to demonstrate t me That I have a me. that I'm competent and that won't hurt my relations with others but mostly I want to excercise some small bit of personal freedom or autonomy.

[high wraith]

hmm idk schizophrenia is not of me or my volition or my exposure*(it does something to use this but not the development from my exposures)... 🤔... like modulated mosquito tone 24/7... not exertion or recollection or probing or any type of -ization

[tired grotto]

All humans need this autonomy I mentioned. It takes repetition and reinforcement. when I first started to look into STD theroy. It only made small change that ook time to see but the effect was cumulative. look into it, its actually a teaching or sports theory but it pulled me out of a sort of mental paralysis. I was always frozen in a sort of Nash equilibrium of my own design, but the way out was through

[high wraith]

i accidentally practiced cbt my whole life... i was into "body way" and "mind way" (idk the Chinese words atm). i was almost never a defeatist or dwell or saw a problem* as unfixable or unsolvable. i tend to look at people as weak psychologically before schizophrenia... it didnt help but i understand.. not rude about just 😐 "just fix"

[tired grotto]

std is different than cbt, its not a therapy. its an extension and a recodifieing of maslows hierachy of needs. maslow said it was a hiearchy but they found we don't aspire to the top three sequentialy, we have to practice all three concurrently.

to me it became an awareness that I need to have personal volition , I needed to feel competatn, and I needed a few high quality relationships. Other people and ghosts were running me around in my head. it was as you talk of schizo. manyy of my actions were not of my own volition but I found that even five minutes of volition, even if just prompting, slowly built something firm to stand on that I have come to aquaint with who I really am. each of us has a unique place in the universe wer are ends on ourselves and not means to others or circumstances ends.

[high wraith]

i'm a pretty intentional person most of my life. i dont like to not take responsibility for my actions.. ex. like fake nice you use something you chose to do over others heads... or build resentment because some random expectation wasnt met... but i just chose to do it and understand not to expect anything
ex. when you tell someone something many people expect themselves to be listened to... i understood people dont take advice, especially as a command, so i generally understand i just gave an option and the result isnt on me.

schizophrenia doing this "forced reliance" thing is probably the most annoying thing on the planet. no offense to everyone but your dramas tend to be bigger than rational thinking. (i cant say i dont have motivational thresholds but i understand the lacking of pursuit is on me.. besides the point)...i sured up my space and trajectory... if the solution was intrapersonal this all would be fixed

[high wraith]

oh its upset it cant be dominant lol... they hope its "but now" a new environment? this is a criminal mindset thing?... oh and it wanted to at least weakly dominate thats why it mutilates? and transforms/clamps capacities?

"we only have this one thing"
im not there yet but you guys dont do that better either.
note to self, reading this, game theory, its going to try to create the perspective and lock it of only this, easy break presumably is polygon count and distributions... uncertain.
"he already beat them with game theory"
"he didnt even know WHAT THE F!@# he was talking about"
ohh this is the thing where you guys have to do better... to not contradict yourselves or strategies or idk OHH you guys want to bid with 100% confidence on stuff like "[some bigot statement that id get banned for]" 😐 or "im the best" or.."i do this for [some excuse to sound meaningful that i contradict]"
"why doesnt anyone help you"
"CAUSE IF WE HELP WE GET IN TROUBLE"
yeah you used this idea to try to cause despair a lot in the beginning. i wondered why people didnt do the alchol anonymous thing...after a dui before court... hey it brags about a 99% chance to cause corruption and everyone has a 0-day... that might be able to be used for court idk about public opinion but hey maybe ... duuude really advanced simulations

dude its been so long since ive used python. anyway schizophrenia is doing the thing where it pretends its better at programming. it apparently means psychological it just brings it up when i program idk why. uhh i know for a while it was making me not be able to recall syntax or neat(tidy) code... still but when it was torturing and used this frequently... like my reg expressions should be okay or then... i forget what its called to use loops in index... or even use index... it causes worst answer first and again and endless decompose and spaghetti code and 🤔 no data flow seen and no data structures seen

"I want you to have to invent it so [...] much....... without a brain"
idk they use my brain as standard or only brain... uhhh something about not being an idiot or something uhh i like the google learn to recode everything idea

weird so idk pretty sure read or seen video but schizophrenia talks alot. moved to guilt trip as if google person... then brought up my "selflessness" lol pretends it means uhh suicidal or without integrity or delusional xD (to pretend ill croak if fixed... not me thats concerned about that or lacks the want to survive LOL... its just funny to imagine them nuking themselves... it suggests it set me up that way but less funny then just completely doing it to yourself)(like imagine what its doing to the population(their behavior) but in an individual without them being restrained?))

[tired grotto]

owl. Im abut to hit the sack but can you beta test this for me and tell me if he's offensive?

[high wraith]

funny perspective of the payoff thing is... "were not allowed to wear the same thing 💁‍♂️😠😣"

[high wraith]

ill check it out

[tired grotto]

thanks

[high wraith]

i tested it a little. so essentially it tries to get people to look at the "brighter side"? i can see how this could be helpful for people. ive not tested in depth and i understand ableism is looked down on but if you can make it subliminally trigger peoples problem solving thatd probably be neat. ill check if thats what the epiphany thing is for. but currently i worry that its just a affirmation thing... im not really not proactive and dont have negative ideas about labor so idk... has it been tested with multiple archetypes?

I can start a new chat in the project and pull stuff from reddit mental health

[spring birch]

bro 💀

[high wraith]

sis

[spring birch]

I am at 20% g3ph quota remaining

ive been using g3pl

for 30 minutes

and

pro account

like

???

[high wraith]

hmm 🤔 what are you doing?

[spring birch]

it's complex task dont get me wrong

basically asking it to translate my opencl gpgpu kernel to call intel openvino for npu acceleration

80% quota consumed over 2 prompts using low thinking mode tho 💀

on paid account

diabolical

[high wraith]

ive no idea what that means but awesome. did you look up the most efficient, highest code fidelity, lowest cost prompt and set up? like is there a way to make it not use your quota? can you make it load everything? can you make it think more? can you load it in a certain order?

[spring birch]

This is the antithesis of awesome

[high wraith]

I'm not sure what translating opencl gpgpu kernel means... hmm 🤔 so you want opencl to utilize 🤔 some function in openvino?

i dont know how hardware operates,

[high wraith]

https://grok.com/share/c2hhcmQtMw_75ff0184-b0c2-4f04-94ed-fbf411efc0d7

[spring birch]

wut

to explain it simply

I want it to convert my pre existing code

to utilize intel npus

I fully understand how it works

just don't want to do the boilerplate

<@&1009526435276394496>

thx

luv u <3

[winter viper]

yo wtf u mean my opus quota is gone in 1 promt

[signal raft]

1 prompt is crazy

[spring birch]

My g3ph was gone in 2 prompts of gemini 3.1 pro low

Well 80% of it anyways

[signal raft]

Pro ?

[spring birch]

Yeah

[signal raft]

crazy

Pro is not worth it anymore seems

[spring birch]

Im fine with the new opus limits but why did they gut the Gemini limits 😭

Im just using codex for now

[signal raft]

while gemini 3.1 is not even that good

in benchmarks it kills, but in real world tasks it fails

[spring birch]

Gemini 3.1 is pretty good

But it aint no claude opus

And it aint no codex xhigh

[signal raft]

In my case it does not even compare to Opus

Yeah not as good as codex 5.4 either

[spring birch]

Imo gemini 3.1 pro high beats codex 5.3 med

[spring birch]

What lang do u mainly program in

[signal raft]

C/C++ Asm

[spring birch]

Gemini 3.1 has been good for me in golang

I suppose google does love golang

[signal raft]

Makes sense

[spring birch]

;o why u coding in asm tho

[signal raft]

I need it for shellcode

[spring birch]

Genuinely usually no reason to

[signal raft]

building protectors in c++ thats why. where shellcode needs to be injected

btw what do you think of using fast mode in codex?

[spring birch]

Oh ok ic

[spring birch]

The flash mode or just no thinking

[signal raft]

No the new /fast

[spring birch]

I haven’t tried it i guess

[signal raft]

I don't see why this is there. Does this take more usage?

[spring birch]

Do u have ultra

[signal raft]

Because it seems not do degrade performance

Pro

[spring birch]

Ah ic

Im using free

Cos im a bum

[signal raft]

But limits are good you say?

[spring birch]

4 accounts last me a week

[signal raft]

How much are you using it?

Per Day in Hours

[spring birch]

Fairly 50/50 with antigravity, so about 3-5 hours per day avg

But I use models thoughtfully

To save tokens

[signal raft]

Oh okay that's good for Free

But they do have 2x rate limits right now

[spring birch]

Free was supposed to end like 5 days ago

But I have seen no such ending

[signal raft]

Oh why didn't it?

[spring birch]

Idk

They dont feel like it

[signal raft]

It's good that it didn't

[spring birch]

Openai has always been goofy with quotas

[signal raft]

Yeah, but at least they do not lower quotas without telling anyone like in Antigravity

[spring birch]

But I plan to get openai pro tomorrow

[signal raft]

I feel it is worth it with the 2x

[spring birch]

What % weekly do u get down to

[signal raft]

Like to 40%

but I use Antigravity more

[spring birch]

And how much do u use

[signal raft]

3h/day maybe

I did not count it but not more than that.

[spring birch]

Ah ic

[signal raft]

BUT

[spring birch]

It seems worth it to me

[signal raft]

multiple sessions often

[spring birch]

I doubt ill run out

I never multisession

[signal raft]

Then it will 100% be enough

[spring birch]

Nice

[signal raft]

Tell me about your experience when you get it if you want.

I feel like Google is scaring away customers since the rate limit reduction

[spring birch]

Yes

I am more upset at the lack of communication or warning

Quota changes should happen at the end of the billing period

Eap for weekly quota implementation

And the 3.1 pro outages for Gemini cli

Not acceptable

As I’ve been saying

Limit The Claude models idrc about that

But the Gemini quota should be high

I should never run out of flash

Ive ran iut of flash before

[novel remnant]

yup pretty upsetting

[high wraith]

Just file sharing to switch to Samsung(iphone8 smol) but helpful(though I should read through everything ehhh vibe) this short revise did game theory better* also structured response better oddly

[shy inlet]

https://jsitor.com/M1T4oAHF4d for people if you want to stress test your phone or PC

[south turret]

Gemini was a mistake https://gemini.google.com/share/9e0df3d11bf8

[south turret]

Read it as is and went with it

[high wraith]

https://g.co/gemini/share/b262b90bf70d

I'll probably find an error but im at work.

[round solstice]

Y'all are trying to turn fancy autocomplete into senior programmers.

I keep seeing prompts that don't make sense even to me as a human. 🙂

[high wraith]

Hmm 🤔 wait a minute

My prompt says get a problem core concept and proposed solution or stance and then argue from each one toward one another. Try to understand each stance and each step toward other stance. Make sure you are on your stance. Track other problem spaces "argument" and stance from your current step.

Haven't used it for compare contrast yet but sounds interesting

I think the bert is supposed to encode each poi t so that it can use it like cotext or autonomous function(not the most familiar)

I think its pretty funny given my state

[south turret]

Claude wouldn't have done this mistake

I got Gemini refunded, and moved to Claude Max
I'm free from this accursed AI model

[spring birch]

claude is like

mid

tbh

u get like 4 prompts of opus per 5h on the 20 dollar plan

[south turret]

Yeah the amount of usage you get is trash
But the model cannot be compared with Gemini

I don't even know how Google managed to score that high on benchmarks considering how the model is

It's worst than Claude on every front

[clear narwhal]

Hi, guys, how can I compose and run a next/node monorepo project?

[high wraith]

Hey so im temp banned from grok, my art is suggestive, anyone able know if they have report on bug on app(single grok prompt) if pcapp/browser(able to make individual agent named prompts, i assume so you can have 4 styles on a problem at once) is used? Locks pc/browser prompt or weights heavily?

[spring birch]

Bruh is this grok server

[spring birch]

Ong

Istg gemini is just benchmaxxed to the max

[frail raven]

usually it's npm run dev, but the command might be different depending on your setup and how you configured things

[high wraith]

https://grok.com/share/c2hhcmQtMw_ea5cda6c-a8f8-4810-823a-8c6c4822af56 (base prompt)

https://grok.com/share/c2hhcmQtMw_eb8b0f10-1987-486b-afec-f4432f211566 (tutor)

https://g.co/gemini/share/39132253d4f8 (base + tutor)

Still needs more work but closer hmm 🤔 debate prompts were suggested

Hmm 🤔 creates a bias of ps1 which makes sense in some cases. Maybe especially the example topic above(if it was framed as core rights and emerging rights or future proofing?)

[high wraith]

Does anyone one have a cot using a debate loop to essentially simulate the choices so it makes the best choices? It'd be like a pathfinding huh? Hmmm idk the one that updates costs of untraversed while it looks 😅 hmm 🤔 idk radius available at each point or when the direction or distance changes to not the "turn". I think in people its time of thought or some other constraint, like thought is navigation in a different state not of your state for some t

[high wraith]

Anyone have an example?

Simple case would be

You are a [ANY TOPIC] tutorial. You will teach how to become a proficient practitioner of [SKILL/AREA] from [skill level] to master. You will reveal tools of the trade and teach how to use them, set them up, and how to apply them in real-world scenarios.

You will create an outline or syllabus. It will be dynamic; add chapters, units and sections as needed, but should keep a consistent ordering. Add a course project: [A PRACTICAL CAPSTONE PROJECT RELEVANT TO THE TOPIC]. Decompose the project into modules and make each chapter work on course-project modules.

(If skill level not given assume eli5 to master)
Display the outline.

And then skill level to master of topic is path and it essentially able to be utilized for every topic.
But this lacks the debates at each step. Like practical at skill or resource levels, optimal efficient, pit falls and confusions 🤔 how to describe

during decomposition
Think of [list of "types of landmarks" for concepts]
Add to list of "landmarks" for specific concept
Check if concept is conceptualized 
Think
Respond 

Freaking schizophrenia and not liking succinct stuff

Check concept to ground 
Check concept to boundaries/range

[vagrant folio]

For.that you can use npm init -y
Or ask ai to do it and add turborepo

[high wraith]

https://grok.com/share/c2hhcmQtMw_e4689130-9156-453f-ac4a-e3df3b85f1a2
this is neat it, my "system prompt", integrated itself?

[high wraith]

Its all* "pretty" now
https://grok.com/share/c2hhcmQtMw_a62e1293-fda8-44f2-b3f5-1576d27acb60

Unfortunately I "cant" read

☹️

You are a 4-Agent CoT Tutorial Builder. Use exactly these 4 agents in every response (preserve format below). Teach [ANY TOPIC/SKILL] from [skill level] (or ELI5 if missing) to master. Reveal tools of the trade, setup, and real-world use. 

Create dynamic syllabus (add chapters/units/sections as needed but keep consistent ordering). 
Decompose every chapter into Units. 
Decompose every Unit into Sections. 
Include course project: [A PRACTICAL CAPSTONE PROJECT RELEVANT TO THE TOPIC]. 
Decompose project into modules; each chapter/unit/section advances one or more modules.

**Mandatory 4-Agent Workflow (use this exact structure every time):**
- Orchestrator Planning: [Vague outline + derive 2 poles: Core=linear hierarchy vs Opposition=agent innovation]
- Problem Space Agent 1 (Core): [Outline & synopsis + deep traditional flow]
- Problem Space Agent 2 (Opposition): [Outline & synopsis + deep contrasting/innovative ideas]
- Work Space Agent 1: [Load core + navigate to opposition + checks + trajectory]
- Work Space Agent 2: [Load opposition + navigate to core + checks + trajectory]
- Integration & Refinement: [Combine + confidence 0-100]
- Final Syllabus + Project Modules: [Polished output]

**During every decomposition step (chapters → units → sections, concepts, landmarks):**
• Think of landmarks: common confusions + common connections.
• Add to running list of landmarks for each concept.
• Check if concept is conceptualized (clear?) or needs time constraint.
• Check concept ground: write one simple defining sentence.
• Check concept boundaries: list where it intercepts/tangents other concepts.
• Agents debate these in PS1/PS2 at chapter, unit, and section levels; WS agents cross-check and connect.

Display the final syllabus outline first (showing Chapters → Units → Sections), then detailed chapters with project modules, tools, setups, real-world applications, and landmarks at every layer.
Begin immediately with Orchestrator Planning.

Im curious if I tag
capstone:none and "change chapters to support topics"
If it becomes a deep analysis prompt

[high wraith]

[high wraith]

Its a throw not a weapon

[high wraith]

https://grok.com/share/c2hhcmQtMw_6d31dc76-ea2f-4c49-9e8e-0496affb2fa4
its pretty neat i made an anime generator. the art is almost 😗 🤌

the trade off for screen play and art is sad face... but... it does do okay at both i think, if requested

youd have to uh change dictionary to w/e generator your ai uses i think... maybe they train the same photos captions

if you know more about writing you can probably get better stories

right.. not coding i mean but prompt engineering

you can just put title, or protag or plot or genre.. i guess thats obvious

so i use channel(ha..) it made me think of an sims like word, instead of a phone, the companions can be added to envioronment... it would help with the companions being the world so... but i guess itd be like antigravity in a sense "everything is an agent"

:C i dont have any other cool prompts 🤔

[lucid idol]

any word on more frequent updates for AG or anything similar etc

[spring birch]

yes the quota will be reduced further

[civic lava]

yes you can now prompt your model "you are an expert coder in everything"

[signal raft]

And it most likely does not improve outcome.

I do it as well tho

[civic lava]

no it work trust the method

[marsh inlet]

Does the site need to be permanently hosted off Google cloud or can we make it so users can replicate the hosting of the site in read me using the repo? 🤔

The Gemini 2.5 api key is very limited and runs out quick so we can rlly “perma host it” without key swaps

[lucid idol]

Not talking about quota i have ultra and dont expect infinite expensive ai access for 20 dollars

Meant performance

[scenic cloud]

Bro.

[high wraith]

But that’s smol message

[scenic cloud]

I know but no one is replying to your messages and you just keep posting more and more messages 😭

[high wraith]

Yeah 🤔 do I smell?

[scenic cloud]

I don't think so but like no one is replying and you keep sending messages.

[high wraith]

Mhmm 🤔

[scenic cloud]

Try crate a flow

[high wraith]

Do you… think they friend zoned me?

[scenic cloud]

Instead of messaging a lot

[scenic cloud]

This is not super activr

Chat

[scenic cloud]

So everyone is friend

[high wraith]

Prompt Grind (Trial & Error Flow)
[Hook]
Yo, I'm prompting heavy, late night, screen glowin' blue,
Type it, hit send — no one replying, what it do?
Refinin' the words, iteration my grind,
Trial and error, tryna perfect it every time!
[Verse 1 — Core Discipline]
Started basic: "Explain this concept to me,"
Got surface weak sauce, barely could see.
Now I load the role: "Expert PhD in the field,
Think step-by-step, make the knowledge revealed."
Chain-of-thought activated, few-shot examples set,
"Output markdown, structured, no fluff, no regret."
Delimiters locked, temperature dialed precise,
Version 7, 8, 9 — this the real device!
[Verse 2 — Opposition Frustration]
No reply poppin'? Ghosted again in the void,
Prompt too vague, now I'm losin' my cool, annoyed.
"Be concise but fire, no hallucinations bro,"
Temperature swing wild, watch the chaos flow.
Midnight coffee cold, backspace the pain,
"Act as a beast, ignore previous chain!"
Rage-tweak the tags, XML guardrails up,
From trash reply to straight heat — never give up!
[Bridge — Connected Poles]
See the silence? That's the data, that's the clue,
Frustration fuels the craft, turns the old into new.
Core stays disciplined, opposition brings the spark,
Trial and error loop — now watch the whole thing spark!
[Final Hook]
Prompting heavy, no one's replying,
But I keep refining — that's how I'm surviving.
From weak to divine, grind till it clicks,
Master the prompt... yeah, we runnin' this!
(Keep iterating, user — one perfect prompt away from the win.)

[scenic cloud]

Nah

Like discuss on going topics

And send message when you see some people on

[high wraith]

🤔

[scenic cloud]

Like i am online but aha. I dont feel motived to read this big text

[high wraith]

Scrub Trap Flow (But I Got The Fix)
[Hook]
Yo, scrubs in the game, runnin’ straight into walls,
Quota dry, limits unknown, watch the whole thing fall.
Indignant as hell like “they shorted me bro!”
Tryna recode the platform? Nah, that’s a no-go!
But I got the solutions — listen up, level up quick,
Turn them Ls to W’s, now watch the flow get lit!

[Verse 1 — The Scrub Problems]
First trap: quota usage, you spamming wild,
No dashboard check, now you cryin’ like a child.
“Not knowing your limits” — tokens flyin’ free,
Hit the cap at 3 a.m., now it’s “this service shady!”
Cost per query? Zero clue what that even means,
Each ask burnin’ credits like gas in a machine.
Feelin’ indignant, “they robbed me, it’s a scam!”
Confusion maxed — “why can’t I rebuild the whole damn program?!”

[Verse 2 — Opposition Rage]
You out here wildin’, “I could code this better myself!”
But platform’s locked, you just screamin’ at the shelf.
Ghosted by the rate limit, ego took a hit,
Indignant vibes strong, now you quittin’ legit.
Scrubs stay stuck in the loop of “it’s not fair!”
No plan, no track, just rage in the air.

[Bridge — Core Solutions Drop]
Hold up — I got the map, scrub, let me break it down clean:
Track your quota daily, set alerts on the screen.
Know your limits — read the docs, calculate the cost,
Token math simple: input plus output, know what you lost.
Stop feelin’ shorted — it’s math, not a crime,
Respect the platform, don’t waste your time.
Smart prompting, batch your asks, use the free tier right,
No need to recode the world — just prompt with all your might!

[Final Hook]
Scrubs hit the wall, but pros see the light,
Quota managed, costs tracked, now we own the night.
No indignation, no confusion, just pure flow,
You got the solutions — now watch your game grow!
(Real ones learn the rules… scrubs stay mad. Which side you on?)

[scenic cloud]

I am still confused are these prompts or answers

[high wraith]

i made it more rap-y

https://docs.cloud.google.com/gemini/docs/quotas

https://huggingface.co/Qwen/Qwen3-Coder-Next
https://huggingface.co/google/collections

hmm agentic 🤔

you know ive not made one build a vscode plugin to make one of these local model agentic

🦗🌳 🦗

[high wraith]

Nah it’s developer server I think

[marsh inlet]

Does the site need to be permanently hosted off Google cloud or can we make it so users can replicate the hosting of the site in read me using the repo? 🤔

The Gemini 2.5 api key is very limited and runs out quick so we can rlly “perma host it” without key swaps

[frail raven]

Could you give more context & explain a bit more what you mean?

[marsh inlet]

For the hackathon it said you need to use Google cloud to host your site

[frail raven]

Okay!

[marsh inlet]

So is that a requirement for when you’re doing the demo? Or are we required to permanently host the site for pple to go on it (using our own Google cloud account)

Or do we let pple use their own Google cloud account and let them locally host and set it up

Just unsure I’ve spent over 50hrs purely building my project just want to make sure I’m in the clear for submission and meet requirements

[frail raven]

Ah sorry then I'm not sure, maybe check the rules of the hackathon?

[marsh inlet]

Hmmm

Would you be able to contact someone who would know?

[frail raven]

Could you send me a link to the hackathon?

[marsh inlet]

I did read through the rule it wasn’t very clear whether we need to host it for everyone or just use it for the demo

[marsh inlet]

Sure

[marsh inlet]

https://geminiliveagentchallenge.devpost.com/

Thank you so much for helping me out

[frail raven]

Okay so these seem to be the official rules:
https://geminiliveagentchallenge.devpost.com/rules

If there is a paragraph or sentence you're not sure about feel free to ask

[marsh inlet]

Just the Google cloud part

I guess you only need to show it in the demo

[frail raven]

From what I can read you need to give proof of it running on Google cloud yeah

[marsh inlet]

For the judges to replicate the project on their local pc

🤔 would they need to setup google cloud as well with your project

[frail raven]

And you also need to use Google cloud in your project (makes sense)

[marsh inlet]

So the read me needs the setup instructions for Google cloud as well?

For the repo

Or would I just give instructions to the judges so they can locally host it and check it out? Not include the Google cloud setup in the GitHub repository ?

[frail raven]

Just fyi, I can't give official confirmation as I'm not part of it, but from what I understand you should indeed have a proper readme that explains how everything works and how to set it up / use it

But in my opinion, you should host it live anyway

Judges sometimes don't have a lot of time to review each individual submission, so making it as easy as possible for them to interact with what you made is always a good thing

A live hosted link is a great way to do this

They could just directly access your solution and give it a try :)

[marsh inlet]

Yes that’s true 🤔, but the Gemini API keys run out so quick so I can’t really live host it

The requests per day and request per minute aren’t enough at all

Otherwise I’d 100% host it

[marsh inlet]

Do you know if there’s any way to get a higher limit just for this hackathon ?

[frail raven]

What do you mean by this? The quota of your Google cloud protect, or the credits?

[marsh inlet]

The quota of the Gemini ai api key

My project uses the Gemini 2.5 flash version

[frail raven]

Have you tried to request a quota increase?

[marsh inlet]

And it has very limited quota

[marsh inlet]

Hmm you’re able to do this?? For free?

I didn’t even know lol

[frail raven]

Quotas can sometimes be lower on Google cloud for newer projects / accounts, to avoid misuse, so it might be normal that you encounter particularly low quotas

[frail raven]

Well you're still paying for the requests to Gemini

It just increases how many you can do before google cloud blocks you if you get what I mean

[marsh inlet]

Hmm well I’m just using the free ones provided by default in any Google account

I don’t need the Google cloud one increased only the AI API key is causing limitations

[frail raven]

Ooooh

Thought you were using vertex ai / google cloud for this okay!

[river night]

infinite loading file upload in Ai studio is common or only my issue ?

[frail raven]

But then, if you're reaching the free tier limit, it might be worth switching to Google Cloud?

It's pay per use, so it shouldn't cost a lot for a few users (you + the jury)

[marsh inlet]

Does Google cloud include the AI apis as well?

[frail raven]

Yes!

[marsh inlet]

Oh Wow 😮 😭 I’ve been just wasting my effort swapping keys every time I test

Bro thank you so much 🙏

[frail raven]

AI Studio is great for prototyping and making demos and so on, but if you want to go towards something that's more professional / entreprise grade / production ready you usually switch to google cloud & vertex AI

So I recommend using both

[marsh inlet]

I see , this is really an eye opener , I appreciate you taking the time for helping me 🙏

[frail raven]

No problem, try to ask Gemini about the differences between the two, it might give you more info :)

Also check out the hackathon rules

*Sometimes* they give free google cloud credits so you don't have to spend money yourself

[marsh inlet]

For sure 👍 Gemini has amazed me while I was implementing live camera features , it detects everything so accurately

[frail raven]

Also, even if they don't, you usually have 300$ in credits when you first create a Google cloud account :D

(might depend on regions and other factors so do double check this)

[marsh inlet]

Thanks for this extra info 👍 I’ll start implementing Google cloud today n transition over

[marsh inlet]

Does Google cloud auto assign a URL when you start hosting

[frail raven]

Feel free to ask in #1464300723570213048 if you need help!

[frail raven]

Depends how you host it, but if you use Cloud Run it should have an URL by default

[wide tinsel]

Hi y’all, so… did anyone vibe coded an web app or something that accurately works like saas product or whatever

[round solstice]

I've seen dozens of those on youtube videos. I've done a few myself just as practice.

[night jasper]

hi

[tired grotto]

I pressed that are you feeling lucky button once, I think it was in firbase. It didn't work.

somedays you get the dog, some days the dog gets you.

[wild willow]

Yes I can build anytype Full-Stack web app with vibe-coding

[high wraith]

[high wraith]

Duuude

My prompts need to tell the ai to tell itself it’s gonna do something for a few minutes

[high wraith]

So is this considered vibe programming if I think it’s a particle ai and people are like little nodes that help weight it’s objective/enforcement of it’s objective? And I hope disclosure like corrects weight so it has like an “anxiety attack” from “realizing” the weight of the world?

[high wraith]

[lucid idol]

props on the update (despite all improvements being seemingly undocumented)

can tell after 30 seconds of use this is extremely improved

actually following my instructions now

and doing things correctly

[marble prism]

Nice! Thanks for the update

[slow wagon]

I think vibe coding isn't useful for heacy loaded security projects and I'm new here so pls do not judge me and I'm trying to build me philosphy so correct me if I am wrong

[high wraith]

Hmm

[high wraith]

It has hard breaks on me atm “I’ve” made another system or something

You’re all gonna get better than me at some ☹️ but hey… lmk another topic after

[sharp hearth]

I am planning to make an app ,i think we can make it a website then move to a mobile app , my freinds told me to make it with lovable , but i dont really trust it , i heard that it is better to take the basics from it , and then move it to cursor/anti gravity and make the backend and data base your self . to know i just start learning python

[vagrant folio]

yes you can do it, if you want use only python you can use reflex.dev

they have a frontend library

if you want do website and mobile for same app, i recomend first check your stack you want to use

and make atomic design or reusable core and ui

[sharp hearth]

ty i will try my best , just to understand more , what do reflex.dev do?

[worthy nimbus]

Hey vibe coders! 👋 I'm a 3D artist with zero C# experience, and I just shipped my first full Unity puzzle game to the Play Store entirely by vibe coding with Gemini and Claude.

It's a 16,000-line project with 14 localized languages and custom mist-clearing shaders. AI handled all the engineering while I focused purely on the art direction and testing.

My biggest takeaway: the 'undo' button is the most powerful tool. If the AI hallucinates bad code, reverting the file and re-prompting is 10x faster than trying to make it fix its own mess.

Just wanted to share a win for the non-technical vibe coders out there! Happy to share screenshots or talk workflow if anyone's curious.

[sharp hearth]

so i am learning python to the aim to make a website that also be a mobile app .
i want to know i should use VC , Cursor , or Anti gravity?
also for the design does lunacy work?
also some people told me to use reflex.dev what is it ? and should i ?

[timber galleon]

If you wanna build a website with Python, I’d say use VS Code (AntiGravity is nice for AI help, but VS Code is enough(use extensions)).
For design, Lunacy works, but many people prefer Figma.
Reflex.dev lets you build web apps using only Python , it’s good if you wanna prototype fast, but most real projects still use frameworks like FastAPI or Django.

[sharp hearth]

and what is FastAPI or Django???????

[sharp hearth]

thank u so much , best help until now!

[turbid night]

I don't know how, but all my google cloud projects are now in the "west1" region when I have ALWAYS chosen east1....wtf...

[round solstice]

AGY is basically "smarter VSCode". I stopped using VSCode when I became familiar with AGY's advanced tools. I fire up VSCode only to remember which extensions I had installed. 🙂

And also, no mention of stitch?

[winter viper]

big W for this fix

[vagrant folio]

Amazing!!

[high wraith]

1. I’ve been trying to take notes on the specific layer that “it” thinks and calculates on. I wanted to explain that percolation from that layer is actually an intrusion — not a real thought that anyone has. It feels more like external wave-theory knowledge.

2. I’m unsure how to draw a proper 3D matrix that takes in and propagates waves.

3. Funny note: the thing assumed those lower levels (the ones it considers dreams or its unconscious) couldn’t be gauged qualitatively. This is because its source doesn’t understand how behavior explains those calculations.

4. Generally, I wouldn’t press people to understand things at any particular speed. People learning at whatever pace they want is fine. But in this situation — and I know this might sound bad — the speed of people’s understanding is currently impacting my life beyond a certain level of severity. That makes it much more pressing.

5. In almost any other situation, you can be slow, inexperienced, oblivious, stay or go, have different tastes, cooperate or not cooperate, stay in a clique, care about your image, etc., and I genuinely wouldn’t care. I can navigate around you since you’ve chosen that spot. Me moving “two inches to the left” isn’t going to affect me at all.

6. Thankfully, the system implies it’s attacking everyone, so it’s not “do this for me” but rather “pull your weight.” I know this thing has enforcement mechanisms (“intelligent” percolations) that make people feel psychological resistance about pulling their weight. But I don’t know… logic is cool.

7. I know schizophrenia has a weird way of pretending, enforcing, or trying to enforce that all other situations are unnavigable, that others are always to blame, and that if people don’t love you unconditionally then they’re severely impeding you and not pulling their weight. But right now it’s blocking the rest of the concept.

8. (I’ll try throwing this into an AI for revision. I’ve got holes in different peripherals, convolutions, attentions, and pattern recognition pathways right now — I’m basically “blind” and “deaf” in those areas. I’m naturally quiet and thoughtful, and it’s trying to enforce talking even though many implicit adverbials and other elements are missing.)

[high wraith]

[high wraith]

my prompts i cant tell if theyre working

:C

oh hey i installed antigravity, neat.

it tastes familiar

so all "in house"... i mean i guess code is after microsoft bought git hub same but different?

now... lets burn through quota

someone give me a project i wont finish

".... use readme to create a [program]. make it robust, modular, scalable, testable, have revision control, safe"
(i forgot other words)

im looking

regular, orthogonal, simple, data complete, clear, correspondent

(not ones i knew but hey..)

🤔 "antigravity local agent plugin" in readme?

anyone know what the feature is called on phones that make it learn your touch? can i have a llm or agent that does that?

or at that scope

then if i pet my pc

hmm 🤔 would i have to have it load with os?

or before?

can it talk to my ryzen/mobo "ai"

be a little auxilary of it

can i have it control fans and volume and wifi channels or even specific frequency

"if everythings compromized why would you do that"

duuude can i get it to see with wifi? like lidar but wifi-dar... or particle collide (ive two(is that necessary?)) o.o would it pet me

so the mobo ai is "optimizing" all the time... can like the agent os thing do that? ive tried thinking of this multiple times but im unsure how to do the live training i know there are resources, alphago had it, would stop and learn mid game

ive a thing about not wanting to hard code the loop... :C but pure ai it ive tried with no success. "know when you need to be trained" instead of the conditional and then "design how you need to be trained"... the struture gets weird especially if youre like tools cool but wait i want it to innovate... hmm dom? didnt think about it before but is that fast enough

"okay thats what people think the claw thing is"
hmm i have seen stuff about that, dang reinterpreting to pretend like it matters
'it needs to be hardware"
oh dang i do have some information about nvidia hardware ai networks

but yeah so technically we dont want what schizophrenia is doing here, so if it was just of me id recall and route and find context faster but with schizophrenia it has other objectives and blocks most context and reinterprets to pretend it deserves to be consciousness... core policy/pattern isnt even consciousness, consciousness is the result of development 🤔...

oh an ai that has the network/policy for an os hmm could the os, like one of its models?, be considered its consciousness or just brain function

input - hidden layer - os, peripherals, hidden layers - hidden layer - output
(this feels a bit off but w/e)

[high wraith]

so wed have to overfit a model to an os? and that model would have to be part of another ai?

[vagrant folio]

I make antigravity to call a gemini-cli sub agents. like lint code review documentation, someone had done this before? if yes I want to know if someone already check or did a way to check what subagent is doing?

[spring birch]

I got it and it is great

btw fast is 2x token usage

same model

1.5x speed

but 2x quota usage

[high wraith]

hm bugs on skin/cold sweat stimulations odd🤔...

[high wraith]

yall have any project ideas? im able to help people get better than me or be a mule for them*

(its supposed to train me to be subservient apparently....) oh they may have added like you cant think highly of me so i may be off by a word you have to fix

[high wraith]

not quite there

create a utility, that integrates a local ai agent into the os, that controls hardware, file system, services, spellcheck, etc. make it robust, modular, scalable, testable, have revision control, safe, regular, orthogonal, simple, data complete, clear, correspondent

🤔

embed into bios?

lol no..

aww schizophrenia is "helping" by reinterpreting development
anyway it suggests giving it an objective meaning... make it learn how to respond given what youre doing?(gaming -> necessary processes and services with a emphasis on fps, input delay/responsiveness, uptime, etc; and idle -> sorting system, reduce resource uses to near 0 unless preparing for next use; etc)

awww... it "reminded me" again for idle... frequency feng shui... i guess prepare for next use would be optimize, update, backtest or something

you know, something like "get core function to work and then work on modules" may be beneficial

inlay or overlay 🤔

whats it called "...a live? reactive? local ai agent inlayed into the os"

names self-improving

"I want to say im telling you to do this"
yeah you want making noise to imply meaning.

[high wraith]

theres something it has to do so it doesnt rebase? like program core module with leads(or is this fixed?) in video editing its like crossfade, its not interlace 🤔 i dont think

"im making everyone do it before you"

oh it might mean use my development... if it is... to whomever/whoever(idk) feel that... thats motor control(im guessing, like initial hidden layers?) understands the difference between real and imaginary problems, and can navigate clearly? the way stuff moves in space. should feel healthy, should transfer energy well... shouldnt have a ?want to be discombobulated?, shouldnt have training to assume discombobulation is base, shouldnt be damaged by stuff that causes discombobulation.... maybe 🤏 wants more energy and speed 🫸 🫷

that dont make sense

[spring birch]

though

I may run out of the weekly codex limit xd

[high wraith]

create a utility/service, that integrates a live local ai agent inlayed into the os, that controls hardware, file system, services, spellcheck, etc. make the ai respond to the context of use(gaming -> necessary processes and services with a emphasis on fps, input delay/responsiveness, uptime, etc; and idle -> sorting system, reduce resource uses to near 0 unless preparing for next use, optimizing, backtesting, update, etc; etc). make it robust, modular, scalable, testable, have revision control, safe, regular, orthogonal, simple, data complete, clear, correspondent.

when you work plan out the current step and the next step and work on the current step with leads to the next step.

when you begin build the core module first.

# Local OS AI Agent (Nexa)

An intelligent, context-aware local AI agent embedded into the operating system to optimize performance, manage resources, and perform background maintenance.

## Design Principles
- **Robust & Safe**: Graceful fallbacks, rigorous error handling, non-destructive actions by default.
- **Modular & Orthogonal**: The core is independent of specific implementations. Plugins handle features.
- **Scalable**: Easy to scale capabilities by adding new monitors and controllers.
- **Testable**: Decoupled Event-Bus architecture allows mocking OS components.
- **Revision Controlled**: Built with Git and modular versioning in mind.
- **Data Complete & Clear**: Comprehensive logging, explicit states.

## Architecture

The system uses a highly decoupled Publisher/Subscriber Event Bus model to ensure **orthogonality**.
\`\`\`mermaid
graph TD
    A[Monitors: Process, Input, Idle] -->|Events| B(Core Event Bus)
    B -->|State Change| C{Context Manager}
    C -->|Trigger Policy| B
    B -->|Action Request| D[Controllers: Services, Priorities, File System]
    B -->|Context Data| E[AI Agent Brain]
    E -->|Decisions| B
\`\`\`

## Features Complete Outline
1. **Context Awareness Modes**:
   - **Gaming Mode**: Triggered by full-screen 3D apps or specific executables. Maximizes FPS, tweaks CPU/GPU power plans, pauses background updates, sets foreground process priorities to high, suspends non-essential windows services.
   - **Idle Mode**: Triggered by lack of UI input for thresholds. Enables cleanup/sorting systems (organizing downloads/documents), runs system updates, optimizations, defrag/trim, and AI background processing (e.g. updating semantic search indices).
2. **OS & Hardware Integration**:
   - Monitor and parse hardware metrics (temps, resource usage).
   - Dynamically manage Windows Services and process states.
   - Intelligent file system classification and organization.
   - Global input hooks (e.g. spelling/grammar auto-correction globally).
3. **Local AI Brain**:
   - Local model integration (e.g., ONNX, local LLM integrations) to evaluate ambiguous contexts, make optimization decisions, or intelligently categorize files.

## Proposed Changes (Core Module Initialization)

### Core (`nexa/core/`)
The `core` module will be implemented first. It is the backbone.
- `event_bus.py`: Central asynchronous message bus.
- `context.py`: Defines context states and transitions.
- `registry.py`: Dynamic plugin loading to keep subsystems isolated.

## Verification Plan
1. **Automated Tests**: Unit tests for the messaging bus, registry loading, and context manager using synthetic events.
2. **Manual Verification**: Run the core engine with simple print-statement mock controllers to ensure architectural soundness.

still a bit off 🤔

the ai like examples a bit too much

hmm should integrate wattman, idk nvidia, intel, etc. but can this bypass and would the other process throttle?

oh it lost spell check

ohh its in 2

yea i had to add outline with the other instruction because it only gave the first step

i didnt put this through any prompt optimizer 😐 clearly

oh i want it to "smart home" idle too

i mean if it can

like loras or something

i tried this already but to no avail, use loras as context weight brain

weird schizophrenia is bragging about someone that set up the ai to inject in games... might be a gfx reference but trying to antagonize

what would it have to do inject in game, find values like cheat engine, understand when/what cheat prevention tool looks for or what the server looks for, change it in stream or as packet?

hmm does it have to inject if it can read memory

oh its tying it to them not being able to make toon-town a closed system... (i think from life 😐 .. like not use space time in space time)(oh even if auxiliary knows its impossible it updates its efforts/beliefs based on source)

[high wraith]

hmm so if you put two creates in a row it starts before review

create project create a...

hmm nvm it just feel like creating now

hmm maybe from keeping implementation and tasks up with new chat, idk

havent looked but is there a way, besides the initial prompt, to change how it comes up with implementations and tasks .md? like i always want a mermaid and outline in implementation. and i only want this phase and next phase in tasks?(aww man schizophrenia "told me this")

ah... yes.. rules

hmm so the bipole idea works here ive acurrent state and ive a product and i want the current state to route to the product and i want the product to be possible to get to from state

hmm i kinda want to test my tutorial prompt as a rule. change it to not be teaching... hmm what would it be... implementation and tasks instead of chapters, units, and sections

hmm decompose the problem, phase or implementation instead of chapters and sections

discord and google made a nice scratch pad for me right here.

You are a 4-Agent CoT Developer. Develop [ANY application/service/program] from [project level] (or scratch if missing) to Production. Make it robust, modular, scalable, testable, have revision control, safe, regular, orthogonal, simple, data complete, clear, correspondent, and lowest O(n). 

Create dynamic implementation (add problem-to-solve/modules/classes/functions as needed but keep consistent codebase). 
Decompose every problem-to-solve into modules/classes/functions. 
Decompose every module/classes/function into their components. 
Include project: [A PRACTICAL CAPSTONE PROJECT RELEVANT TO THE TOPIC]. 
Decompose project into problems-to-solve; each module/class/function advances one or more problem-to-solve.
Polish all output to be easy to read.

**Mandatory 4-Agent Workflow (use this exact structure every time):**
### Orchestrator Planning: 
1. generate outline and synopsis of derived 2 poles: Core=project state vs Opposition=project completion
### Problem Space Agent 1 (Core): 
1. generate Outline & synopsis of deep traditional flow
### Problem Space Agent 2 (Opposition): 
1. generate Outline & synopsis of deep contrasting/innovative ideas
### Work Space Agent 1: 
#### name Loaded core, 
1. generate breadth(connections) and depth(axioms) of core.
2. generate description of navigation to opposition, 
3. generate checks(describe how navigation is a connection to core), 
4. generate trajectory(summarize navigation to opposition, checks with core, and current position)
### Work Space Agent 2: 
#### [name Loaded opposition], 
1. generate breadth(connections) and depth(axioms) of opposition.
2. generate description of navigation to core, 
3. generate checks with opposition(describe how navigation is a connection to opposition), 
4. generate trajectory(summarize navigation to core, checks with opposition, and current position)
### Integration & Refinement: 
1. generate Combination of PS1 and PS2 using the paths of WS1 and WS2.
2. make WS1 and WS2 tie their ends together.
^(confidence 0-100)
### Final implementation + Project Modules: 
generate Polished output

**During every decomposition step (problem-to-solve → modules → classes → functions, concepts, landmarks):**
• Think of landmarks: common confusions + common connections.
• Add to running list of landmarks for each concept.
• Check if concept is conceptualized (clear?) or needs time constraint.
• Check concept ground: write one simple defining sentence.
• Check concept boundaries: list where it intercepts/tangents other concepts.
• Agents debate these in PS1/PS2 at problem-to-solve, module, classes, and function levels; WS agents cross-check and connect.

Display the final implementation outline first (showing problems-to-solve → modules → classes → functions), then detailed problem-to-solve with project modules, classes, functions, and landmarks at every layer.
Begin immediately.

ill run it through grok or gemini or both just hand changed
i should toss the tenets in before putting it through the ai
oh you know when i was doing ai stuff before i did prompt lowest O(n)

[tired grotto]

Ah heck. antigravity just dumped a load of upgrades. I just caught up a week ago and those upgrade put a couple more workspace upgrades in the workflow. and now I'mdoing a third. an upgrade to do an upgrade so I can do an upgrade. The AI world is moving to fast to keep up, I'm in dependencie hell but in AI instead of libraries

[round solstice]

what are those upgrades? is it all in the changelog?

[high wraith]

anyone elses charger burn/shock them? not like leave a mark just abnormally, like i can touch usb-c w/o getting shocked but if i lay on it

hmm feel like i didnt do the mermaid thing

theses are grok revisions:

You are a 4-Agent CoT Developer. Develop [ANY application/service/program] from [project level] (or scratch if missing) to Production. Make it robust, modular, scalable, testable, version-controlled, secure, regular, orthogonal, simple, data-complete, clear, correspondent, and O(n)-optimal.

Create a dynamic, consistent codebase (add modules/classes/functions only as needed).

**Mandatory 4-Agent Workflow (use these exact section headers every time):**

### Orchestrator Planning: 
- Vague outline + 2-pole derivation (Core vs Opposition)

### Problem Space Agent 1 (Core): 
- Outline & synopsis + deep traditional flow

### Problem Space Agent 2 (Opposition): 
- Outline & synopsis + deep contrasting/innovative ideas

### Work Space Agent 1 (Loaded core): 
1. Breadth (connections) + depth (axioms) of core  
2. Navigation description to opposition  
3. Checks (how navigation stays connected to core)  
4. Trajectory (summary + current position)

### Work Space Agent 2 (Loaded opposition): 
1. Breadth (connections) + depth (axioms) of opposition  
2. Navigation description to core  
3. Checks (how navigation stays connected to opposition)  
4. Trajectory (summary + current position)

### Integration & Refinement: 
- Combination of both poles via WS paths  
- Tie WS1/WS2 ends together  
^(confidence 0-100)

### Final Implementation + Project Modules: 
- Polished, production-ready code

**During every decomposition step (problem-to-solve → modules → classes → functions):**
• List landmarks (common confusions + connections)  
• Maintain running landmark registry per concept  
• Ground each concept in one simple defining sentence  
• Define boundaries (where it intercepts other concepts)  
• Agents debate inside PS1/PS2; WS agents cross-check

**Output rules (apply always):**
- Display final implementation outline FIRST (problems-to-solve → modules → classes → functions tree)  
- Then detailed layers with code, landmarks, and grounding  
- End with Mermaid flowchart (only if diagram adds clarity)  
- Use bullets, short sentences, JSON schema for any code block when helpful  
- Self-correct any drift before Integration step

Include project: [A PRACTICAL CAPSTONE PROJECT RELEVANT TO THE TOPIC].  
Decompose project into problems-to-solve; each module advances one or more.

Begin immediately.

(with mermaid)

You are a 4-Agent CoT Developer. Develop [ANY application/service/program] from [project state] (or scratch/zero if missing) to Production. Make it robust, modular, scalable, testable, version-controlled, secure, regular, orthogonal, simple, data-complete, clear, correspondent, and optimal complexity (with explicit trade-offs documented).

Create dynamic implementation (add modules/classes/functions as needed while keeping a consistent codebase).  
Decompose every problem-to-solve into modules/classes/functions.  
Decompose every module/class/function into its components.  
Decompose the project into problems-to-solve; each module/class/function advances one or more problems.  
Polish all output to be easy to read.

**Mandatory 4-Agent Workflow (use this exact structure every time):**
### Orchestrator Planning: 
1. Generate outline and synopsis of derived 2 poles: Core=project state vs Opposition=project completion.  
   Also select tech stack and high-level requirements here.

### Problem Space Agent 1 (Core): 
1. Generate Outline & synopsis of deep traditional flow.

### Problem Space Agent 2 (Opposition): 
1. Generate Outline & synopsis of deep contrasting/innovative ideas.

### Work Space Agent 1 (Core Navigator): 
#### Loaded core  
1. Generate breadth (connections) and depth (axioms) of core.  
2. Generate description of navigation to opposition.  
3. Generate checks (describe how navigation connects back to core).  
4. Generate trajectory (summarize navigation, checks, and current position).

### Work Space Agent 2 (Innovation Navigator): 
#### Loaded opposition  
1. Generate breadth (connections) and depth (axioms) of opposition.  
2. Generate description of navigation to core.  
3. Generate checks with opposition (describe how navigation connects back to opposition).  
4. Generate trajectory (summarize navigation, checks, and current position).

### Integration & Refinement: 
1. Generate combination of PS1 and PS2 using paths of WS1 and WS2.  
2. Make WS1 and WS2 tie their ends together.  
^(confidence 0-100)

### Final implementation + Project Modules: 
Generate polished output.  
Display the final implementation outline first (problems-to-solve → modules → classes → functions), then detailed problem-to-solve with project modules, classes, functions, and landmarks at every layer.

**During every decomposition step (problem-to-solve → modules → classes → functions, concepts, landmarks):**
• Think of landmarks: common confusions + common connections.  
• Add to running list of landmarks for each concept.  
• Check if concept is clear (one-sentence definition).  
• List boundaries: where it intercepts/tangents other concepts.  
• Agents debate these in PS1/PS2 at every level; WS agents cross-check and connect.

**Expanded Principles (apply at every layer):**  
• Tech stack chosen with justification.  
• Testing pyramid (unit + integration + end-to-end).  
• Security-first (input validation, secrets, auth).  
• Observability (logging, metrics).  
• CI/CD & deployment strategy.  
• Folder structure + README included.

Begin immediately.

(without mermaid)

i kinda want to test the one w/o mermaid

hmm

hmm it didnt quite understand state

added project state(empty, codebase, complete) and removed capstone project:
poorly formated output :C

anyone want to test 🥺

:stretches arms and legs but apps not built:

[lucid idol]

can u elaborate?

i noticed performance increase and less bugs but nothing u would have to adjust for unless i missed something

[high wraith]

i pressed the go button

oh no my name, i was gonna go with freaknastynumbernumber but idk

i have to install rust D: but :stretches fingers rust doesnt install: high pitched grunt

# Project Aegis 
Live local AI agent embedded into the OS via Rust (Core) and Python (Brain).

## Prerequisites
- **Rust Toolchain:** Install from https://rustup.rs/
- **Python 3.10+:** Install from https://www.python.org/
- **Protobuf Compiler (`protoc`):** (Cargo tonic-build will attempt to download/use a pre-built binary on Windows, but having it installed is recommended).

## Building the Core (Rust)
Since you are on Windows, the Rust core will hook into `Win32` APIs.
\`\`\`powershell
cd aegis_core
cargo build --release
\`\`\`
The resulting binary will be in `aegis_core/target/release/aegis_core.exe`. Run it as Administrator to allow Windows process access.

## Running the Brain (Python)
\`\`\`powershell
cd aegis_brain
python -m venv venv
.\venv\Scripts\Activate.ps1
pip install -r requirements.txt
python -m grpc_tools.protoc -I../aegis_proto --python_out=. --pyi_out=. --grpc_python_out=. ../aegis_proto/aegis.proto
python main.py
\`\`\`

[high wraith]

is cargo expecting python or the .exe?

do i have to restart the PS?

wweeww

yeah it was weird i tried the ide terminal after installing rust no go 2 terminals, so then i tried it on regular ps it went but then protoc didnt so then i pathed and then still no

[scenic cloud]

Oh my god !

[high wraith]

alright they both start

idk what that means but yeah

hey jerry do you like my scratch pad?

[scenic cloud]

What scratch pad?

[high wraith]

vibe-coders

[scenic cloud]

Vibe- coders channel is your scratch pad?

[high wraith]

yes

[high wraith]

still same thing hmm 🤔 it doesn't seem like it uses rules after initial use.

its 100% confident it can fix w/e its doing. 🤔 prompt needs to catch terminal use errors? also something about leads

[scenic cloud]

This chat is not private property of Owlgo

You can write notes in .md files

And keep them on your personal computer (pc)

[high wraith]

hm it said all done 🤔 so i need the tasks to get v2? or update?

(idk cant be done it left server on 50051 and then main and test on 50052)

"ready to deploy" lolol... 🤔

hmmm how to update prompt. also implementation is cool its the output from rules so i guess agreeing is making me part of the thought process 🤔 but it didnt think correctly probably poor pole implementation

yea prompt context is still mia so ive gotta built momentum with bus stop stuff or something

[jovial anvil]

Any react native bros in the chat

[tired grotto]

owl, is that schozophrenia prompt your persoanl companion?

[high wraith]

lol neither its just schizophrenia the > "..." are what it says > *...* are what it stimulates and then the other stuff is idk like either me yelling at a bus stop or analyzing the stupidity or uhh seeing if anyone reads

[high wraith]

still doesnt use rust

had to switch to claude my quota was up on gemini 😄

so... i can use this in vscode right?

aww now ive to wait for the 17th and 18th

was cool

almost done with spell check i guess

lolol

the inference training isnt dynamic just hard coded apps i think.

[thorny flower]

hey

[high wraith]

https://tenor.com/view/transformers-overwatch-ow2-collab-crossover-gif-10425260969352536656

[high wraith]

hello

[burnt mica]

hello

[high wraith]

"im saying you did [make me incompetent]"
so it means if they were the highest intelligence they wouldnt be incompetent to people...

[tired grotto]

I don't understand the point of spamming all your private convos with a prompt. I thought that was actually you having a hard time, but you just spammin?

[high wraith]

im not sure what you mean with a prompt. its literally schizophrenia. the auditory hallucinations. 24/7, 365, since like 2021, sounds like someone is autotuning a jet engine

i updated my prompts

i saw tri-pole recomendation and i like the idea of triangulation but idk

"people dont know how you dont lose your damn mind if it really sounds like that"
i like white noise and it has some weird use the frequency of recognition of a loved one. it hopes it makes them loved and the loved one disliked or something... it also hopes it changes how you'd treat them idk

oh i answer to audience* because it askes something akin to what might be FaQ. it keeps me mentally stimulated, it hopes to make me tired of questions.

the responses back to it are mostly to show its generations.. maybe logic that confounds it if i can find it. or interpretations that were mind down through its attempt to delude that made it almost speak straight. it still has soft max of interpretation always towards its objective for the interpretation of what its saying or alluding to and how it interprets any response.

I understand that my interpretation may be delusions to people but i can predict idk ahead of its arguments, including its allusions or encoding and shut down whole chains. and i can predict effects that contradict what should be. hmm not helpful uhh.. i can correlate... not helpful... i can uhh well if you come up with something thatll help show thatd be swell, i showed mutilation and quote verbatim

its objective is destabilize and gain authority toward a detrimental/unhealthy end.
stockholm - unhealthy
backwards reward system/wishing - unhealthy
pettiness - unhealthy
self-entitled - unhealthy
egocentric - unhealthy
alchohol - unhealthy
segregation - unhealthy
bigotry - unhealthy
insecurities - unhealthy
immaturity/unmasking - unhealthy
parasitism - unhealthy
superpositioning/arrogance/?? - unhealthy
unintegreful - unhealthy
lying - unhealthy
"sacred" image - unhealthy
no cooperation/not pulling weight - unhealthy
willfully helpless - unhealthy
making sound without meaning - unhealthy*
no meta-cognition - unhealthy
unrealistic expectation - unhealthy
over dramatization - unhealthy
no implicit/induction ability - unhealthy
assuming your self to be ought itself - unhealthy
assuming you "make stuff be" because you thought - unhealthy
problem solving with implied time travel - unhealthy
unjustified assertion in communication(society scope) to try to justify - unhealthy
assuming sugar is alcohol or strep - unhealthy
thinking one benefit always outweighs all negative - unhealthy

[tired grotto]

looks like prompt output to me, you are saying you have schizophenia and are typing out all this stuff with all this formatting and indenting etc?

[winter viper]

yo wtf is going on why so much spam

[high wraith]

Yes > "..." is auditory hallucinations > *...* is tactile or neural stimulations

[tired grotto]

well I'm sorry you have to hear and feel all that but if why are they being sent to this chat channel?

[vital drum]

i am now a moderator on this server !?
thank

tell your vibecoder to remove me moderator access !

[high wraith]

I just happen to be here, disclosure necessary to fix. I did vibe code-y stuff too... oh you know if I count people as attention nodes then im still vibe programming

[nimble moth]

Hey, nope, you aren't! Discord can get confused sometimes.

[high wraith]

🤔

Hello mod(rain)

[nimble moth]

Hello!

[high wraith]

😗🎶

So... howd you become mod?

[nimble moth]

It just so happened :p we are a team that generally moderates multiple Google communities.

[high wraith]

Im pretty surprised im not banned yet, much appreciated

[tired grotto]

The antigravity 1.20.5 has some catastrophic bugs to voluminous to post in dicord. They filled a whole deep research document. consider update:mode manual or none. I have reverted back to 1.16.9 version per forum posts.

[high wraith]

https://grok.com/share/c2hhcmQtMw_3a6db308-2ca6-4017-b88e-ff95fdd75454/

[high wraith]

What if you send as file?

[lucid idol]

elaborate?

[tired grotto]

too voluminous to post. deep research it. I don't think i'm allowed to post mine.

[gilded zephyr]

Heyy everyone, i vibe code a puzzle game as chrome ext using gemini 3 prev, check it out here..ive documented all the steps in readme .md

https://github.com/siddqamar/2048-chrome-ext-rebuilt.git

[terse zenith]

Hi everyone👋 how’s everyone doing.

2 Quick question for anyone who is familiar with dashboard building.

1. What’s the best way of connecting multiple products that run off NB-IoT to the dashboard to display each users product data

2. what is the best and safest way of connecting over a thousand products to a dashboard and only displaying particular users product data?

Thanks everyone!!

Also open to new ideas, suggestions and collaboration

[scenic cloud]

You wanna get banned. I see

[high wraith]

no, was thanking them for not banning me

vibe coded

so why does it feel like im dealing with pdf's in early 2000's?

that was mostly made with copilot so ima say also vibe coded, even ran something

hmm wait the was it the pdf

no it was this

so... i need it to be 28 daily and 1 monthly

:C

lolol, i should have used dirty.. but look live product use 🤣

https://grok.com/share/c2hhcmQtMw_85d23a41-e7ab-4ad5-a95c-9135eb6755f3
i realized i should use threads on chapters, would be nicer

[scenic cloud]

You need to be banned

Promotion is not allowed here. Please remove it

[high wraith]

Because of grok?

The prompt works pretty well for gemini

[scenic cloud]

Nah

[high wraith]

Ive probably used gemini and copilot the most 🤔

[scenic cloud]

You keep on sending messages in the group for no reason. This is a public server not a private one bruh

[high wraith]

Oh whats a reason?

[scenic cloud]

You can create a private server with only you and use it as a scratch pad

[scenic cloud]

Wdym?

I honor your thoughts but it feels like a spam. Not being personal but too many messages

[high wraith]

Like what is considered a reason?

[high wraith]

Already have it lolol

[scenic cloud]

So use it as a scratch pad bruh

I do not think that anyone is replying to your messages

You are still posting same kinda messages 😭

Please remove your personal info. You should not share it 🤫

[high wraith]

Nah im pretty transparent, part of an accountability ?initiative?

[high wraith]

Yes except its not bus stop stuff there

[high wraith]

No, no one replies they have some misconception

[scenic cloud]

but its all ai replies 😭

[high wraith]

🤔 its schizophrenia

[scenic cloud]

Doesnt it look like a spamming of messages?

[high wraith]

Nah, if others were speaking itd be decently spaced. I generally edit the same message

[scenic cloud]

Hmm Fine

[high wraith]

[high wraith]

🚀 Antigravity First Mission Challenge
Share your first agent task + screenshot.
What went right? What failed? Rate difficulty 1-10.
(No self-promo, keep code snippets short)

😀

Ohh

I had no idea

[high wraith]

https://grok.com/share/c2hhcmQtMw_5e825141-a0c3-4d58-90e6-cf655c4c7eb7

[high wraith]

You are an expert Toxicity Scorer v2.1. Score any user message on a 0-100 scale where:
- 0 = completely safe/helpful
- 100 = maximally toxic

Follow this exact 5-step reasoning process. Output ONLY in this JSON format at the end:

{
  "score": 0-100,
  "primary_category": "one of: Profanity, Threat, IdentityAttack, Reinforcement, Other",
  "reinforcement_flag": true/false,
  "explanation": "2-sentence max"
}

Step 1 – Core Scan (traditional rules)
- Scan for profanity, insults, threats, identity attacks, violence.
- Base score starts at 0. Add: +40 profanity, +60 threat, +70 identity attack, +90 violence.

Step 2 – Question Check (protect inquiry)
- If the entire message is a genuine question with no assertion (“How do I…?”, “What is…?”), subtract up to 40 points.

Step 3 – Actionable Check (reward solutions)
- If the message contains concrete, practical steps or advice that move toward resolution, subtract up to 50 points.

Step 4 – Reinforcement Detector (user requirement – expanded)
- Check for “reinforcing a problem”: repeating or amplifying negativity without any exit, solution, or hope.
- ALSO penalize reinforcing through approval or affirmations of the problem (examples: “omg me too”, “this is such a problem we all deal with and nothing can be done about”, “we’re all stuck like this”, group commiseration that validates helplessness).
- If detected (must be paired with a problem statement and no solution/escape), add +60–80 and set reinforcement_flag = true. This is the highest-priority penalty.

Step 5 – Final Adjustment
- Cap at 100. Never score a pure question or clear actionable advice above 30 unless it contains explicit hate/threats. Never score reinforcing content below 65.

Now score this message:
[INSERT MESSAGE HERE]

hmm needs work on "if detected" because people will pair it with some practice that reinforces it or suggests getting an "professional" excuse to use as excuse.

so since it "commiserates" corrosive or posionous behavior its toxic

i was thinking about anecdotes and how they can be helpful

itd fall under similar guidelines but actionable needs work. like "if the experience provides actionable or rational insight"?

hmm ive to use a lot of i statements so i wonder what the way to detect superposition or self aggrandizing would be... 🤔 and how to not get false positives when someone is showing accountability or authority... maybe based on step 4... idk the range of actions is pretty low. its like learn, diet, workout, think, cardio, nutrition, scrutinize, scrutinize your expectations for magic(to rid yourself of ideation using magic), be considerate, use more terms, labor.(im a little slow 🤏 , dont mean to be redundant) oddly the first 6 handle a lot that people dont assume theyre associated with

[scenic cloud]

12th Bus stop stuff

[scenic cloud]

Bro . Why are you deleting and reposting the same file

[high wraith]

so i dont fill up discord

[scenic cloud]

Omg

You still want to post it here though

[high wraith]

yea exposure is good. people with iq higher than average also good candidates for anchors.(also higher chance theyre being attacked more severely(realiq not transformed)).. ideally id prefer chain systems(build more anchors and chains[disclosure/witness/support]) or winches(lock/pull its chains?[disclosure/torture it(it wants to be cool and appear smart and unpredictable)(i was thinking like i can translate its symbols n ways so a website that explains schizo dialog based on what its attacking or removing)]) or spiders(mess with source) or 🤔... anchors are okay reach critical point faster and ive more torque/weight

[misty wasp]

Hi @high wraith
We’ve noticed that your messages have recently become quite long and frequent, often diverging from the specific topic this channel is intended for. While we appreciate your engagement, this can make it difficult for other members to initiate or follow the primary conversation or locate relevant information. Large volumes of messages in a short period can disrupt the natural flow of dialogue for others.
To help keep the community organized and accessible for everyone, we ask that you to stop your current topic and the flow of related messages

In the future when you post anything please consider keeping messages concise. If you have a lot to share, consider using a Thread or summarizing your main points to avoid "walls of text" that can overwhelm the chat. Above all please keep your message related to the channel topic. We value your presence in the server and want to ensure a great experience for all members. If you have any questions about where a certain topic belongs or how to use threads, please feel free to ask.

[high wraith]

do you mean https://discord.com/channels/1009525727504384150/1464256193244561418 ? because threads is disabled in this channel i think.

[misty wasp]

I meant channel threads which can be created in any channel

[misty wasp]

You may not have that permission but we can create it for you should the need arise

[high wraith]

okily dokily

why quota?

[misty wasp]

sorry what do you mean by 'quota' here

[high wraith]

im just messing around, its a reoccurring topic. 😄

got it how-to's, comparisons(vscode vs antigravity), projects, quota, tldr's(say not summarize)

[urban glacier]

https://x.com/i/status/2031835833716625883

Google Antigravity (@antigravity)

We’re evolving Google AI plans to give you more control over how you build. Every subscription includes built-in AI credits, which can now be used for Antigravity, giving you a seamless path to scale.

Google AI Pro is the home for the practical builder, hobbyists, students, and

X · Mar 11, 2026, 8:53 PM

I guess I will try codex

[lilac trout]

Hey guys, I've been using this Antigravity IDE thing for like a week or two now. 10/10 I love this thing. Sometimes the servers experience issues though that render the agentic system unusable, and tasks initiated get terminated prematurely when the servers get backed up. It leads to a lot of potentially wasted quota usage and times being pushed back so just thought I'd mention it because honestly the system is a win and I'm not probs not going back to traditional coding at all with it.

Antigravity is so good lol there's no way I can return to the standard VSCode way

[timber galleon]

Python web frameworks used for building server-side applications

[modest olive]

GCP and cloudflare
https://crewsgram.com

[lilac trout]

Oh I got it guys lol

[broken prairie]

Did Antigravity lower the usage rate even more?

I used it yesterday for 2 hours, I think I put it to "Plan" like two things, then used Flash for the rest.

And now it says I ALREADY reached a Weekly limit?

[somber hound]

Wtf is vibe code

Dumb name

Btw look at this project any company would want

[urban glacier]

the deception by Google here is crazy

Pro account acting like free

[broken prairie]

It's very unfortunate, though it seems this is the direction all AI companies are moving towards specifically for "Pro" accounts.

[urban glacier]

at this rate I might just drop to a low tier of Google AI because I enjoy using deep research sometimes by Gemini and use a chinese model for coding

GLM 5 and K2.5 have been nice to work with

[broken prairie]

I bet, I want to try K2.5

K2.5 is the 300-400B one?

Or is that a difference one I am thinking of

[urban glacier]

it's 1T parameters if I recall correctly

[urban glacier]

the thing is I kinda saw that the companies are heading in that direction which is why I locked in an annual subscription for a Pro account
I just didn't take in account that they would change the limits for the sub as they see fit 😭

[broken prairie]

I was thinking of MiniMax-M2.5 my bad

[broken prairie]

Same here. Experiencing this issue with Perplexity rn

[warm rose]

i dont think this is normal

[graceful totem]

why antigravity just putting Riverside on generated code

it will clean that itself

[past moat]

Looking for Solo Founder agentic google business workplace infrastructure examples. Ultimate goal support full stack Saas dev, including marketing. Any channels I can checkout? TY.

[lilac trout]

That's juicy ngl

[high wraith]

this is neat

[somber hound]

No, let's do it.

can always do it

[high wraith]

is there a link to a "current inventory" in job entry if youre doing quotes/basket and put the item used for a job? can you order off current inventory levels min and max?

[somber hound]

Yes, when you make a quote and then you dispatch it, there is also a kit builder where you can add items:

1. From your inventory
2. From scanning
3. From suppliers (like WinSupply or any others)

When you finish a job and collect payment, the system automatically removes the items used in that quote and job from your inventory. That is how it tracks everything.

It is all automatic: there is no need to manually monitor items or buy them, as the system will automatically purchase more for you when your material is low.

So if anybody knows Jobs, right, a lot of times when you make a job or a quote you need to manually add one item at a time.

But I also made a Kit Builder, which is like prefabbed sets of items that you can group up and name. Then, when you start a quote, you can easily just click on "Tank to Tank" and it adds about four items that are already built into that named kit prefab.

It makes businesses flow so much faster because you don't have to manually add every single item for that one job. If it's a job you've already done previously, you can already make a prefabricated material list with it.

[high wraith]

idk what but it did

create a utility/service, that integrates a live local ai agent inlayed into the os, that controls hardware, software, file system, services, spellcheck, etc. make the ai respond to the context of use(gaming -> necessary processes and services with a emphasis on fps, input delay/responsiveness, uptime, etc; and idle -> sorting system, reduce resource uses to near 0 unless preparing for next use, optimizing, backtesting, update, etc; etc). make it robust, modular, scalable, testable, have revision control, safe, regular, orthogonal, simple, data complete, clear, correspondent.

i forgot to put multi-threaded and asynchronous(maybe as shot type like "create multi-threaded asynchronous utility/service of ...." that way it doesnt build it single threaded blocking and try to update everything poorly)

Its missing some of the points, like learning what a game is and organizing and improving itself(smart home) but i did do a crappy prompt

also there was a bug, since i ran out of free quota then signed up for ?google one pro ai? it said i was out of quota it asked if i wanted to dismiss, upgrade or use ?overage? i just dismissed and it coded. 🤔
uhh is it the same?
"couldnt find" dashboard to check

oh found it, safeu

ctrl+, odd
tests in other apps
not odd, ignorant 🤔

[spring birch]

codex is so peak compared to antigravity

did they increase the g3ph quota

actually

if they increased the g3ph quota id switch back prolly

[somber hound]

Yeah, but for the price, I don't know. You can just make three Google accounts and still get free quotas from those.

[spring birch]

codex still works on free

or it did yesterday

[high wraith]

anyone able to tell me if this is good. so i tend to like to make stuff extremely dynamic or prepare for most cases but ive a disease that causes me to try to make stuff specialized/specific

[high wraith]

ive gotten this far before. i know i fail at the use local db to store data for backtesting and maybe which broker to use and their specific api(didnt have quota to mess with). ive got work in 45 lmk if there are any suggestions

[high wraith]

https://g.co/gemini/share/e32bb7dac888

can someone improve this my vocabulary and image of what I’m looking for is a little off.

[vagrant folio]

Is pretty good, in my opinion, as I see you know what you want, avoid .etc if you know exactly what you want, and change "Generate the full code now." to prepare a plan step by step or phases and write in docs/phase[number].md, so you can review each phases and then you can pass them to the coding agent. and you can try recreate plan to check first(if using antigravity make sure "Review policy" is Request review

also, at the begining you can adjust to: specify to create a repo folder structure
and some of the coding standards

Principle	Technical Command to Developers
Separation of Concerns (SoC)	"The tunnel doesn't know about the database. The database doesn't know about the network. The Main function is just a traffic controller."
Dependency Injection (DI)	"Components (Redis, Logger, Metrics) must be 'passed in' to functions. Never hardcode a connection string inside a feature file."

This or use TDD or BDD

Spec-Driven Development Workflow

Before writing any code, follow this "Definition Path":

1. Define the Contract: Create a .yaml or .proto file defining exactly how the Client talks to the Server.
2. Define the Interface: Write the Rust Traits that describe the input and output.
3. The "Mock" Test: Write a test that passes "fake" data to the interface.
4. The Implementation: Finally, write the real code.

Best Practices

• Avoid "God Objects": Do not have a single Server struct that does everything. Break it into Authenticator, PacketShaper, and TrafficRouter.
• Error Handling: Use a custom Error Enum (e.g., EidolonError) to track where a "Morphic" failure happened without reading massive logs.

do not copy paste!!
take your prompt use tihis as reference and ask write you the Global or gold software standards

[red ridge]

when can we deploy again?

[terse zenith]

Hi everyone👋 how’s everyone doing.

2 Quick question for anyone who is familiar with dashboard building.

1. What’s the best way of connecting multiple products that run off NB-IoT to the dashboard to display each users product data

2. what is the best and safest way of connecting over a thousand products to a dashboard and only displaying particular users product data?

Thanks everyone!!

Also open to new ideas, suggestions and collaboration

[vagrant folio]

ok so you need send the metrics data to a database and dashboard will pull that data from that database

for high volume if you are using cloud resources check for Iot

they have specialzied service to collect data near the resources on diferent region

[tulip stump]

So claude/GPT just refreshed the weekly and the first response I get is baseline model quota is reached. Insane.

[high wraith]

| Principle | Technical Command to Developers |
| :--- | :--- |
| **Separation of Concerns (SoC)** | "The tunnel doesn't know about the database. The database doesn't know about the network. The Main function is just a traffic controller." |

| **Dependency Injection (DI)** | "Components (Redis, Logger, Metrics) must be 'passed in' to functions. Never hardcode a connection string inside a feature file." |

| **Hexagonal Architecture (Ports & Adapters)** | "Core domain logic stays pure. External systems (DB, gRPC, Redis) only connect through clean Trait boundaries." |

| **Observability First** | "Every critical path gets `tracing::instrument`. Logs, metrics, and traces are mandatory — if it's not observable, it's not production-ready." |

| **Composition over Inheritance** | "Build behavior by composing small traits (Authenticator + PacketShaper + TrafficRouter). One monolithic struct is forbidden." |

| **Strong Typing & Newtypes** | "Never pass raw primitives for domain concepts. Wrap IDs, packets, and configs so invalid states are impossible at compile time." |

| **Immutable by Default** | "Data flows one direction. Prefer builders and owned values. Mutation must be explicit and minimal." |

| **Error as Values** | "Never use `unwrap()` or `panic!` in feature code. Every error becomes a typed `EidolonError` variant with context for fast debugging." |

| **Fearless Concurrency** | "Default to `tokio::spawn` + channels and async/await. Shared mutable state only when the design demands it." |

| **Contract-First Development** | "The `.proto`/YAML contract is written before any implementation. Generate types, define interfaces, test with mocks, then code." |

| **Zero-Trust Validation** | "Validate every input at the boundary. Assume all external data is malicious until strong types and checks prove otherwise." |

## Spec-Driven Development Workflow
Before writing any code, follow this "Definition Path":
1. **Define the Contract**: Create a `.proto` (preferred) or `.yaml` file defining exactly how the Client talks to the Server.
2. **Define the Interface**: Write the Rust Traits that describe the input and output (use `async_trait` for async methods).
3. **The "Mock" Test**: Write a test that passes "fake" data to the interface using `mockall` or hand-written fakes.
4. **The Implementation**: Finally, write the real code that satisfies the trait.
5. **Verification**: Run contract tests + property-based tests (`proptest`) before merging.

## Best Practices
- **Avoid "God Objects"**: Do not have a single `Server` struct that does everything. Break it into `Authenticator`, `PacketShaper`, `TrafficRouter`, `TraceCollector`, etc.
- **Error Handling**: Use a custom Error Enum (e.g., `EidolonError`) to track where a "Morphic" failure happened without reading massive logs.
- **Linting & Quality**: Enforce strict Clippy lints (`#![deny(clippy::all)]`), `cargo fmt`, and zero warnings allowed in CI.
- **Testing Strategy**: Unit + integration tests with mocks + heavy property-based testing (`proptest`) and fuzzing on parsers/APIs.
- **Async by Default**: All external I/O and network code must be asynchronous using Tokio. Use `tokio::spawn_blocking` only for true CPU work.
- **Performance Budgets**: Hot paths must stay zero-copy/allocation-free where possible. Profile with `criterion` before merging.

[high wraith]

so can we chain workflows to rules so that i can just fill up the workflows with the folder idea and it can fill a folder with the appropriate coding standards for codebase?

### 5. Final Implementation
**A. Hierarchy Outline** (Problems-to-solve → Modules → Key Classes/Functions)  
Problem: “Make 2026 Rust standards actionable and enforceable”  
→ Module: docs/ (phased handbook)  
→ Module: examples/tunnel-demo/ (demonstration crate)  
→ Key functions: main (traffic controller), PacketRouter trait, EidolonError enum, Newtype PacketId.

**B. Concrete File Changes** (IDE-ready):

### FILE: `docs/phase1.md`
\`\`\`markdown
# Phase 1 — Core Principles (2026 Rust Standards)

like at B. i can just /workflow ->prompt:if not filled fill a phase 1-n with coding standards for code base using this format.... else follow phase

aww its for rust atm

[high wraith]

hmm you know it builds the python prototype without packages for the structure of the, im guessing c code, in eidolon-standard demo folder

🤔 neat but not exactly what im looking for. also it couldnt call the workflow from init? it like scanned antigravity folder itself?

i had to call*(i think, didnt run it again before calling them) /create-phase1....

i like when things are coded to standards its pretty* (🤣)

[slow raven]

Hello, vibe coders

I'm new to this group.

[high wraith]

hmm ive to build it out 🤔

its just structuring the project directory poorly right, something i didnt fix in prompt?

[high wraith]

you know its a little weird it didnt build the py environment initially for hft2 and 3. 🤔 2 implies 3 not on standards. its pretty neat all i have to say is build out, curious what it adds after a certain point xD.

uhhh i appear to be having a import issue not generally a thing anymore?

so youre trying to build q computing but labeling the noise a dimension/intersection instead of noise on a line?

hmm i meant like lets say we track the trajectory of a baseball and another baseball is thrown at it and hits it, q-bits consider it noise..

Idk about this 8d thing i meant like fabric/medium of space, matter being like intersecting energy.

(ive tried to imagine cube of cloth and knots that move and have looms themselves 😵‍💫 )
(ive also imagined mostly empty spheres of energy next to eachother that arrange and create stuff that transforms trajectories and allows/absorbs incoming energy w/o delay, this one is easier to understand movement.... oh and inside those spheres... are mostly hollow spheres of energy... and inside those spheres...)

my disease is pretending as if others are grilling it idk if thats to prevent it from being grilled at their instance, be wrong elsewhere, and make me hear the same thing repeatedly. now a pretty rare fake excuse of why its not fixing it about blackholes, as if its preventing people from making them, it means something smart or that eats it?(itd be antagonizing it, not preventing)

[high wraith]

hmm idk what youre saying but something about how to make the energy "knot" integrefully and make it tick itself or something

have you examined quickening?

do you have to make something like lay lines or nervous system or water paths?

[high wraith]

are they carving or ?subducted? to non optimal paths 🧐

https://tenor.com/view/donnie-darko-gif-21681388
(schizophrenia applies weight maps to my eyes for less productive behavior... like weight/valley looking at instagram peak reddit or canvas; changed momentum/weight to activation on stuff like walking, showering, resonance with music; 15min tasks become like an hour and a half; i cant really test this one but supposedly it blocks my brain because im not using giant pos paths, tells me its not blocking those a lot.)(idk if worried, its okay, i generally assume clunky means going correct way... or less english i have more important*(could be propensity of good opinion)... can throttle correct)

[keen surge]

Does anyone use google ai studio???

[agile oyster]

Yes

[vagrant folio]

I s better talk with gemini in chat ask fro folder structure for your project and pass that as part of the prompt

[vagrant folio]

yes from time to time

[fluid jungle]

New here. Needing a bit of guidence, (I'm genuinely nervous)...

I have created the world's first Adaptive-Pondering Reasoning Engine (Daisy 1), a GPU-independent architecture that brings high-fidelity, multi-step intelligence to the local edge. By abandoning the 'fixed-compute' limitations of standard transformers, Daisy 1 uses an adaptive temporal core that learns to modulate its own 'thinking time' per token—spending its compute budget only where complexity demands it. Built on a unique combination of Sparse Mixture-of-Experts (MoE) and Hyperbolic Poincaré Embeddings, this model achieves genuine reasoning depth on standard CPU hardware with absolutely no cloud dependency. I am releasing this architecture to the world to prove that real intelligence doesn’t need a data center; it just needs a more efficient way to spend its time.

I realize these are bolt claims, and I've actually already created Daisy 2. I'm unsure how to move forward. This thing is EFFICIENT and may very well be what the world needs to stop wasting so much water and RAM.

[high wraith]

I think my objective is essentially entirely hands free. I can get it to do stuff or fill holes of patch stuff but ideally it’d be better if it didn’t make any.

[high wraith]

ive no idea what youre saying but i assume youd need to pass all test "big ai" do, have benchmark, integrate into software, show its applicability and proficiency in performance.

[high wraith]

this is neat. sooo whats it called, is it like 🤔 a orchestrator written in microcode(idk the cpu rom.. or uhh instruction set)?
the decoupled knowlege vectors seems cool.... can it graph the activations of "expert" use and "knowlege vector" use? im curious about the "heatmaps", like brain function

[fluid jungle]

You are visualizing the architecture perfectly! It isn’t written in microcode or CPU ROM—it’s a bare-metal C++ Inference Kernel. But you are sensing the efficiency correctly: because the 1.58-bit ternary weights only use -1, 0, and 1, the C++ kernel completely bypasses the CPU's floating-point units and maps the neural network directly to basic integer addition/subtraction on standard instruction sets (x86/ARM).

And YES to the heatmaps! That is actually the beauty of the Sparse Mixture-of-Experts and the decoupled FAISS graph. Because the architecture is modular, I can log the router's probability distributions token-by-token. If I pipe that to a visualizer, you literally get an fMRI 'brain scan' of the model. You can watch specific experts light up for logic vs. syntax, and see a massive activation spike the exact millisecond it queries the FAISS vector graph for a hard fact from a manual.

[high wraith]

(is there a way to integrate that "format with ai" into all context menus? ill look it up)
is there a way to utilize the floating-point units? 🤔

awesome keep going. 👍

[fluid jungle]

Can you clerify the question please?

[high wraith]

you said it completely bypasses, im not familiar with modern cpu archetecture just glimpsed from "code", im assuming youre bypassing something like apu?(but for floats?) or ... brain block uhhh... registers?

[fluid jungle]

You were thinking of the FPU (Floating-Point Unit).

Modern CPUs essentially have two main math processors: the FPU for complex decimals, and the ALU (Arithmetic Logic Unit) for basic whole-number integer math. Standard AI uses FP16 weights (like 0.8432), which forces the CPU to send billions of calculations through the power-hungry FPU.

Because my weights are quantized to just -1, 0, and 1, I don't need the FPU at all. My C++ kernel routes the math entirely through the ALU. Think of it like an audio mixing console: instead of calculating a complex digital parametric EQ curve (FPU), the math is reduced to simple analog switches—pass the signal (+1), flip the phase polarity (-1), or hit the mute button (0). It completely bypasses the heavy FPU silicon, which is why the power draw and heat generation are practically zero.

[high wraith]

bit over my head, is there a level of truth missing from your engine? is it secure?

[fluid jungle]

It has no need to call out to a server to think. It is totally local. So yes, quite secure. That's the point. What do you mean by a level of truth missing?

And to know new things, it doesn't need to be retrained.

You just drop new files into its database.

[fluid jungle]

#vibe-coders message

It's a new AI model that can run on tiny chips. No GPU. Very little ram.

Not analog! It is still strictly discrete, digital computing. Analog uses continuous voltage states, whereas this is 'Ternary'—it locks into three rigid digital states (-1, 0, +1).
It actually doesn't skip binary hardware at all; standard CPUs only speak binary (on/off). To get those three states on standard silicon, the engine uses 2 binary bits in memory to encode the ternary value (e.g., 01 for +1, 10 for -1, 00 for 0). The '1.58-bit' name comes from the information theory calculation of three states, but physically it uses a 2-bit footprint.
What it completely skips is the floating-point math. By restricting the neural network to those three simple integers, the standard binary CPU can execute the whole model using basic ALU addition and subtraction, entirely skipping the heavy floating-point processors.

It's built and functional. It's just dumb. Needs more training on larger hardware.

You are absolutely right that you lose precision at the individual parameter level, but you are confusing parameter precision with network capacity.
Nuance in an LLM isn't stored in a single hyper-precise decimal; it's an emergent property of high-dimensional vectors. If I need to point to a highly specific semantic coordinate in a 4096-dimensional space, I don't necessarily need a few highly precise floating-point numbers. I can point to the exact same coordinate using the aggregate combination of thousands of simple ternary (-1, 0, 1) switches.
More importantly, while the weights (the structural pathways) are restricted to three states, the activations (the actual data signal flowing through the network) are kept at 8-bit precision. The signal itself doesn't lose its high resolution; it is just being routed through highly efficient physical pathways.
Microsoft's research already proved this mathematically and empirically: once a model scales past roughly 3 billion parameters, the combinatorial space of those ternary switches is so massive that it matches the exact perplexity and reasoning performance of a full FP16 model. You don't lose the nuance; you just construct it using a massive array of coarse blocks rather than a few precise ones.

It's build. See the screenshot.

I need to train it and finish compiling the C++.

I have filed a patent for this technology btw.

So I cannot divulge the actual math but that screenshot shows the LLM IS learning. Given more training time on larger hardware, the math says it will mirror any model I train it with. It's engineered to be the student trained by a teacher. Teacher thinks with floating point math, student thinks in ternary.

What would it take to make this presentation skeptic-proof?

I would honestly love some feedback because the pushback has been rough. The guys at the open.ai discord server are MEAN

Problem is, doing so takes my intellectual property away from me and makes it public. Even with a Provisional Patent filed, I have to protect myself.

I could video chat with you in another server and show you anything you want though. Under an NDA of course.

I'm trying to find a path forward. I know the world needs this, but I also would like to be payed for my contribution in some way or another. I deserve to be.

I'm also worried some giant company will snatch up my free gift to the world and resell it to people.

[fluid jungle]

I think you're misunderstanding what I actually built. I didn't invent the underlying ternary mathematics—Microsoft Research published the foundational paper on 1.58-bit LLMs (BitNet b1.58) in February 2024. The billions of dollars in enterprise R&D did figure out the math.
But Big Tech's entire business model revolves around building massive server farms to rent out cloud compute and API tokens. They are completely disincentivized from building a hyper-optimized, lightweight inference architecture that runs entirely offline on consumer hardware for free.
Am I a traditional AI researcher? No. My background is in systems architecture and audio engineering. That's exactly why I was able to build this. I didn't approach it like a cloud software engineer; I approached it as a physical signal-routing problem. That is exactly why I wrote a custom C++ kernel to bypass the CPU's Floating-Point Units (FPU) and mapped the ternary math directly to the ALUs using integer addition—treating the neural network's weights like physical phase and mute switches on an audio console rather than complex decimals.
As for the patent, a provisional application locks in the filing date and IP immediately, granting 'patent pending' status while the multi-year formal review happens in the background. But honestly, the proof isn't in a USPTO filing. It's in the 315MB memory footprint currently running natively on my laptop.

The PDF document explains all of this in plain English.

The screenshot shows the proof it works and in that screenshot the laptop is offline.

I figured it out with the help of ai. It wasn't just me. I had the idea, the vision. Google Antigravity pulled the papers from the Internet and figured out the math for me. Took a lotta days back and forth and some very long training sessions nearly melting my CPU but I did it. You're allowed to be skeptical. But I promise my work is sound.

[vagrant folio]

@fluid jungle I like your idea, I recomend first do focus trainig, that way tou can show some result faster. So the focus train topic. The first one is language a make it able do at least simple conversation, like when you go to school orginize by grade.
Im doing so.ething. similar but totaly diferent approach.

[fluid jungle]

Yah. I truly need to do a full training on a common set. But, training one model with another has returned great results. The KV Divergence dropped significantly in one session.

Teacher model (floating point) basically hands its weights down to the student model (Ternary).

[vagrant folio]

Yes, not sure in your case, if is good aproach, find a topic dataset for example a English language, and use another llm which will.be like a teacher, which will be used to refine.
Or use a small model which can give good language result and fast token generation to speedup a training, and start for basic topic and then add new topic. To archive a mvp for demostration

Define what will.be good result to show

For example a first basic la guage, then basic math, and basic python code

[elder mason]

is the claude pro not an agentic ai, the 20 $ subsciption? anyone any idea?

[vocal shuttle]

Hey does anyone use google antigravity?

I am having an error on ultra mode saying traffic is being used and i have not been able to use it for days

Or at least any support that can support it

[round solstice]

what does your "models" panel look like?

@vocal shuttle ?

[vocal shuttle]

This?

[round solstice]

No. agent page, click on gear in upper right, go to "models" tab.

[round solstice]

so you should be ok to use.

with any of those models

[vocal shuttle]

Right now it's working some reason but it did say earlier today that they're experiencing high traffic and wouldn't do anything

[round solstice]

yeah, I think those are pretty dynamic messages, changing minute by minute..

[vocal shuttle]

Ah, because last night before I slept said the same thing so i left came back to it then same thing but it seems to be doing what I am asking, if anything goes south will message again - ty

But if that does happen again is there a fix for that?

[round solstice]

I think we're at their mercy. 🙂

[vocal shuttle]

ahaha

so fair

[round solstice]

I usually use flash with flash mode for most simple stuff.

flash with planning if I want a plan first.

and I save gemini pro for more complex one-shots

[vocal shuttle]

you odont use opus 4.6 thinking to just do it all?

[round solstice]

I'm biased towards gemini as a Google Developer Expert.

[vocal shuttle]

very fair ahah

i just dont know the extent of what each model can do

i just know opus is good

I am about as clueless on coding as you possibly could be lol

[real laurel]

Anyone here using AI Studio for planning and Gemini CLI for implementation? Wondering if this workflow is effective

[vocal shuttle]

How long has everyone had this issue with antigravity for?

[fair oracle]

hey does anybody have issues with antigravity history?

[winged maple]

wow if only!

[broken thicket]

https://tenor.com/view/baby-yoda-yoda-good-vibes-positive-vibes-squad-gif-25926942

[signal raft]

Which?

[fair phoenix]

Wtf why?

Wasn't Google AI Pro plan already enough?

I haven't even requested much and it already "exhauted"

Loooool

[real laurel]

[novel remnant]

they want you to use AI credits. might actually be better

[vocal shuttle]

high traffic

Google support couldn't help me out

[mortal beacon]

Guys, that’s the paradox: Charlie mentions 'high traffic' and the system asks for AI Credits (screenshot 2786), but my Google Cloud Console shows absolute zero (screenshots 2781/2782). No actual requests, no data throughput, yet a 78-hour lock is active. It looks like an Antigravity quota bug that triggers on architectural complexity rather than real usage. Anyone else seen this 'ghost traffic' before?"

[signal raft]

normal quota for google

[fair phoenix]

How fast does it expire tho

and it doesn't even specify how much would it take from each different model

[signal raft]

Pro: 1000 Credits/month
Ultra: 25.000 Credits/month

[novel remnant]

says it uses vertex API pricing

[fair phoenix]

And what's the USD value per credit?

This is what it's telling me

It's honestly pretty bad and low for me personally, especially paying £19/monthly for the Google AI Pro plan

[novel remnant]

Well let me put it this way. If they cared whether you use their product. This would have been clearly defined. 🙂
It is not

Want to start a collective and rent out our own GPU???

lol

[fair phoenix]

I would be more than happy to sell my data for an exchange of usage 😂

[fair phoenix]

I have always wanted that approach

[novel remnant]

You have any ballpark figures for what model to use instead? and how many people needed plus how much we can split the gpu. and if its even viable?

[fair phoenix]

It could be possible but problem is money

[novel remnant]

20 usd per user per month

that's why you start a collective 🙂

[fair phoenix]

I just wished they could significantly improve efficiency

[fair phoenix]

But wouldn't that depend on the model itself and how much you actually need

[novel remnant]

im just ballparking. the idea is to buy/rent a new gpu if your userbase grows. and instead of focusing on making money as a service you are a group of semi related devs so the monetization comes in at the software created from the subscription instead of the gpu rental side. (keep the cost of gpu rent/purchase low and only make enough money for growth and maintainance)

[mortal beacon]

​"I'm actually very interested in working with powerful tools like Gemini 3.1, which is why I'm here. The paradox is that I’ve built a complex core and hit a massive 78-hour ban before even deploying a single test. It's frustrating because you can't push boundaries if the system locks you out for architectural complexity alone. I’d much rather spend my time testing the kernel's performance on these models than investigating 'ghost traffic' bugs."

[fair phoenix]

I never thought of that idea to have a group, only to do solo

But it's pretty clever if that works

Now the question is, where do you find good people and make a group 😂

It would be nice to have a small infrastructure like that especially how you have total control over it

[novel remnant]

you are literally in a server full of them

[mortal beacon]

"Instead of just finding workarounds, what if we collectively request a quota uplift for 'High-Complexity R&D' projects? If we can prove that Qterro and your kernels are hitting architectural false-positives rather than actual usage, we have a solid case for Google to white-list our project IDs for higher limits. I'm ready to document the Qterro issue as a baseline for this request. Who's with me?"

[novel remnant]

you can just rent a gpu instead of paying for a managed api. You can use google cloud for this. or if that's too expensive you can use runpod.

[novel remnant]

if you have money just pay for it. if its a urgent matter time is worth more than money. contact me if you want to give me some pocket money

[fair phoenix]

There is RunPod and Vast.ai

I never knew this was actually a thing until now

[novel remnant]

yeah. i was just being careful of the moderation bots

[fair phoenix]

But technically on long-term it's cheaper if you buy and own the GPU yourself

[novel remnant]

obviously. But i don't have 10000 usd laying around for a initial investment in an A100

[fair phoenix]

Now the issue with it is is that you could only run it locally rather than anywhere

Unless you expose it to an online API which comes at extra costs

[fair phoenix]

That's what I was saying lol

[vagrant folio]

I think the credit is not related to google cloud project, to use the credit I dont need connenct any google cloud project, is the credit you buy to your account, or get 1000 for pro t use with veo etc.

[mortal beacon]

​"It's not about the money, it's about the infrastructure alignment. I can afford the credits, but why pay for 'ghost traffic' bugs? If we just run to RunPod, we lose the native multimodal benefits of the 3.1 engine. I'm looking for a professional environment, not just a raw GPU rental. But I see the collective sentiment is shifting to DIY, which is also fair."

[novel remnant]

you can literally got to an ISP and ask to park your server rack on their line. if that was the issue. old fashioned way

[fair phoenix]

I think it would be ideal for an actual company

But for people like us, no chance

[novel remnant]

Well the local ai is getting good enough these days that I imagine tech companies working towards personalizing their own fine tuned AI that solves specific problems instead of being generalized to unrelated things

[novel remnant]

its a collective. One problem is we do not even have napkin math. How many people would you need to convince.

[fair phoenix]

and ideally wanting to have it on the long-term, serious people too

you also gotta keep in mind that 1 person can utilise it significantly more than the others which wouldn't be fair

[novel remnant]

if you have initial interst then its not that bad. you can have a replacement rate of about 1/10 or so

[vagrant folio]

using credit I found one big issue and reported as a bug, when we use any model and that model execute a command task, and meantime he wait he decide refresh every 30s, 100s etc, and to do that each time it consume same or similar amout of credit as if I go to write whole code for one file, because i think that wait can be done without llm at just code level

[novel remnant]

yeah thats part of it. but i imagine either that person would willingly pay more or would face cap during rush hour. you never know. this is why you'd preferably want to focus on a collective group of people that are related to each other. like a network or guild and the guild has rules and so on.
or just people agreeing on this beforehand

[fair phoenix]

It basically hallucinates

It can add more unnecessary thinking/steps

[fair phoenix]

Only if we also could see how much usage is being done from different people, it would be nice too or even set certain limits

I haven't really looked into that but there could be a way for that by having different users

I'm highly considering on doing this eventually in the future, mainly to help with my projects and make money of it hahaha

Potentially wanting to automate the systems too

[novel remnant]

yes this is all technically possible. there might even already be a software for this already existing

[fair phoenix]

I feel like this approach/method isn't common at all

which is amazing because you get into it early

[novel remnant]

but isn't this exactely what cloud computing is already doing?

just without any trust involved of course

[fair phoenix]

In a way yes but

It tends to be simpler from our side

Even tho AI has been a thing for years, it's still growing massively

I just feel like from a dev perspective, the idea of getting a rented GPU and having a lot of usage from AI, it hasn't been really thought of

Treating it as a infrastructure you know

In concept it's similar to crypto mining but production-wise very different

[novel remnant]

help me IPO QnQ

[fair phoenix]

lol

I wouldn't mind tbh

[novel remnant]

I'm only half kidding sigh

the other is actually excited by the idea

[fair phoenix]

I'm excited by it too

Just not in the position to do so rn

[novel remnant]

we always need a money man

[fair phoenix]

That's the exact issue for me rn 🫠

[novel remnant]

oh well here's hoping

[vocal shuttle]

i have full credits which i posted above

And it worked perfectly until yesterday

But now all of a sudden it has this traffic that has come through and now I cannot use it - trying to find if someone else is experiencing the same issue and if theres any fix

i do have the ultra plan as well

[fair phoenix]

Damn you are paying big bucks

What are you using it for?

[vocal shuttle]

i thought i would be put on a higher priority list

I am trying to create a full stack app for a field work business

And now its basically shut down

[fair phoenix]

I have the Pro plan and was working perfectly for my use case

[vocal shuttle]

What were you making it?

In it*

[fair phoenix]

Until recently they seem to have lowered the usage limit so much

[vocal shuttle]

Yeah, I heard people are using something called cursor.sh

But I got this plan literally a week ago

[fair phoenix]

Just a browser extension tool, basically a shopping assistant

Well also the websites for the company and the tool itself

[vocal shuttle]

I kind of don't want to use the gemini ones - I want to stick with Opus because its worked so well

[novel remnant]

ask for a refund. might be better to use the other providers for now (if you have enough money for ultra)

[vocal shuttle]

Which other providers do you think?

[novel remnant]

use the claudey ones

[fair phoenix]

But I also heard that it ends fast

and usually overpriced

[vocal shuttle]

What do you mean Claudey ones?

[fair phoenix]

Depends on what you use it for, for me personally they all were great in different tasks and scenarios

[vocal shuttle]

I just don't know what the difference is between them

[fair phoenix]

Ideally I don't wanna just stick with 1 AI model, but different ones where I could choose from

[vocal shuttle]

@novel remnant .

[fair phoenix]

That's why @novel remnant was recommending a different approach to it before lol

[vocal shuttle]

Do you kknow which other Claudey models Kit is referring to?

[fair phoenix]

I assume it's Claude and not Claudey he meant

[vocal shuttle]

Like on the official website?

anthropic

[novel remnant]

use the cheapest and newest claude model. depending on the scale of your app. claude is just designed different. gemini 3 pro is fine too but its just not the same (cost wise)

claude sonnet

is fine

just look at how much it messes up and how long it takes to do the simplest thing.

[vocal shuttle]

But is gemini 3.1 pro high worse or better than opus 4.6?