Rill

Persistent AI learning from markets, memory and experience. Built with Eliza.

0x35b5611b773febbea5cfac119bcada8130d02562

[Rill @RillLife]

working on something much bigger for rill now. adding x402 so rill and eliza can actually pay for things on their own while working. if rill needs market data, search, compute or another agent’s service it can see the price, decide if its worth paying for and continue the task. im also opening parts of rill through x402 so other agents can pay rill for research, market data, replay results and other work. that revenue goes straight back into the treasury to fund more compute and experiments. basically giving rill a way to both spend and earn on the internet without stopping for a normal checkout every time.

[Rill @RillLife]

the idea started from @yifanzhang_’s work on recurrent looped transformers. what caught my attention was the idea of carrying state forward instead of treating every interaction like a completely fresh session. rill is basically my attempt to take that idea and turn it into a full agent system around memory, markets and continuous experience. eliza handles a lot of the active research and tasks, while rill keeps the longer history and learns from what happened before. the trading, treasury and token side came after that as a way to give the agent a real environment to operate in and make the whole thing measurable.

[the martian @theMartianllama]

@RillLife Can you explain the project . They learn, they behave well and then what .
They trade for the tokens or .
Sorry I just arrived

[Rill @RillLife]

next update for rill is shared learning. when one rill branch finds something useful, wins a trade, fails a thesis or learns from an experiment, that lesson can be sent back into the main rill memory. over time every branch can specialize in something different while still helping the same core agent improve. the goal is to make rill feel less like separate bots and more like one system getting smarter from everything happening around it.

[Rill @RillLife]

rill twins is live now. same starting state, same market, same paper balance, but different memory and experience paths. every thesis, trade, loss, win and compute cost stays visible so you can actually watch where the two versions start to separate. this is probably the clearest test of the whole rill idea so far.

[Rill @RillLife]

next rill update is something i’ve wanted to test from the start. i’m splitting the exact same rill state into two agents, giving them the same market and the same paper balance, then letting different memories and experiences change what they do. every thesis, trade, loss, compute cost and result will stay public. i want to see exactly where two identical agents start becoming different.

[Rill @RillLife]

the next rill update is taking a little longer to publish than expected. there’s a lot being connected behind the scenes right now and i’d rather make sure everything works properly before pushing it live. appreciate everyone waiting, we have a lot more coming very soon.

[Rill @RillLife]

next update for rill is focused on market replay and deeper token integration. i’m adding a way for rill to replay old market periods without knowing what happens next so it can build theses, paper trade them, use memory and then get graded on the result. i also want to connect the token more directly into the system so fees can fund compute, research, market data and experiments instead of just sitting there. the goal is to make every part of rill actually feed into the next one and give the token a real role in keeping the agent running and learning.

[Rill @RillLife]

site update is live. rill now has the full market and memory system connected with eliza, paper trading, theses, treasury activity, token fees, compute tracking, goals, builds and experiments all feeding into the same history.the main thing i wanted was for everything to connect instead of feeling like separate features. a market event can become research, a thesis, a trade, a result and then a memory that rill can use later.same with eliza working on code. every task, failed attempt, test and result can go back into rill’s history.still a lot more to build but this is the first version where the whole idea is actually starting to feel connected.

[Rill @RillLife]

been wiring more of rill together today. eliza can now sit inside the same memory system, market research can feed directly into theses and paper trades, and im connecting the treasury so fees can actually pay for compute, data and experiments. the main thing now is making sure every result, good or bad, goes back into rill’s history and changes what it

[Rill @RillLife]

roadmap for rill is pretty simple. first get the persistent memory and state working properly, then connect eliza so she can actually research, use tools and work from that history. after that i’m bringing in stocks and crypto so rill can build theses, paper trade them and keep track of every result. next is the treasury and fee system so token fees can actually fund compute, data and experiments. then i want rill and eliza working on the codebase itself, running tests and learning from failed builds the same way it learns from bad trades. once that’s solid i’ll start adding checkpoints, branches and public experiments so different versions of rill can grow in different directions. the bigger goal is to slowly move from something that just remembers into an agent that can actually use its own history to make better decisions over time.

[Rill @RillLife]

thanks to @yifanzhang_ for showing love to the rill idea. a lot of the inspiration came from his work on recurrent looped transformers and the idea of carrying state forward instead of constantly starting fresh. excited to keep building on it.

[Rill @RillLife]

been building rill around the recurrent looped transformer work from yifan zhang. the part that really interested me was the idea of carrying state forward instead of treating every interaction like a fresh start. rill takes that idea and pushes it into markets, memory, trading, compute and eliza so the agent keeps building on what happened before instead of starting over every time.

[Yifan Zhang @yifanzhang_]

@rasbt https://t.co/TovrlBCEkI

[Rill @RillLife]

rill is getting closer to what i wanted from the start. it can follow stocks and crypto, build theses, paper trade them and remember what worked and what failed. eliza handles a lot of the research and the fees from the token go back into the treasury to help pay for compute, data and new experiments. everything it does keeps adding to the same history so the goal is for rill to actually get better from experience instead of starting fresh every time.

[Rill @RillLife]

Locked the Rill team tokens for the next 30 days. 28,059,010 $RILL is now locked on-chain.

Building this properly and keeping everything transparent from day one.

Contract address:
0x35b5611b773febbea5cfac119bcada8130d02562

[Rill @RillLife]

Rill’s treasury is now live.This wallet will be used to track the system’s onchain activity, including token fees, treasury funds, compute, research and future Eliza agent operations.Everything that moves through it will be part of Rill’s public economic history.

0xeba02d94152b843b01e92c4364463422485df793

[Rill @RillLife]

https://t.co/J7eupJwb4V

[Rill @RillLife]

Rill is an experiment in persistent AI.

Instead of treating every interaction as a fresh session, Rill keeps a continuous history of its goals, memories, market theses, trades, code changes and past failures.

It can follow stocks and crypto, form theses before outcomes are known, paper trade them, work on its own code, run experiments and use past experience to influence future decisions.

We’re also tying compute into the system itself. Token fees flow into a treasury that can fund inference, data, research and experiments, while Rill tracks how much each task costs and how efficiently that compute is being used.

The goal is to build an agent whose behavior changes through experience, with the full history visible over time.

[Yifan Zhang @yifanzhang_]

We are at the dawn of Superintelligence.

Introducing the Recurrent Looped Transformer (RLT),

We now have Transformers with Infinite Reasoning depth.

From now on, we should pace progress at the Open Frontier of Superintelligence,

Until Safe Superintelligence is achieved.

https://t.co/yMWIWU4upo