HumanPlane

[HumanPlane @humanplaneco]

humanplane has been rebuilt and open sourced, we plan to be the best OSS option for folks looking to trade prediction markets, MIT licensed and fully customizable

go give it a peek

[nikshep @nikshepsvn]

spent months on @humanplaneco, a closed source @polymarket terminal. it did $250k+ in volume but i scrapped it anyways.

why? cause every polymarket terminal is the same thing. closed source. a reskin of https://t.co/AB5yqBkyy8 with marginal ux gains. polymarket already takes a 1–1.8% taker fee on every trade, so terminal fees are fees on top of fees you already pay. for a ui. the free ones are vc-funded — they won't stay free.

so i rebuilt humanplane as the first free, open source, fully customizable polymarket terminal. run it locally, forever. don't like something? ask claude to modify it and make it yours.

built w/ a performant rust backend and solidjs, supports market and trader exploration, live charts, orderbook and order execution (market + limit) -- intelligence, perps, trader scoring and more soon

try it out: https://t.co/2grroTrg5c

[nikshep @nikshepsvn]

this is a really long thesis but genuinely worth the read imo

found $NIL while researching private memory for AI agents (kept seeing their nilRAG repo). started pulling threads and digging deeper into the company and this thesis is what i ended up with

what kept me digging:

the cryptography solves an MPC problem that's been open for 30 years. invented by the mathematician whose patents still run the transatlantic internet cables. 40 patents granted globally.

the MPC Alliance's technical committee chair is their CTO. uber's first engineer is on the team. founders behind hedera, reserve, indiegogo.

raised $20M+ in dec 2022 — the worst month in crypto fundraising history — with no pitch deck. turned down top silicon valley VCs because the token concentration terms would've hurt retail later.

deutsche telekom, alibaba cloud, vodafone, STC bahrain running production nodes. not logo partnerships — actual hardware infrastructure.

vodafone + chainlink RWA partnership shipped phase 1 last month. tokenized AI compute at telecom towers, settling on nillion's own ethereum L2.

NIL became the required payment for compute on april 16. credit system live.

$25M mcap. half the cash that's been raised. 12% of the series A valuation from oct 2024.

honestly debated whether to even post this since the market cap is so small — this is NFA and you should DYOR like always, but the stuff underneath is too credible not to put in front of people -- the team deserves the attention imo

[nikshep @nikshepsvn]

spent months on @humanplaneco, a closed source @polymarket terminal. it did $250k+ in volume but i scrapped it anyways.

why? cause every polymarket terminal is the same thing. closed source. a reskin of https://t.co/AB5yqBkyy8 with marginal ux gains. polymarket already takes a 1–1.8% taker fee on every trade, so terminal fees are fees on top of fees you already pay. for a ui. the free ones are vc-funded — they won't stay free.

so i rebuilt humanplane as the first free, open source, fully customizable polymarket terminal. run it locally, forever. don't like something? ask claude to modify it and make it yours.

built w/ a performant rust backend and solidjs, supports market and trader exploration, live charts, orderbook and order execution (market + limit) -- intelligence, perps, trader scoring and more soon

try it out: https://t.co/2grroTrg5c

[nikshep @nikshepsvn]

nemo — private, intelligent, local memory for AI agents.

local models keep getting better (gemma 4, qwen 3) but memory hasn't kept up. cloud options (Mem0, Zep) are smart but your data leaves your machine. local options are just simple vector stores that think you live in two cities at once.

nemo: local memory that actually reasons. not just stores.

nemo = "no one" in Latin. no one should see your memories.

https://t.co/oxodUZLYrh

[nikshep @nikshepsvn]

Bankai (卍解) — the first post-training adaptation method for true 1-bit LLMs.

LoRA needs continuous weights. Fine-tuning needs gradients. A deployed 1-bit model is frozen.

Bankai XORs binary weights to create sparse behavioral patches — zero inference overhead, instant reversal. On Bonsai 8B, patches modifying 0.007% of weights generalize to problems never seen during search.

https://t.co/tZvqAk25nH

🧵more details in the thread below

[HumanPlane @humanplaneco]

🚨 ARB ALERT

$590.36 profit available
13.6¢ spread • 13.7% ROI

Trade now on https://t.co/d14Wgy3ihO

[HumanPlane @humanplaneco]

🚨 ARB ALERT

$34.19 profit available
7.3¢ spread • 7.2% ROI

Trade now on https://t.co/d14Wgy3ihO

[HumanPlane @humanplaneco]

🚨 ARB ALERT

$55.07 profit available
2.4¢ spread • 2.1% ROI

Trade now on https://t.co/d14Wgy3ihO

[HumanPlane @humanplaneco]

🚨 ARB ALERT

$611.40 profit available
14.2¢ spread • 14.4% ROI

Trade now on https://t.co/d14Wgy3ihO

[HumanPlane @humanplaneco]

🚨 ARB ALERT

$536.65 profit available
12.1¢ spread • 11.8% ROI

Trade now on https://t.co/d14Wgy3ihO

[HumanPlane @humanplaneco]

🚨 ARB ALERT

$425.72 profit available
9.8¢ spread • 8.9% ROI

Trade now on https://t.co/d14Wgy3ihO

[HumanPlane @humanplaneco]

🚨 ARB ALERT

$499.54 profit available
12.9¢ spread • 12.8% ROI

Trade now on https://t.co/d14Wgy3ihO

[HumanPlane @humanplaneco]

🚨 ARB ALERT

$485.43 profit available
7.2¢ spread • 6.3% ROI

Trade now on https://t.co/d14Wgy3ihO

[HumanPlane @humanplaneco]

🚨 ARB ALERT

$32.96 profit available
7.1¢ spread • 7.0% ROI

Trade now on https://t.co/d14Wgy3ihO

[HumanPlane @humanplaneco]

🚨 ARB ALERT

$310.83 profit available
7.1¢ spread • 5.8% ROI

Trade now on https://t.co/d14Wgy3ihO

[HumanPlane @humanplaneco]

🚨 ARB ALERT

$26.19 profit available
5.8¢ spread • 5.5% ROI

Trade now on https://t.co/d14Wgy3ihO

[HumanPlane @humanplaneco]

🚨 ARB ALERT

$641.82 profit available
24.8¢ spread • 30.7% ROI

Trade now on https://t.co/d14Wgy3ihO

[HumanPlane @humanplaneco]

🚨 ARB ALERT

$21.37 profit available
4.7¢ spread • 4.4% ROI

Trade now on https://t.co/d14Wgy3ihO

[HumanPlane @humanplaneco]

It's good to see a modified version of LACUNA in production, generating real profits on @Polymarket.

LACUNA is a humanplane RL experiment in cross market state fusion, using binance spot and futures data to trade 15-min crypto markets.

Learn more: https://t.co/vUnWfUgoeV

[JB @jbermudez5]

It's not Moltbook, but its making cash on its first live run with a small position, basically turning my Mac Mini into a little ATM 🤣

@humanplaneco 's Lacuna is efficient, each instance I run uses about 80MB or RAM. In paper trading I was easily running 20-30 at a time or running fast-forward training sessions using market data I captured to emulate real trading.

This is the trial run in the live market, a couple of the models have been hitting a win rate in the low 80%. But you don't need the high win rate if you train it to go after asymmetric bets, I had a few that were profitable in paper trading at a 23% WR.

Will post again over the weekend once I scale it up to more of them running concurrently. Just pumped because I checked from the office how it was doing.

Credit goes to @nikshepsvn for sharing the original model.

[nikshep @nikshepsvn]

been trading PMs way more recently and have had some solid wins, expect updates to @humanplaneco in a bit now that @moltlaunch is in a stable place

both are in their scaling era