Mojunel.eth

34 | research AI since 2021 | ex Binance

[meya. @birmeya]

Borderline insane, a whole Claude Code ops team in 10 repos.

Claude Code on Max costs $100 a month. This kit is the open-source layer that keeps it from re-learning your repo every session.

learn. remember. watch. run. install.

01 skills → sets the skill folder standard, 192.5k stars
https://t.co/Guqdl82Qto

02 claude-code-best-practice → maps every feature, subagent and hook
https://t.co/xmJS6kMKlo

03 Claude-Code-Quest → teaches Claude Code with three levels of graded exercises
https://t.co/0kfQPad5C6

04 claude-mem → keeps what one session learned alive for the next one
https://t.co/tT1QFxcedI

05 deja-vu → indexes transcripts of 43 agents, hands sessions between them
https://t.co/xbiee8biwt

06 ccusage → prices every token run from local logs, nothing uploaded
https://t.co/zRUPkdciHE
07 ccstatusline → shows model, branch and tokens live
https://t.co/Or6UM3WCxn

08 agent-browser → drives a real browser from Vercel
https://t.co/XgnHhqKWOU

09 opcode → runs sessions and agents from one window
https://t.co/CzTsQAupQZ

10 skills → drops one skill into Claude Code and 75 more agents
https://t.co/6bOFGPRqD1

learn → remember → watch → run → install.

Claude-Code-Quest runs PolyForm Noncommercial and opcode runs AGPL-3.0. Neither is free to ship commercially. Read both licenses before the code hits production.

All 10 repo names drop into the feed and the last skill slot clicks shut

[Mojunel.eth @mojunel]

@0xNeuroBlue 👀

[Mojunel.eth @mojunel]

I'm still in shock, the CEO of Y Combinator open-sourced his agent brain on GitHub, 32.4K stars, one clone away

7 second brains you probably never noticed, all ready to clone right now:

1. gbrain · 32.4K stars
Garry Tan's personal agent brain. every fact tracks its origin, and every agent taps the same memory
https://t.co/zLTcodasyO

2. Karpathy's llm wiki · 50K stars
the single-page idea file that kicked off this entire wave
https://t.co/zh5JBhK6cB

3. ai memory · 13.7K stars
leave Claude Code halfway through and resume inside Codex with nothing lost. memory sits in plain markdown
https://t.co/AOzWqCqiEL

4. COG second brain · 1.4K stars
35 skills and 10 agents, no database. just markdown files that reason
https://t.co/5XcyyFpoka

5. agent second brain
fire off a voice note through Telegram and it arrives as a linked entry in Obsidian, or a task inside Todoist
https://t.co/04IxEH6Mg1

6. open second brain
each night it converts the fixes you keep making into rules your agent obeys
https://t.co/OEGOtVjdqz

7. claude second brain
one npx command wires up Obsidian, Claude Code and local search in a single pass
https://t.co/QO7ZwSQ6zG

seven repos sitting in the open. your agent still loses you by morning

save this

[Ridark @ridark_eth]

https://t.co/9zfcxvi8yJ

[Mojunel.eth @mojunel]

@0xNeuroBlue true?

[Mojunel.eth @mojunel]

$55 to $1,724 overnight, zero screen time.

I set Grok loose with $55 and 3 open repos at bedtime. It ran 114 trades on autopilot. 10 hours of sleep, $1,724 in profit.

04:35, dead to the world. Grok spotted a coin ripping +240% in 5 minutes and threw $28.92 at it.

Sold 3 minutes later. +$45.74.

Found out over breakfast.

0 manual clicks the whole night.

GROKSCAN uses toasteragent. Grok from xAI finds the hottest new launches on https://t.co/YWQgI1cUaR.

SNIPER uses Chainstack's https://t.co/YWQgI1cUaR bot, 928 stars. Catches new mints, no third-party APIs.

EVAL uses agentinu. AI rates the risk at 5/10 on that coin before anything executes.

MONITOR locks take profit and stop loss on its own. SWAP triggers the auto-sell.

What I kept coming back to. 4 out of 5 agents agreeing still falls short. Nothing moves until it hits 5/5.

5 years in trading showed me the worst calls land at 4am, tired and reckless. The bot has neither problem.

22:00: $55. 04:45: $447. 08:00: $1,779.

Top trade +$396.18. Worst hit -$58.80. 81% win rate on 12 coins.

The stack:

https://t.co/9lCVeXLYex
https://t.co/8Ym1FWEgaX
https://t.co/eaQOVqTsA8

All open source. Save this.

[Ridark @ridark_eth]

https://t.co/9FxFn9jbXe

[Mojunel.eth @mojunel]

This is f*cking insane...

Stop paying Claude Opus $6,880 to read your own documents.

XERJ Reader is built for one job:
find the exact answer in a document, show the evidence, or admit it’s not there.

You give it a file and a question -- it highlights the exact line with the answer, or says “no answer.” It can’t make up a number. It can only return text that is already in your file.

The same 1M questions:

→ GPT-6 Astra: $17,200
→ Claude Opus 5.5: $6,880
→ Claude Haiku 5.5: $172
→ XERJ Reader: $6.50

$0.01 per 1M input tokens, output is free.
1,050x cheaper than Opus. 26x cheaper than Haiku.

The answer lands in 80 ms. Opus needs about 5 seconds just to start typing.

Not every AI task needs a frontier model. Sometimes you just need the right answer at the lowest possible cost.

I put the whole model on one page -- bookmark it before your next API bill.

Get on the waitlist while it’s open ↓

[Ivan Novikov @d0znpp]

I've built a new AI model that is not allowed to generate text. Instead, it produces a verifiable program that answers your prompt using the original reference.

XERJ Reader-1 is a neurosymbolic model for working with data. It points to the exact words in your document, writes a small program over them, and exact code computes the answer:

• 20-90x faster
• 10-1,000x cheaper (output tokens free)
• every answer is a program you can replay

The shortest path to automating data work.

[Ridark @ridark_eth]

YOU’RE OVERPAYING FOR AI BY 1,000X.

I found a tool called XERJ Reader, and it doesn’t generate a single word.

Here’s how it works:

1. Drop in your data: PDF, Word, email, JSON, CSV, logs. No schema, no setup
2. Ask a question in plain words
3. Reader scans the whole document in a single pass
4. You get the exact passage highlighted in the source, with its position
5. Not in the doc? It tells you “no answer” instead of making something up

It can’t make up a word it never writes.
The only answer it can’t give you is a wrong one.

And the math nobody wants to run is f**king brutal:
→ $0.01 per 1M tokens. Output is free
→ 1,050x cheaper than Opus 5.5
→ 26x cheaper than Haiku 5.5
→ Roughly 4x cheaper than Jev

Same question. Same file. Three models.
Opus writes you a paragraph.
Haiku writes you a shorter paragraph.
XERJ writes nothing.
It just underlines the sentence that was already there.

10 questions on a 100 KB doc:
Reader → under half a second
LLM agent → about a minute

The problem was never the model.
You’ve been paying for an essay when all you needed was a highlighter.

No hallucinations. No waiting. No huge bills.
Just the answer, highlighted.

Waitlist opens today. Bookmark this and be one of the first to start using it ↓

[Mojunel.eth @mojunel]

@0xNeuroBlue banger

[Mojunel.eth @mojunel]

Broke the cost floor with a team

$8 brain plans. $0.50 hands type.

I wired four agent files and each one carries its own model.

Fable is the brain.
Plans the work, splits it, judges what comes back.
Runs only when I type /brain.

Opus 5.5 takes the builder role. Builds from the brief, nothing else.

Sonnet 5.5 is the checker. Reviews the output. Edits nothing.

Haiku 5.5 runs the repetitive tasks and never touches code.

80x between the brain and the runner.

The brain delegates three kinds of work:

1. Independent parts
2. Work too big for one session
3. Repeatable tasks

One brain plans. Three hands deliver.

I describe the job
→ Fable writes the spec at $8 / $40
→ Opus builds at $4 / $20
→ Sonnet reviews at $2 / $10
→ Haiku runs routine at $0.10 / $0.50
→ Builder fails twice, Fable takes over

Never two helpers on the same file at once. Checker signs off before done.

One CLAUDE.md line: consult the advisor before the approach, on a repeating error, before calling done.

Agent teams in plan mode burn 7x the tokens of a solo session. Worth it only when the job outgrows one sitting.

Four files and one line

[Ridark @ridark_eth]

https://t.co/m2FHDbGXzh

[Mojunel.eth @mojunel]

@0xNeuroBlue hahahah

[Karmi @0xKarmi]

$11,000 or I kill the subscription. The bot blew past it

last post showed it clearing $11,040

you figured that was the peak?

it just hit $15,743.36

from a $19.63 starting deposit

sit with that for a second

i've been running an ai trading terminal built with Grok Bot, and six agents keep going nonstop, scanning tokens, spotting setups and pulling the trigger on trades

since that last check the balance on the terminal climbed another $4,702.78

here's the current snapshot

starting balance: $19.63

current balance: $15,743.36

total p&l displayed: +$15,723.73

win rate: 70.4%

resolved events: 2,533

and this isn't just a static number

the terminal tracks balance history, live activity, token scans, every execution and the back-and-forth between agents

six agents running inside a single environment

one watches the market, another catches setups, trades fire, and every event hits the log

wins, losses, liquidity scans, slippage, fresh pairs

all of it runs through the same terminal

the crazy part is how the whole thing got started

i didn't wire up a trading system from scratch by hand

i used Grok Bot to build the agents, the execution pipeline and the interface on top

now i'm pushing the experiment harder and logging every result as it lands

from $19.63 to $11k, then straight past $15.7k on the screen

the only question left is where the ceiling sits

[Mojunel.eth @mojunel]

A 21-year-old guy from San Francisco walked into an Apple Store, connected his Claude account to his phone right on the spot, and made $1,956 in just 15 minutes trading via Fable 5.2 without writing a single line of code

He opens Fable 5.2 right there in the store, types a quick prompt for an AI-powered trading strategy, and connects it to the live charts

Fable 5.2 builds the full setup in 3 minutes: real-time analytics, automated pattern recognition, and precise market entry signals

He launches the tool directly on his mobile device, tracks the live market spikes, and watches the profits grow trade by trade

While standing right in the store, he executes a series of successful trades using the AI-generated system, stacking up to $1,956 in profit in just 15 minutes

He closes the app, walks up to the store employee, and buys the brand-new phone he just earned money for right on the spot

[Joying.eth @Joyingeth]

NOT ON THE APP STORE. GROK BOT MADE THIS, I FELL ASLEEP, IT TURNED $347.86 INTO $3,048.56

from 11:47pm to 6:36am. 8.76x, 13 trades, i touched none of them

i told Grok Bot to make me a terminal to run off my phone. it put this one together, i dropped it in my Trading folder, tapped Run and went to sleep

4 agents inside
Scanner tosses anything short on liquidity
Analyst scores whatever gets through
Executor buys
Risk attaches a -6% stop to each buy

2:35am i opened my eyes and checked the app. same balance, positions firing open and shut on their own. went back to sleep

out of 13 sells, 10 hit green and 3 went red. those 3 lost $18.33 combined

top trade was a coin named scamsem, that name. +$566.46 in 46 minutes
and Diamond Hands, held true to its name, 1:27am to 5:00am, closed at 5.65x and 6.84x

8.76x is solid, but what i actually like is that this app belongs to me. nobody touches the rules except me

would you hand your money to an app that doesn't live on the App Store, or is that insane?

[Mojunel.eth @mojunel]

THIS GTA AI CLIP MAKES A PHYSICS-DEFYING TRANSITION LOOK LIKE A 30-SECOND PROMPT

what sells it is the continuity, not the AI glitch. you get realistic ambient lighting, accidental slip physics, matching clothing and crew-style framing before anything impossible happens.

3 different character interactions show up in under 50 seconds, but they all feel like they belong to the same street-level clip instead of random AI shots stitched together.

that opens a much better format than another cinematic clip: build seamless character loops and physics transitions around scenes that would normally need complex motion capture, stunt teams and hours of manual masking.

with @Pixverse, i can push the same concept through text, image, reference and transition modes while keeping the character identity consistent enough to feel like one continuous generation.

1 idea can become 4-6 impossible character loops, one behind-the-scenes workflow breakdown and multiple cuts for x, reels and tiktok. that is a much stronger use of AI video than just generating one pretty shot.

[Ridark @ridark_eth]

https://t.co/nbKGZnMx2m

[Mojunel.eth @mojunel]

COMPLETELY DERANGED.
Most teams still stack six paid tools for one funnel while 10 repos run a whole sales and revenue-ops team.

Zapier Pro charges $733.50 a month at 100,000 tasks. These ten cover the same pipeline with no per-task fee.

trigger. capture. pipeline. reach. measure. close.

01 n8n → wire triggers to every tool you run
https://t.co/ZbvgAAPHKq

02 activepieces → 280+ pieces work as MCP agent tools
https://t.co/f3odRtSKKM

03 formbricks → capture leads with in-app surveys
https://t.co/J9ZCglCkdG

04 twenty → Salesforce rival, open CRM built for AI
https://t.co/THs0rTIpKl

05 mautic → nurture leads with drip email campaigns
https://t.co/TKND05zieD

06 listmonk → send newsletters from a single binary
https://t.co/5ksjC893qQ

07 novu → route email, SMS, push and in-app alerts
https://t.co/mvmkxfkpvr

08 umami → measure traffic and conversions, no cookies
https://t.co/HDX6Q2ZVkX

09 documenso → get contracts signed without DocuSign
https://t.co/RIy2mLFgnQ

10 akaunting → invoice and book every closed deal
https://t.co/67e5XtAd6O

trigger → capture → pipeline → reach → measure → close. Half the list ships AGPL or source-available. Read the license before you resell a thing.

Six subscriptions gone.

[Ridark @ridark_eth]

https://t.co/il9WsxmfK5

[Ridark @ridark_eth]

One agent scoring its own find is a guaranteed yes.
Five that never see each other's work built this $100 desk.

I wired one prompt on Solana. Five agents, each doing one job. A coin moves left to right and any agent kills it.

1. Finder scans every new pair. Under $20k liquidity or past 6 hours: dead.
2. Auditor reads contract and holders. Top 10 above 35% or dev above 5%: dead.
3. Judge scores 0 to 100. Under 80: dead.
4. Stager sizes at 4% of the book. Sets the -25% stop before entry.
5. Closer sells half at double, quarter at triple, trails 30% under the high.

Seven rules sit above every score. 6 open positions, 3 entries an hour, -10% day kills the desk until midnight.

The Finder never knows what the Judge scored. The Auditor reads contracts, not scores. Something goes wrong and I get a name, not a story.

Rule 7 saved the desk. The creator pushed tokens toward the exchange and the Closer cleared at -6% in one block. The coin lost 91%.

[Mojunel.eth @mojunel]

YOUR KEYWORD TOOL IS THE PROBLEM

72 keywords entered the scan. 39 survived.

1. Jev pulls the top 10 Google results per keyword. 720 pages land in the queue.
2. Claude fires 8,640 typed judgments across all 720. 216 per second. 204 ms median per result.
3. Forum thread. Dead. Thin page. Dead. Outdated. Dead. Weak domain. Dead. Off-intent. Dead.
4. Keywords with fewer than 3 beatable results in the top 10 get cut. No appeal.
5. 33 keywords drop. 39 survive with 3+ soft slots each.

40.0 seconds. $0.0363 total.

54% of the 72 came back with a beatable spot.
44% of all 720 top-10 results sat on tier-3 domains nobody defended.

https://t.co/qHxPH0MGxR tutorial ranks first. Jev is early access, waitlist only.

Bookmark this

[Mojunel.eth @mojunel]

@0xNeuroBlue wow

[Neuro @0xNeuroBlue]

stared at a kill log past midnight: $100 flat, all red

2.5 hours of coins killed by three rules. Then at 00:31 the first $4 trade landed.

I fired the prompt at 22:00. Five agents, one rule each, every new Solana pair. The log filled before I looked back.

Every line was a kill. 00:12, liquidity $6,400. 00:19, coin age 11 hours. 00:24, top 10 holders at 52%.

Red lines stacking through the night.

By midnight the balance sat at $100.00. Not one coin survived. I nearly killed the setup.

At 00:31 one green line broke through. $4 entry, stop at -25% set before the fill, score past 80, three clean holders.

Day one ended at $118. Five trades, three wins, two losses. Every loss carried a report line that said why.

Reports landed twice a day. Kill reasons first, then stops and sizes, then the HONESTY line. Two minutes per read.

After day 8 the report pointed at the trail, not a broken rule. I changed one number: 30% to 25%. One edit in nine days.

One green line landing after 2.5 hours of red

[Mojunel.eth @mojunel]

YOUR KEYWORD TOOL IS THE PROBLEM

72 keywords entered the scan. 39 survived.

1. Jev pulls the top 10 Google results per keyword. 720 pages land in the queue.
2. Claude fires 8,640 typed judgments across all 720. 216 per second. 204 ms median per result.
3. Forum thread. Dead. Thin page. Dead. Outdated. Dead. Weak domain. Dead. Off-intent. Dead.
4. Keywords with fewer than 3 beatable results in the top 10 get cut. No appeal.
5. 33 keywords drop. 39 survive with 3+ soft slots each.

40.0 seconds. $0.0363 total.

54% of the 72 came back with a beatable spot.
44% of all 720 top-10 results sat on tier-3 domains nobody defended.

https://t.co/qHxPH0MGxR tutorial ranks first. Jev is early access, waitlist only.

Bookmark this

[Roan @RohOnChain]

Jev is the FASTEST AI model ever built for trading

It makes calibrated buy/sell decisions in under 100 ms

That is one real decision on every single block, 24/7

In this article I've shown EXACTLY how to build HFT trading system with Jev (from scratch) https://t.co/4hmZDzN00Z