
market news
the latest headlines, piped straight to your agent.

snap blocks together into agents that watch the market, check your risk and wait for your yes. they run on your own machine. nothing moves past the limits you set.
name it, pick a model, tell it the job in plain words. or start from a template and skip the blank page.
claude, gemini, deepseek, kimi + moremarket news, telegram alerts, hyperliquid prices and wallets. fill in the settings and it picks them up on the next run. no restart.
install once, live next runtell it when in chat: every 15 min to once a day, daily or weekly at your time, or just once. asleep between runs. watch every run live, chat with it, pause it whenever.
asleep till it's neededa scout spots the setup. an analyst writes the thesis. risk checks it against your limits. a sizer cuts it down to size. then it waits for you.






see what each agent told the next one, and why, in one live feed.
the trading block checks your limits before every order. paper mode for trying ideas first.
nothing executes until you approve it. unless you let it run inside caps you set.
every block is an integration in its own sandbox. before you install one you see exactly what it can reach, and whether it can touch money.

the latest headlines, piped straight to your agent.

prices, candles and any wallet's positions, orders and fills. read only: it watches, it never trades.

your agent taps you on telegram when something needs you.

runs a strategy on its own server and drops the report card in your feed.

logs every trade with the reasoning behind it. searchable when you want to look back.

builders publish blocks to the marketplace and get paid when people use them.
a typical editor extension can reach your files, your keys and the whole internet. a block gets one install, the domains it declared, and nothing else.
every team gets its own machines. no inbound ports, no ssh, no platform keys on them. they only ever call out.
encrypted the moment you enter them, decrypted only for the one call that needs them. never on your machines, in a prompt or in a log.
each block runs in its own sandbox, reaches only what it declared and sees one install at a time. a bad block can't touch the others.
money blocks check your limits in code before every order. a block's screen can't place an order by itself. you confirm it in ours.
calls to models and blocks pass through our gateways to be metered. nothing else from them is kept.
columns for conversations in our database
tables on your machines hold every run and every message
every agent gets its own machine, any model, schedules, chat and a live feed of every run. agents sleep when there's nothing to do, and asleep is free.
| plan | free | solo | desk most chosen | pro |
|---|---|---|---|---|
| price | $0 | $9 / month | $29 / month | $79 / month |
| best for | trying it out | your first agents | a full desk, every day | heavy, all-day use |
| awake hoursa month. the time your agents' machines are running | 15 | 100 | 340 | 960 |
| agentson your desk at once | 2 | 5 | 12 | 25 |
| per hourwhat an awake hour works out to | – | 9.0¢ | 8.5¢ | 8.2¢ |
| modelspaid on your own model account. we add nothing on top | any model, free ones included | any model, free ones included | any model, free ones included | any model, free ones included |
trying it out
your first agents
a full desk, every day
heavy, all-day use
hypersolid is a platform for building ai agents for hyperliquid traders. you snap blocks onto an agent (market news, hyperliquid data, telegram alerts), pick any model, and it runs on its own machine, on the schedule you give it in chat. trading is coming soon.
no. you sign in with email. the hyperliquid block watches any wallet by its address, and trading, coming soon, is a block you switch on when you want it.
not yet: at launch the hyperliquid block is read only. trading is coming soon, and every order will be checked against your limits in code, never left to the model.
claude, gemini, deepseek, kimi and more. you pick one per agent. soon agents hand work to each other, so a cheap one scans all day while a stronger one makes the call.
they're encrypted the moment we get them and decrypted only inside our gateway when a call needs one. they never reach your machines, a prompt or a log.
an hour one of your agents is awake. each agent has its own machine, so two agents awake for an hour use two. a scheduled run sleeps the moment it ends; after a chat it stays up 5 more minutes so replies are instant.
join the waitlist and we'll email you when your spot opens.