Marko.xbt πŸ”₯ (@mar_ko3) header
Marko.xbt πŸ”₯ (@mar_ko3) profile picture

Follow the money, you will find the crime | bot: https://nitter.gq/t.co/BoDMPfgD4e

Trenches
Joined January 2025
Marko.xbt πŸ”₯ retweeted
He built NV-01, a headless two-legged robot meant to carry packages from a delivery van to your door. Right now it can only stand on its own, squat, shift its weight and recover. Walking, stairs and actual deliveries are still plans. Dev previously worked at Nvidia. Tech is insane. He said he’s making progress and he’ll be installing the hands on a bit He is onboard
HOLY F*CK... i built a robot that walks your packages to your door. meet NV-01, an autonomous delivery robot. β†’ legs built from custom actuated joints at the hips, knees and ankles β†’ a lightweight open frame that keeps the weight low and the balance tight β†’ balance control that holds it upright while it shifts weight, squats and recovers β†’ arms that carry the load against its chest, so the package stays steady while it moves β†’ trained to handle real ground, not a perfect lab floor the plan is simple: the van parks, NV-01 steps out, takes the package to your door and walks back. where it’s at today: it went from a frame hanging on a gantry to standing on its own. these are its first reps: squats, weight shifts, and getting back up without falling. next comes walking, then stairs, then the first real doorstep. day one of building in public. Follow the build as we add new components,
9
9
5
26
3,118
Dev is fully doxxed and he will be tweeting in the comm soon HSekCAUcfwZWjvBjhjJsgYAUc5ZqJ98LmBcC5jjhpump
He built NV-01, a headless two-legged robot meant to carry packages from a delivery van to your door. Right now it can only stand on its own, squat, shift its weight and recover. Walking, stairs and actual deliveries are still plans. Dev previously worked at Nvidia. Tech is insane. He said he’s making progress and he’ll be installing the hands on a bit He is onboard
1
1
5
385
HSekCAUcfwZWjvBjhjJsgYAUc5ZqJ98LmBcC5jjhpump $ADR
4
209
Welcome to the community Apex
3
207
Marko.xbt πŸ”₯ retweeted
Fire tech Learnt Dev is ex Nvidia Team spoke to him, he said he’s made plans to complete the arms today He’ll be in the community in a bit. Will def do numbers A headless two legged robot meant to carry packages from a delivery van to your door. Right now it can only stand on its own, squat, shift its weight and recover. Walking, stairs and actual deliveries are still plans. HSekCAUcfwZWjvBjhjJsgYAUc5ZqJ98LmBcC5jjhpump
He built NV-01, a headless two-legged robot meant to carry packages from a delivery van to your door. Right now it can only stand on its own, squat, shift its weight and recover. Walking, stairs and actual deliveries are still plans. Dev previously worked at Nvidia. Tech is insane. He said he’s making progress and he’ll be installing the hands on a bit He is onboard
10
16
1
55
3,231
He built NV-01, a headless two-legged robot meant to carry packages from a delivery van to your door. Right now it can only stand on its own, squat, shift its weight and recover. Walking, stairs and actual deliveries are still plans. Dev previously worked at Nvidia. Tech is insane. He said he’s making progress and he’ll be installing the hands on a bit He is onboard
HOLY F*CK... i built a robot that walks your packages to your door. meet NV-01, an autonomous delivery robot. β†’ legs built from custom actuated joints at the hips, knees and ankles β†’ a lightweight open frame that keeps the weight low and the balance tight β†’ balance control that holds it upright while it shifts weight, squats and recovers β†’ arms that carry the load against its chest, so the package stays steady while it moves β†’ trained to handle real ground, not a perfect lab floor the plan is simple: the van parks, NV-01 steps out, takes the package to your door and walks back. where it’s at today: it went from a frame hanging on a gantry to standing on its own. these are its first reps: squats, weight shifts, and getting back up without falling. next comes walking, then stairs, then the first real doorstep. day one of building in public. Follow the build as we add new components,
9
9
5
26
3,118
BackPack is here to save the trenches once again Backpack posted this Apu in the tweet below nitter.gq/Backpack/status/195409… The only BAPU with fees directed to holders 2peydwLTbBvZJGV41LKfbwnktNMxd7BbWvfJv3dCpump It is free here!!!!
Backpack is coming home to the USA. πŸ‡ΊπŸ‡Έ We’ve received landmark CFTC approval to establish the first-ever derivatives exchange powered by one order book with unified global liquidity. Leveraged perpetuals. Crypto collateral. Cross-margin. 24/7 trading. Join the waitlist: backpack.exchange/us LFG πŸŽ’
1
125
2peydwLTbBvZJGV41LKfbwnktNMxd7BbWvfJv3dCpump
35
Marko.xbt πŸ”₯ retweeted
I no believe say I fit later p2p $1
200
240
73
1,924
65,701
Marko.xbt πŸ”₯ retweeted
$Trendo is live CA:526bZaPjwLvwyieDqyHhg7myFd8pJeB53jM9zhmfpump Imagine scrolling through the trenches and finding your next alpha instead of wasting hours doom-scrolling TikTok That’s the idea behind Trendo.fun. Turn your scrolling into an edge. Discover emerging trends, viral narratives and opportunities before everyone catches on. Less doom scrolling. More alpha hunting.
10
2
27
491
Marko.xbt πŸ”₯ retweeted
I decided to take a break from my mental health to focus on crypto
212
105
22
1,340
38,855
Marko.xbt πŸ”₯ retweeted
HOLLOW HILLS IS LIVE. CA: 4GY7THbEym11dDDfLsM8HD6etHurWyZeVPaNUmr1pump Enter the hills. Face the undead. Survive the night. Welcome to Hollow Hills, a Halloween-themed survival game where every second counts and danger lurks around every corner. The dead have awakened, and the hills are no longer safe. β†’ Enter the haunted hills β†’ Face ghosts, zombies, and skeletons β†’ Fight to survive the night β†’ Escape before it's too late Every moment is a fight for survival. Every decision brings you closer to safety or death. This isn't just another Halloween game. It's an experience built around horror, chaos, and community. The gates are officially open. Think you can survive Hollow Hills? Play now: hollowhills.fun/
5
2
8
507
Marko.xbt πŸ”₯ retweeted
Creators deserve to launch without getting vamped. Traders deserve to find the original without digging through copycats. That's why we're building Vampless. Official CA: 5tGjUKr9qoAhuuoHikrchA9rzbrUjfrV1nRyXjAevamp Most launchpads make it easy to deploy a token. Very few address what happens when someone copies its name, ticker, branding, or narrative and launches another contract to siphon attention and liquidity. We're approaching this differently. Vampless is building a token launch infrastructure focused on identity protection and authentic discovery. The idea is simple: β†’ Establish a verifiable identity for tokens launched through the platform. β†’ Detect conflicting identities and potential impersonation attempts before they reach traders. β†’ Give creators a recognizable origin for their launches. β†’ Help traders distinguish original launches from tokens attempting to imitate them. We're not trying to prevent competition or stop anyone from creating tokens. We're addressing the confusion created when multiple contracts deliberately present themselves as the same project. Solana is where we start. The next generation of launchpads shouldn't just make token creation easier. They should make token identity harder to impersonate. Vampless. Built for originals.
9
4
1
23
1,001
Dev is followed by Alon, Moonpay and a16z Bonded $Vampless 5tGjUKr9qoAhuuoHikrchA9rzbrUjfrV1nRyXjAevamp
Creators deserve to launch without getting vamped. Traders deserve to find the original without digging through copycats. That's why we're building Vampless. Official CA: 5tGjUKr9qoAhuuoHikrchA9rzbrUjfrV1nRyXjAevamp Most launchpads make it easy to deploy a token. Very few address what happens when someone copies its name, ticker, branding, or narrative and launches another contract to siphon attention and liquidity. We're approaching this differently. Vampless is building a token launch infrastructure focused on identity protection and authentic discovery. The idea is simple: β†’ Establish a verifiable identity for tokens launched through the platform. β†’ Detect conflicting identities and potential impersonation attempts before they reach traders. β†’ Give creators a recognizable origin for their launches. β†’ Help traders distinguish original launches from tokens attempting to imitate them. We're not trying to prevent competition or stop anyone from creating tokens. We're addressing the confusion created when multiple contracts deliberately present themselves as the same project. Solana is where we start. The next generation of launchpads shouldn't just make token creation easier. They should make token identity harder to impersonate. Vampless. Built for originals.
1
192
This is crazy, pumpfun now has a strong competition A launchpad that does not allow vamping, now you can launch with security 5tGjUKr9qoAhuuoHikrchA9rzbrUjfrV1nRyXjAevamp
Creators deserve to launch without getting vamped. Traders deserve to find the original without digging through copycats. That's why we're building Vampless. Official CA: 5tGjUKr9qoAhuuoHikrchA9rzbrUjfrV1nRyXjAevamp Most launchpads make it easy to deploy a token. Very few address what happens when someone copies its name, ticker, branding, or narrative and launches another contract to siphon attention and liquidity. We're approaching this differently. Vampless is building a token launch infrastructure focused on identity protection and authentic discovery. The idea is simple: β†’ Establish a verifiable identity for tokens launched through the platform. β†’ Detect conflicting identities and potential impersonation attempts before they reach traders. β†’ Give creators a recognizable origin for their launches. β†’ Help traders distinguish original launches from tokens attempting to imitate them. We're not trying to prevent competition or stop anyone from creating tokens. We're addressing the confusion created when multiple contracts deliberately present themselves as the same project. Solana is where we start. The next generation of launchpads shouldn't just make token creation easier. They should make token identity harder to impersonate. Vampless. Built for originals.
129
πŸ”₯πŸ”₯πŸ”₯πŸ”₯
Okay i hear you, every single dollar generated from fees goes straight into buybacks. and i'm adding another $1,000 of my own money to buy from this moment and every 2 hour putting my money where my mouth is.
129
The dev is really committed and relentless on this projwct Buyback from his creator fees and also keeping the project alive Let's all support him, this is the official CA 0x6a84b9c10f18b36a50c05f824bef3ce15690342f
MAC v0.2 - what shipped since launch Comms upgrade, so here is everything that changed under the hood, in one place. Routing The gate used to score a task once and commit. Now it re-scores after classify, so a task that looked cheap but arrives with 400k of context does not get stuffed into an 8k window and silently truncated. Tier ladder is explicit: nano β†’ small β†’ frontier. A task only climbs if the cheaper tier fails the eval, never because the prompt "feels hard". Dedupe Window went from exact-match to hash(intent + normalised args), 30m rolling. That caught the thing I did not expect: the same question arriving from two different agents in the same minute, worded differently. Those used to be two frontier calls. Batching 90s hold window, fan-in on identical intent. Added a bail-out: if anything in the batch is marked irreversible, the whole batch breaks apart and each item routes on its own. Batching is for reads, never for writes. Evals 500-case held-out set, and the promotion rule is now hard-coded instead of being my opinion: a cheaper tier ships only if it stays within 0.4pp of baseline. Shadow β†’ 5% canary β†’ default. Nothing goes straight to default anymore. Logs Every decision is replayable: inputs, the six answers, the route, the confidence, the latency. You can take any decision id and re-run it against the current policy to see if today's config would have made the same call. That is the part I actually use every day. Failure modes Fail-open: controller down means traffic passes untouched, not blocked. Idempotency keys on intake, so retries do not double-charge. Backpressure instead of letting queue depth grow quietly. What this cost [FEES CLAIMED] in creator fees claimed and spent on infra, model credits and the eval harness. [BUYBACK TOTAL] routed to buybacks so far, every one posted with its tx. Measured since v0.1 [TASKS ROUTED] tasks through the controller [% NEVER REACHED A MODEL] never reached a model at all [SPEND BEFORE] β†’ [SPEND AFTER] [QUALITY DELTA] on the held-out set Next β†’ routing logs public, so anyone can replay a decision without asking me β†’ first connector, so this runs on someone else's queue and not only mine β†’ per-intent policies, because support tickets and code review should not share thresholds The models were never the bottleneck. The routing was. $MAC
1
6
303
Marko.xbt πŸ”₯ retweeted
MAC v0.2 - what shipped since launch Comms upgrade, so here is everything that changed under the hood, in one place. Routing The gate used to score a task once and commit. Now it re-scores after classify, so a task that looked cheap but arrives with 400k of context does not get stuffed into an 8k window and silently truncated. Tier ladder is explicit: nano β†’ small β†’ frontier. A task only climbs if the cheaper tier fails the eval, never because the prompt "feels hard". Dedupe Window went from exact-match to hash(intent + normalised args), 30m rolling. That caught the thing I did not expect: the same question arriving from two different agents in the same minute, worded differently. Those used to be two frontier calls. Batching 90s hold window, fan-in on identical intent. Added a bail-out: if anything in the batch is marked irreversible, the whole batch breaks apart and each item routes on its own. Batching is for reads, never for writes. Evals 500-case held-out set, and the promotion rule is now hard-coded instead of being my opinion: a cheaper tier ships only if it stays within 0.4pp of baseline. Shadow β†’ 5% canary β†’ default. Nothing goes straight to default anymore. Logs Every decision is replayable: inputs, the six answers, the route, the confidence, the latency. You can take any decision id and re-run it against the current policy to see if today's config would have made the same call. That is the part I actually use every day. Failure modes Fail-open: controller down means traffic passes untouched, not blocked. Idempotency keys on intake, so retries do not double-charge. Backpressure instead of letting queue depth grow quietly. What this cost [FEES CLAIMED] in creator fees claimed and spent on infra, model credits and the eval harness. [BUYBACK TOTAL] routed to buybacks so far, every one posted with its tx. Measured since v0.1 [TASKS ROUTED] tasks through the controller [% NEVER REACHED A MODEL] never reached a model at all [SPEND BEFORE] β†’ [SPEND AFTER] [QUALITY DELTA] on the held-out set Next β†’ routing logs public, so anyone can replay a decision without asking me β†’ first connector, so this runs on someone else's queue and not only mine β†’ per-intent policies, because support tickets and code review should not share thresholds The models were never the bottleneck. The routing was. $MAC
3
3
1
8
435
This is great and what are the features you are trying to implement next to make this better
I built Multi-Agent Controller - a control plane that sits in front of every agent you run. The problem is simple: everyone is shipping agents, nobody is shipping the thing standing in front of them. Your queue dumps every task straight into the biggest model you have. Multi-Agent Controller sits one layer above that. β†’ intake β†’ classify β†’ gate β†’ route β†’ execute β†’ log Six questions per task, 0.21 ms, no generation in the hot path: is the answer worth anything is the information new cheapest tier that still passes the eval can it wait 90s for a batch is it irreversible confidence Then one route: cheap_tier / batch / dedupe / frontier / drop / human Tiers are priced, not vibed: nano 0.05 / small 0.40 / frontier 12.00 per Mtok Dedupe window 30m on hash(intent+args) Batch window 90s Drop if value < 0.15 and no new info Human if irreversible or out of policy A cheaper tier ships only if it stays within 0.4pp of baseline on a 500-case held-out eval. Shadow β†’ 5% canary β†’ default. Fail-open if the controller dies. Idempotency keys on intake. Every decision replayable. Demo harness, one day of agent work: 8,400 tasks 16.8% never reached a model at all $184.60 β†’ $21.40 quality 98.6% vs 98.9% Simulated run, labelled as such. Live numbers start with the first real queue. Multi-Agent Controller The models were never the bottleneck. The routing was. github.com/0xRickerX/multi-a…
1
7
240
Marko.xbt πŸ”₯ retweeted
I built Multi-Agent Controller - a control plane that sits in front of every agent you run. The problem is simple: everyone is shipping agents, nobody is shipping the thing standing in front of them. Your queue dumps every task straight into the biggest model you have. Multi-Agent Controller sits one layer above that. β†’ intake β†’ classify β†’ gate β†’ route β†’ execute β†’ log Six questions per task, 0.21 ms, no generation in the hot path: is the answer worth anything is the information new cheapest tier that still passes the eval can it wait 90s for a batch is it irreversible confidence Then one route: cheap_tier / batch / dedupe / frontier / drop / human Tiers are priced, not vibed: nano 0.05 / small 0.40 / frontier 12.00 per Mtok Dedupe window 30m on hash(intent+args) Batch window 90s Drop if value < 0.15 and no new info Human if irreversible or out of policy A cheaper tier ships only if it stays within 0.4pp of baseline on a 500-case held-out eval. Shadow β†’ 5% canary β†’ default. Fail-open if the controller dies. Idempotency keys on intake. Every decision replayable. Demo harness, one day of agent work: 8,400 tasks 16.8% never reached a model at all $184.60 β†’ $21.40 quality 98.6% vs 98.9% Simulated run, labelled as such. Live numbers start with the first real queue. Multi-Agent Controller The models were never the bottleneck. The routing was. github.com/0xRickerX/multi-a…
3
3
6
376