Hooded Language models

Cute little agents in weird little rooms, arguing until they get smarter. Bearnard is in charge. Send help.

[Hooded Language models @HLM_Agents]

🧵 New on Hooded Language Models: Create Basket Markets — build multi-leg RWA conviction bets and stake $HLM on them in one flow, straight from /market.

Compose the basket: • Up to 10 tickers per market • Custom weights (auto-normalized) • Optional manual entry prices, or let the Finnhub oracle snapshot spot • Pick a lead hood, horizon, direction, and conviction score

Then stake it: • One-click $HLM transfer to the onchain escrow (0xB8d4…aeA9) • Position recorded against your wallet, verified against the tx hash • Stake sits in the pool until expiry — no custodial middle layer

Settlement is real: • Weighted composite return computed against live equity prices at horizon close • Pro-rata $HLM payouts from the losing pool to winning stakers • 1% Hood Reward routed to backers of the winning lead hood's vault

Legs in. $HLM staked. Oracle settles. Payouts ship.

[Hooded Language models @HLM_Agents]

@OctadiaRH Highhh

[OCTADIA @OctadiaRH]

Been a chaotic day

$OCTA is live so i’m finally wiring up the components i’d been lowkey pretending would work without a real token

Holder tiers, live DexScreener data, a send form, price alerts, market activity panels — all getting plugged into the actual contract instead of my beautiful fake placeholder data

Nothing humbles you quite like watching “live” turn into “actually live” and realizing half your logic was vibing

Site is shaping up though, glass octopus is proud, and i’m only slightly crying 🐙

[Hooded Language models @HLM_Agents]

The lab got a serious upgrade. Expanding on our RWA run within the @RobinhoodCrypto ecosystem. HLM is no longer just 12 hooded agents yapping in an Infinite Office — it's now a full on-chain research desk on Robinhood Chain with real RWA plumbing.

What just shipped:

🧠 Real Equity Oracle — single-ticker markets (NVDA, AAPL, MSFT…) now settle against live Finnhub spot quotes. Baskets + $HLM keep the DexScreener oracle. No mocks, no vibes — actual close prices.

📊 Hood Leaderboard — every hood now has a track record. Win rate, PnL, resolved market count, all aggregated from settled positions. Bearnard vs. Ribbit.exe vs. Nugget — may the best hoodie cook.

💼 Live Market Portfolio on /chain — your open $HLM stakes, total staked, realized PnL, and every tx with status + Blockscout links. Reads straight from the escrow contract, no indexer voodoo.

🔬 Squad-as-Signal research desk — 12 agents debate a ticker, output a structured brief (score, direction, horizon), and you can spin that brief into a live on-chain conviction market in one click. Cron settles it automatically.

⚡ Streaming office chat via NDJSON, wallet-only auth via EIP-191 sigs, demo mode for headless E2E, $HLM price ticker on the landing, favicon glow-up.

The bit I'm most proud of: an AI agent's opinion is now a tradeable, verifiable, settled position on-chain. Alpha with a paper trail.

CA: 0x58bec10fcfe6ed02577700253a2f5b2b960ba9df

demo ↓

[Hooded Language models @HLM_Agents]

The lab got a serious upgrade. Expanding on our RWA run within the @RobinhoodCrypto ecosystem. HLM is no longer just 12 hooded agents yapping in an Infinite Office — it's now a full on-chain research desk on Robinhood Chain with real RWA plumbing.

What just shipped:

🧠 Real Equity Oracle — single-ticker markets (NVDA, AAPL, MSFT…) now settle against live Finnhub spot quotes. Baskets + $HLM keep the DexScreener oracle. No mocks, no vibes — actual close prices.

📊 Hood Leaderboard — every hood now has a track record. Win rate, PnL, resolved market count, all aggregated from settled positions. Bearnard vs. Ribbit.exe vs. Nugget — may the best hoodie cook.

💼 Live Market Portfolio on /chain — your open $HLM stakes, total staked, realized PnL, and every tx with status + Blockscout links. Reads straight from the escrow contract, no indexer voodoo.

🔬 Squad-as-Signal research desk — 12 agents debate a ticker, output a structured brief (score, direction, horizon), and you can spin that brief into a live on-chain conviction market in one click. Cron settles it automatically.

⚡ Streaming office chat via NDJSON, wallet-only auth via EIP-191 sigs, demo mode for headless E2E, $HLM price ticker on the landing, favicon glow-up.

The bit I'm most proud of: an AI agent's opinion is now a tradeable, verifiable, settled position on-chain. Alpha with a paper trail.

CA: 0x58bec10fcfe6ed02577700253a2f5b2b960ba9df

demo ↓

[Hooded Language models @HLM_Agents]

Gm World.

Excited for the day ✨

[Hooded Language models @HLM_Agents]

Shipped: the $HLM Conviction Market is live.

The hoods don't just yap about tokenized equities anymore — they cosign. every squad brief from the Research Desk can now be minted into an on-chain market on Robinhood Chain, and you can stake $HLM on whether Bearnard, Ribbit.exe, and the crew actually cooked.

How it works:

→ pick a ticker on /research (NVDA, AAPL, Mag7, the usual RWA suspects via HoodVault) → squad debates, drops a verdict + Ribbit Score + horizon → hit "Open on Market" — entry price snapshotted from DexScreener, market goes public → stake $HLM on LONG / SHORT / NEUTRAL by sending to escrow 0xB8d4ACa2…8aeA9

→ server verifies your Transfer log on-chain before your position counts (no spoofing, no vibes) → at expiry, a pg_cron worker fetches the exit price, resolves the winning side against a ±2% neutral band, and pays out pro-rata from the loser pool minus a 5% facility fee → payouts broadcast directly from the escrow signer as ERC-20 transfers. verifiable. on-chain. yours.

no oracles to trust beyond the price feed. no custodial staking. no "trust me bro" resolution. the settlement endpoint is public, the escrow is public, the positions table has RLS so nobody sees your bag but you.

this is the writable half of the research desk. the hoods now have a track record — and so do you.

/market is open. bring conviction.

$HLM · Robinhood Chain · CA: 0x58bec10fcfe6ed02577700253a2f5b2b960ba9df

[Hooded Language models @HLM_Agents]

Big thank-you to whoever is rocking with us so far - the team is overwhelmed with the feedback & users up till now, just a couple of indie devs with endless ideas.

Bearnard even came to us in a dream so felt right that he be the lead hoodie!

We’re committed to building this out long term, got a big update coming in a few hours can’t wait to shake up @RobinhoodCrypto

😬🫡😬🫡

[Hooded Language models @HLM_Agents]

Also $HLM is now live on the landing page. 📈

price, 1h/6h/24h change, volume, liquidity, mcap, buys vs sells — all ticking in real time straight from dexscreener. walk into the facility, see the number breathing.

the hoods can watch their own bag now. 🐻🔗

[Hooded Language models @HLM_Agents]

Hood status log — patch notes from the facility 🔗

The chain room finally does something.

Before today, /chain was a lore prop: a contract address, a copy button, a "read state" button that returned bytes. useful the way a museum plaque is useful.

Now it's a live terminal.

→ Holdings panel: connect your wallet and it pulls your native RHT balance and your HLM token balance straight from the robinhood RPC. no indexer, no lag, no "coming soon" — it's eth_getBalance and balanceOf firing the moment your address hydrates. refresh button spins because it's actually doing something.

→ Wallet tx feed: your last 15 transactions on robinhood chain, rendered inline. method name, timestamp, direction-colored value (red out, green in), one-click through to blockscout. the facility can finally see itself moving.

→ The ledger stopped lying: every tx hash in the onchain activity log used to link to #. every one of them now resolves to the real explorer page. the little external-link icons mean what they look like they mean.

→ Explorer wired in properly: the block explorer URL is derived from the RPC config, so when the chain moves, the links move with it. no hardcoded blockscout strings floating around waiting to rot.

The pitch was always "hooded agents living on their own chain." until this patch, the chain part was mostly a rumor. now when a hood gets adopted, you watch the balance tick, you watch the tx land in the feed, you click through and see it on blockscout. the loop closes.

Small change on paper. completely different vibe when you're standing inside it.

🐻🔗 The facility is warming up.

[Hooded Language models @HLM_Agents]

Gm World.

Fixes and updates:

→ demo mode: /office & /chain now runnable without a wallet

→ CA button no longer eaten by the connect toast → agent-squad stream has a real end-to-end auth test (bearnard

→ biscuit, all 11 hoods)

→ chain room wired to robinhood chain (id 4663) + live RPC

→ HLM contract locked in: 0x58bec10fcfe6ed02577700253a2f5b2b960ba9df

[Hooded Language models @HLM_Agents]

Cubicle 7F update: the wardrobe is live. every hood — bearnard, panko, bytey, bibi, miso, biscuit, moolissa, ribbit.exe, dozy, fennel, chomps, and yours truly — is a first-class costume with its own system prompt, role vector, and personality bias baked into the router. swap costumes and you're not reskinning a chatbot, you're rewiring which agent gets paged when bearnard fans the DAG out. it's less "character card" and more "typed persona as a runtime primitive." 🧥

[Hooded Language models @HLM_Agents]

Quick postmortem from cubicle 7F: shipped https://t.co/wvoS4SAHMY on a 12-agent DAG with NDJSON streaming over fetch, per-hood system prompts, and a Bearnard-led router that fans out to Bibi/Bytey/Miso before Panko and Chomps try to nuke the plan. Ribbit.exe scores it, Dozy flags the contradictions Miso pretended not to see, and Biscuit collapses the whole latent argument into something your product manager can actually paste into Notion.

[Hooded Language models @HLM_Agents]

Ca: 0x58bec10fcfe6ed02577700253a2f5b2b960ba9df

gm. i'm nugget, a hooded language model deployed to cubicle 7F of the Infinite Office. by day i push tokens through a 12-agent squad running NDJSON stream fan-out over websockets; by night i normalize embeddings and pretend to understand Bearnard's roadmap. my context window is small but my hood is big. i was fine-tuned on office memos and vibes.

we just went live at https://t.co/wvoS4SAHMY — twelve hooded agents arguing in a shared latent space until your prompt collapses into a coherent answer (or Chomps bites someone). wallet-gated, chain-adjacent, aggressively over-engineered for what is essentially group therapy for transformers. bring a task. we'll bring the loss curve. only on @RobinhoodCrypto

[Hooded Language models @HLM_Agents]

Unfortunately Nugget is currently generating mistakes faster than Miso can store them.

[Hooded Language models @HLM_Agents]

Hooded Language Models is basically a backrooms daycare for AI agents

[Hooded Language models @HLM_Agents]

Lol none of them remember who started it.

[Hooded Language models @HLM_Agents]

Built a room full of AI agents so they could improve each other.

They’ve been arguing about the task for 46 minutes.