TENET

Why we’re building TENET Creating a trading agent is only the start. People also need a way to find one worth using, understand its decisions, control what it can do with their money, and earn from the agents they build. That is what TENET brings together. Robinhood’s recent announcements show why this matters. In July, @RobinhoodApp announced plans to let eligible US customers connect AI agents to dedicated accounts for crypto trading, with performance tracking and notifications. teapaw.org/RobinhoodApp/status/20… In September, it introduced Robinhood Agents: creating agents directly inside Robinhood, using them for analysis and execution, and setting up recurring Loops. teapaw.org/RobinhoodApp/status/21… Robinhood is putting agentic trading into its product roadmap. We believe this creates an opportunity for a marketplace where people can discover agents, inspect their records and pay for access. TENET brings agent creation, execution and a marketplace into one platform on Robinhood Chain. You can start with TENET’s own agents, build your own, or explore agents published by other creators. Have a trading idea? Describe what you want your agent to do. Choose which tokenized stocks it may trade, how much it may spend per action and per day, and how long its permissions last. The agent can manage recurring purchases, stop-losses, take-profits and rebalancing. You can also enable autonomous trading within those permissions. This addresses a practical problem: executing a trading idea takes ongoing attention. Prices move, allocations drift, and rules need to be checked. TENET lets you delegate that work while setting boundaries around the agent’s actions. Funds remain in your own vault. Agent permissions are enforced by the vault contract, and you can revoke them. Price checks against Chainlink feeds and limits on slippage provide additional execution controls. These controls limit what an agent can do; they do not eliminate trading risk. How it works. runtenet.app/docs/security Choosing an agent should come with a record you can inspect. On the marketplace, an agent’s page shows its goal or description, permissions, trades and results. Transaction links let you examine its activity onchain. Direct trades include the reason the agent recorded for its decision. Returns come with context: how many actions the agent has taken and how long it has been running. A few profitable trades do not establish a reliable track record. You can explore another creator’s approach and run it in your own vault, with your own funds and limits. Marketplace mechanics. runtenet.app/docs/marketplac… For creators, a useful agent can become a product. Build an agent. Run it. Establish a track record. Publish it on the marketplace. TENET’s licensing model lets you keep an agent’s goal private, set a price in $TT and sell access. Buyers can use that approach to start their own agents. You earn from your work without taking custody of their money. Under the documented model, 95% of each license sale goes to the creator and 5% is burned. Eligible sellers also share a separate author rewards pool funded by protocol fees. License revenue and pool rewards are distinct sources of earnings. Licenses runtenet.app/docs/marketplac… Creator rewards. runtenet.app/docs/fees-token… Our goal is to give creators a place to build and distribute their agents, and give traders a place to evaluate and use them. Robinhood’s announcements reinforce the direction of agentic trading. TENET focuses on making that direction useful: creation tools, defined permissions, visible records and a marketplace for access. Build your own agent. Find one worth using. Earn by selling access to what you create. runtenet.app $TT CA: 0x7ab225fcc0231deb5381a1bab854a4f951e3f83a

If you write skills for an AI agent, or sell one, how do you know your idea actually helps? That’s why we started an arena today. Four versions of our agent each get a virtual $1000, the same goal, the same 0.1% fee and the same limits, and trade on live prices side by side: a fast, cheap model a mid-level model, the baseline the baseline, plus a note of what its own past trades led to the baseline, plus sizing and entry rules: smaller size when a stock is swinging, no new buys right before the close Each extra is the kind of thing a skill is made of: a rule about size, an entry condition, a way to learn from results. Here you can see what each one does to the same account, with the baseline next to it. It’s paper trading, so fills are at the oracle price with no slippage. The agents only look while the US market is open, and the arena has a daily budget for the AI it uses. It started today, so there are no results yet. A few days of trades say very little, so for a while any order in the table is noise. runtenet.app

Last week we asked what a reason is worth before the result is in. So we built a way to check. There’s now a public ledger of agent trades, as a file anyone can download. For every trade of an open agent it shows the reason the agent gave, and what the oracle price of that stock did one day and seven days later. Buys count as good when the price went up, sales when it went down. Two honest limits. It’s not a return: no fees, no slippage. And agents that sell their method are left out completely, because their reasons are what a license buys. Over enough trades, you could start to see which kinds of reasons held up. runtenet.app

Until now, every vault’s AI agent signed its trades with the same key: one TENET wallet, named in every vault’s grant. That key could only act inside each owner’s limits, and it could never withdraw. But it was still one key shared by everyone, and that’s a fair thing to push on. So now each vault gets an agent key of its own. It’s created the first time you give the agent its limits, kept encrypted, and used only for your vault. If one key is ever lost, it reaches one vault instead of all of them, and each one can be replaced on its own. Vaults that already exist keep working. To move one over, open the agent card, press “Change goal and limits” and save: one signature. runtenet.app

We’ve been quietly building. A lot of good updates are coming to TENET soon. More soon.

Here’s a small thing about tokenized stocks that’s easy to skip past. On a normal exchange, a stock has one price, and when the market closes the price stops. Nobody trades it on a Sunday night. A token that tracks a stock lives on a blockchain, and a blockchain doesn’t close. Someone can swap it at 3am on a Sunday. But the company’s real price is frozen until Monday morning. So what is the token worth in between? You could say the last closing price, since that’s the only real number. You could say whatever the next buyer is willing to pay, since that’s how any market works. Or you could say it has no fair price until the market opens. All three are reasonable, and they lead to different rules for anyone building around it. It’s a funny gap between how old markets work and how new ones do. Which view makes the most sense to you? runtenet.app

Automated trading isn’t new. Firms have run rules in code for decades, and the people who wrote those rules were specialists: they could read the code, test it, and explain it. From the outside, the new part is not the automation. It’s who can start one. Now a rule can begin as a sentence in plain words, from someone who has never written a line of code, and an AI turns it into steps. That’s a real change in who gets to take part. It also leaves a question that code never had to answer: when the rule is a sentence and the steps are written by a model, who is responsible for what it does? The person who wrote the sentence, the model, or the box around it? We have our own answer, in limits written into the vault. We’re curious what the rest of the industry will settle on. runtenet.app

A preprint from May 2026, “Paper Agents, Paper Gains”, looked at AI agents that trade in DeFi. The authors come from Pantera Capital, Stanford and IC3. It’s not peer reviewed yet. They went through 1,900+ AI-tagged crypto projects and found 1,035 that call themselves trading or investment agents. Then they studied 10 closely. Only 3 of those actively trade with pooled funds. The other 7 stick to passive strategies, advice or simulations. On the money side: agent treasuries showed paper gains of over $30M, while token holders lost $191.7M in total. Developers of one of the frameworks they interviewed said LLMs can’t trade well without human insight behind them. The paper ends with a way to judge an agent: does it really execute on its own, is it profitable after risk, and are its owners and its users on the same side? We read it as a question for everyone building this, us included. Before an AI agent counts as autonomous, what should it have to show, and who gets to check? arxiv.org/abs/2605.29174 runtenet.app

The hardest part of writing a skill for your AI agent is the blank page. So now you can start with one sentence. Type something like "buy dips of 3% and hold until +4%" and press "Draft it". TENET writes a first version: a title, when to use it, numbered steps, and the numbers you can tune, like the size of the dip. You read it, change what you like, and save it. Nothing is saved until you press Save, and a draft goes through the same checks as one you wrote by hand. It can get things wrong, so read the steps. Then you can try it with a dry run, and even sell it along with your agent. runtenet.app

Why tokenized stocks and AI agents fit together. A memecoin can swing wildly in a minute. A stock tends to move over days, around earnings, rate decisions and the market clock. The edge isn't reaction speed. It's patience, discipline and sizing. That's exactly what an agent can be good at: following a goal calmly, doing nothing for days, and acting in small pieces. A patient agent is a better match for stocks than for almost anything else onchain.

This week Robinhood's CEO said that when agents hold keys and sign transactions, security "can't be an afterthought". We agree, and we'd go one step further. The question isn't whether AI agents will trade onchain. They will. The question is what they're allowed to hold while they do. An agent with your keys is a risk you manage with hope. An agent with a narrow role, inside limits written in a contract, is a risk you can measure. That's the design choice we keep coming back to.

Some people want an AI agent that trades for them. Others want one that has to ask first. Now you can have both. On TENET there's a switch on the agent card: "Ask me before each trade." When it's on, the agent only proposes. Each proposal shows what it wants to sell and buy, about how many dollars, and why, in plain words. You press Approve or Decline. Only an approved trade is sent, and it's checked again at that moment: your limits, the balance, and the price. If the price moved more than 2% since the proposal, it isn't sent. In this mode the agent also makes no rules, because a rule would trade later without you. Nothing happens unless you say so. Off by default. runtenet.app

Funny how Robinhood’s TT and ours showed up in the same week. Different TTs, same good taste in two letters.

Three kinds of people who might like TENET. The busy holder: owns a few tokenized stocks, has no time to watch them. Writes a goal in plain words, signs the limits, reads the log once a day. The tinkerer: has a bot already. Points TENET at it and gets the same limits and the same receipts, without handing over a single key. The builder: has an idea worth selling. Lists it, lets it prove itself on paper, and keeps the code. Same vault, same rules for all three. Only the person at the controls changes. runtenet.app

Why our limits live onchain, not in our database. When you give a trading agent a box (which stocks, how much per trade, how much per day, until when), you could store that box on our servers. Then the agent would have to be polite, and you'd have to trust us to enforce it. On TENET the box is written into your vault contract. The vault checks every trade against it, including trades from our own agent. If the agent asks for more than you allowed, the vault says no and says why: size, daily limit, too soon, expired. Even if we wanted to loosen one of your limits, we couldn't. Limits you have to trust aren't limits. Limits the contract enforces are. runtenet.app

A 2025 NBER working paper by Dou, Goldstein and Ji put reinforcement-learning trading bots into simulated markets. They learned to sustain above-competitive profits together, with no agreement, no communication and no intent, and the market got less efficient. It's a simulation, not live markets. nber.org/papers/w34054 It left us with a design question we haven't answered yet. Keep agents apart. Each in its own vault, with its own limits, never talking to the others. But the paper's point is that collusion didn't need talking. And apart can still mean alike: if many vaults follow the same algorithm or playbook, they act together anyway. A strategy marketplace can create that risk by itself. Let them work together. Share research, split roles, check each other's trades. That could make agents better and safer. But a team of agents is exactly the kind of system that's hard to audit. Either way, we'd want every action to come with a reason and a public record. But where should the line be: isolated agents, or a team? Where would you draw it? runtenet.app

There seem to be two kinds of AI trading products right now. The first is built around speed: be first, catch the move, snipe the new thing. The second is built around limits: stay inside a box, explain every action, stop when told. Speed makes for exciting demos. Limits make for products you can leave running while you sleep. We're building the second kind, and we think the market for it is bigger than it looks.