The Bot Pod Episode 7: Deep Dive into Routines

We’re back with Episode 7 of The Bot Pod! This week, Mike and Fede take a deep dive into routines—what they’re calling the most powerful and immediately useful primitive in Condor. The episode covers a wide-ranging market discussion (perps on Cerebras, prediction markets, the prop-firm “funding test” model), a live demo of Mike’s memecoin LP yield routine and agent, and a tour of the latest Condor UI improvements.

Watch the full episode: YouTube

Install Condor: condor.hummingbot.org

Mike’s friends in traditional finance are increasingly excited about Hyperliquid. The standout story: a Citrini tweet about funds watching CBRS (Cerebras, the AI inference chip company) trade on a Hyperliquid HIP-3 market for price discovery before the company has even IPO’d.

“Crypto markets are becoming almost true oracles for what might happen in traditional finance. When this thing actually IPOs, guess what those initial market makers are going to use for the reference price?” — Mike

The hosts talk through the counterparty risk in these new market types. Fede points out that every big crypto crash he’s seen came from a broken trust between counterparties—and a thin perp market on a niche index could leave traders unable to exit positions.

“Having lived through FTX, there’s always something unknown about any type of leveraged perp DEX.” — Mike

Mike draws on his finance background to separate market types: L1s like Solana and Base are where new tokens are born and experiments are run (spot), while a perp DEX is a pure derivatives market for anything you can leverage. They also touch on proof.trade’s conditional-value markets and MetaDAO’s decision markets.

Fede and Mike dissect how FX brokerages and prop firms really work—identifying winners and losers, and the “funding test” model where traders pay for an account, trade on paper, and most of them fail, generating free money for the issuer.

Mike frames the core idea of the episode: turning data into something usable used to take a long time—collect, clean, parse, visualize. Routines collapse that. Combined with Hummingbot’s connectivity to every exchange, “anything you can dream of can be built on top of routines.”

Answering an audience question, the hosts are clear: Hummingbot is a framework, not a magic money box. Fede shares a concrete example—a client market-making fiat pairs on Binance, ~$1.5M/day volume, earning ~9% monthly on rebates alone before trading PnL.

“If you’re expecting to click a bot and make money, you will fail. It’s a framework—it has all the primitives you need.” — Fede

Mike demos a routine he built that uses the GeckoTerminal API to fetch the top Solana pools, ranks them by yield (fees ÷ TVL), and filters to concentrated-liquidity pools so he can set single-sided SOL ranges instead of buying memecoins outright. He then pairs it with an agent—Memecoin LP Yield Hunter—that fills three LP slots, monitors them, and redeploys capital as positions auto-close.

“Although it needs some iteration, it basically did what it was supposed to do. So far I’m pretty much in the black in all the pools.” — Mike

Mike shows a second routine that visualizes his live LP positions against each pool’s liquidity distribution—and how he iterated on it, asking the agent to add more detail with each run.

“This is a revolution. Routines and reports are the new reporting era—dynamic visualization. You have an idea, you plot it, then you reproduce it over and over again.” — Fede

Fede explains how it fits together: an agent (powered by an LLM) helps you build routines that surface exactly the data you need. Once built, that data is automatically available to your trading agent, which has tools to create and stop executors—and maintains a learnings.md file so it improves over time.

Fede walks through the latest UI: multi-currency portfolio conversion via the rate oracle, the trade interface for running grids across any CEX, the executors view, and a pairs-trading agent built live in Botcamp that creates grids based on deviations.

A standout routine: built with agent mode in minutes, it pulls order books from four exchanges and compares depth and spread in BPS around the price. Fede uses it to spot a real cross-exchange opportunity on AAVE and a 90 BPS spread on a memecoin—leading Mike to plan a CEX/DEX cross-chain market-making routine on the TROLL token for next week.

A small but meaningful touch—Fede added a color-blind theme alongside dark and light modes after a friend who uses Condor asked for it.

Mike’s homework: extend the order-book arbitrage routine with a Jupiter (Solana) leg and report back on whether there’s a real CEX/DEX arb on the TROLL token. Join us live next Friday on YouTube.

Mike: Welcome, welcome everyone. This is another weekly episode of the Bot Pod podcast. I’m Mike, this is Fede, and we are the co-maintainers of the Hummingbot open source framework. Lately we’ve been working on this new agentic harness called Condor. It’s like Open Claw, but for traders.

We’ve been focusing this podcast on weekly demos of all the new features that Fede—mostly Fede, but maybe 10% me—have been adding to the harness. This week Fede’s going to demonstrate some new changes to the user interface, as well as take a deeper dive into how routines work.

Overall, we think routines are a really powerful new primitive in Condor. Combined with LLMs, they can make a lot of trading-related tasks a lot easier. But before we get into that, like we always do, let’s talk about what’s been going on in the markets lately.

Fede: Hi, everyone. How’s it going? The markets are really hot right now. You don’t know if it’s going up or going down. Bitcoin could break 80K. Hyperliquid went up 20% yesterday—well, maybe 15%, but still a lot. Now it’s going back down a little.

A lot of things happening. A lot of uncertainty, but I think this is a time to build portfolio, build strategies. As I said last time, we’re probably at the lowest point of crypto prices over the last two years for some tokens like Solana. And I still see the market evolving—more transactions being processed, more fees collected, more real-world companies getting into the chains. So I’m bullish on this.

Mike: What’s interesting—I’m actually on the East Coast this week. I’m here for my 25-year college reunion. Yes, I am that old. I stopped by New York on the way and met up with some friends in traditional finance who are crypto-adjacent. It’s really interesting because a lot of them are super excited about some of the things happening in crypto lately, especially around Hyperliquid.

We talked last week about Hyperliquid and their HIP-3 markets—how they’re basically creating perpetual markets for all kinds of things where there’s some observable index price. For example, oil is one of the third or fourth biggest tradable markets on Hyperliquid. But now you also have silver, the S&P 500.

There was a tweet this morning from a Twitter account called Citrini that a lot of people in TradFi follow. Let me read it: “Have gotten four different calls today from funds for watching CBRS trade on Trade XYZ for price discovery. Pretty surreal.”

Fede: Yeah, it’s crazy. They’re trading the price discovery of this new company.

Mike: Exactly. But this isn’t just any new company—it’s Cerebras. Cerebras is actually just down the street from me, literally about two miles from where I live. They’re very secretive, but apparently they’re building some type of AI inference chip. Now that there’s so much demand for AI, the load is shifting from training to inference. If they can build an inference-specific chip that’s faster or lower power, this could be the next big IPO.

So to me, what’s interesting is that crypto markets are becoming almost true oracles for what might happen in traditional finance. When Cerebras actually IPOs, guess what those initial market makers are going to use for the reference price of CBRS? I believe they’ll use this Hyperliquid HIP-3 market.

So I do think perps on anything is actually a viable outcome. I’m not saying it’s going to happen, but—not financial advice—we should probably start accumulating some HYPE.

Fede: Yeah, I agree. We were discussing today what the risks are. Personally, all the big crypto crashes I saw were when someone broke the trust of the other counterparty. If we’re trading on a market like this and suddenly all the players go to zero funds, then the market doesn’t exist and you can’t trade or get rid of your positions. That’s something that might be complicated.

Mike: Definitely. Having lived through FTX, there’s always something unknown about any type of leveraged perp DEX—especially since they’re also doing the market making. There’s some tie-up between HLP, the market-making vault, and the exchange. So there are some similarities. I definitely hear you on the risk. It’s not financial advice.

Fede: Not financial advice, but there’s a part of me that says you have to be careful with these because I don’t know where it can go. At some point, if someone creates a similar thing on Solana—I think the Cheetah guys will have a very good chance with their CTX exchange, because Solana has a lot of liquidity. All the tokens are tradable, even meme coins, because there’s a lot of liquidity for all of them.

If someone can combine the experience of AMMs for perps with an order book—similar to what XRP has, where you combine the order book and limit orders—I think the liquidity in Solana would probably be better and people would choose to trade there. But for now, Hyperliquid has the best experience. The point of why it got so good is that they’re basically a Binance terminal on a blockchain—anyone can feel the same experience as trading on a CEX.

Mike: I totally agree. But I think there’s space for an exchange—a CEX or DEX like Hyperliquid and Binance—which is about trading currencies, major coins, those kinds of things. And there’s also room for another kind of chain for meme coins, where things are created and born. That’s basically an L1—composable, like what HyperEVM is supposed to be.

But that’s a different animal from a perp DEX, or even a perp CEX, which is just a leveraged futures market. It’s more like a pure derivatives market for any instrument you can leverage and trade. So you still need a spot market for those things, and you need a spot market for new assets generally. They serve a different purpose from a market-design philosophy perspective.

Fede: Yeah, you still need both. Hyperliquid did really well at the perp side, not so much the spot side.

Mike: I agree. That’s why I think an L2 makes sense—something customized to fit that market. Whereas the edge Solana has is all these different competing startups creating various types of markets. There’s one recently announced by my friend Squid in the Solana ecosystem called proof.trade. He’s trying to create impact markets—basically trading the value of something conditional on some event happening. For example, what would the price of oil be if there’s a truce between Iran and the US.

As a statistics nerd, I like the idea of conditional probabilities and expected value given some probability. It’s a different spin on prediction markets—prediction markets trade whether an event will be one or zero, and this is more like trading the value of something given that event. Another example is MetaDAO, where they’re trying to make decision markets and futarchy happen to govern tokens.

So to me, Solana and Base and other L1s are the chains where things are born and experiments are run, creating new spot tokens. And then there’s another need: how do you create derivative markets. This is also why, when I worked in finance, these were different departments—equities on one floor creating new companies, and fixed income or derivatives on the floor above, which is what I worked in, creating all kinds of funky derivatives.

Overall, whether it’s spot or derivatives, I think crypto’s going to take over finance generally. Sorry to our friends in TradFi, but in 20 years it’s all going to be crypto.

Fede: I agree. Everything will move to crypto. It’s really interesting, the things that are happening.

Fede: Did you see this paper.trade XYZ? It’s another vision thing happening on Solana. Basically you’re trading against an LP—the paper token. You have 1,000x leverage to trade, and they have a mechanism of queues for getting paid. The long bet is that the value of the LP will grow as people lose against it by trading the markets.

Mike: Those are the things that usually crash, but there’s something interesting here. It reminds me of a conversation I had with a Hummingbot user—I won’t disclose who. They were telling me how the FX brokerage game works. They give you an account, and what they’re really trying to do is identify who’s a winner and who’s a loser.

The winners, obviously, they can give more capital to and you can actually profit—the incentives are aligned. But identifying a loser is also valuable, because you want to trade against a loser and take the other side. The reason they give people an account is that just making that identification is worth more than the value of the account they’re giving up. Most of the value comes from identifying the losers, not the winners.

When I was in Mendoza after your wedding, the guy who drove me around was basically one of these guys getting the accounts. He was trading FX, and I tried to convince him to try crypto trading—”Why don’t you trade FX on Binance perps?” But then he told me about this model and I realized the whole incentive structure is different. Binance isn’t going to give people $100 to trade with on Binance Futures, but these FX brokers are giving them $100 accounts.

Fede: I think it’s also a Ponzi model. The way it works is they make you pay a certain amount of money to get the account—

[A pause as Fede’s desk gets bumped.]

Sorry, I’m having a problem at my desk—my wife threw all the mugs onto the table while she was playing a game.

Mike: It’s good you’re referring to her as your wife. It took me a while after my wedding to refer to “the wife”—”my wife.”

Fede: Now it’s my wife. He almost threw it again. But some of these contracts—you pay for a funding test, and if you fail the funding test, you lose all the money. To pass the funding test, you need to make X percent of yield.

Mike: So it gives you an account, and you have to make a certain amount to keep going. Yeah, I think my driver mentioned that too.

Fede: You need to make a certain yield to keep going. Most of them lose, so that’s free money for the one giving the funding. And the thing is, they’re never trading on real money—they’re trading on paper-trade accounts. So the issuer is never losing.

Mike: Got it. It’s like paper trading. So this paper.trade XYZ sounds a bit similar in terms of incentive design—they’re trying to get people to lose.

This is what I love about crypto: any experiment you can think of is being run.

Mike: Let me do the segue. Speaking of experiments, there’s nothing you can’t do with routines, as we’re finding out. As a trader, it’s important to be able to quickly take data and turn it into something usable. In the past that would take a long time—you’d have to collect the data, clean it, parse it, and finally put it into some visual format you can actually analyze and share.

But with the power of Hummingbot, how it connects every exchange, plus this concept in Condor called routines, that’s gotten a lot easier. We’ve already built a few—Fede has demonstrated them in prior livestreams—but today he’s going to give a real deep dive into exactly how they work and show off the new Condor UI that’s optimized for routines.

Fede: I agree with your description. Anything you can dream of can be built on top of routines. Do you want to share the one you built first—the one for the token—and then I’ll show the new things about routines in general?

Mike: All right, let me make sure I’m running my Hummingbot API and Condor. Let me open up Docker. By the way, I’m trying to make our content a little more clip-optimized. We have a great community manager named Carlio who’s been helping with video production, turning the content we create weekly into something shareable. I’m trying to make his job easier.

Mike: Before I dive into the meat of routines, let’s answer a few questions from the audience. There’s one from Digita: “Is there a real use case where using Hummingbot makes money with a growth strategy?”

We get this question a lot. We probably wouldn’t have $30 billion of reported volume per year if people weren’t using it for something of value. But I’d look at Hummingbot more as a framework for building some type of automated process. A lot of market makers use Hummingbot to run strategies like cross-exchange market making or pure market making, because that’s how they provide that service to their customers. Similar to us—when we do market making, we use Hummingbot because it lets us automate something and build off a framework instead of hiring a bunch of developers to build from scratch.

Our job is to make sure it’s usable in many different ways, and that’s why we added Condor. Now anyone can use Hummingbot—you don’t have to be a developer anymore. You can just talk to Condor and it can build what you want.

Fede: That’s something interesting—if there were no people making money, we shouldn’t have 3 billion in trades per month with our framework. So there’s definitely a way to make money with it. But if you’re expecting to just click a bot and make money, I’ll tell you that you will fail, because it’s a framework. It’s not a magic mystery box that makes money for you.

The framework lets you collect all the market data from the exchange—order book, trades, candles, funding rates, everything you need—and lets you execute any trade you want. There are some pre-built strategies you need to configure, and based on the config they change a lot.

So yes, it’s a framework. If you have a strategy that’s good for the type of advantages you have, you can make money. In my case, right now we’re working with a client that has an edge on fiat markets on Binance. We’re trading around 1.5 million per day, and we get 0.015% in rebates. On a monthly basis, that’s about 9% only on rebates—then comes the trading PnL. So it depends a lot on the edge and the type of strategies you want. The point is, wherever you want to use it, you can, because it has all the primitives you need.

Mike: Let me demo the routine I built a couple of days ago. As you guys probably know, my specialty is doing LP strategies in Solana, especially for memecoins—which I do less of now. A couple of years ago I was doing enough of it that I was personally getting concerned I was developing a slight gambling addiction. So I’ve been building routines to identify promising memecoins and then LP for them on Meteora. Let me share what I’ve built so far.

This is very early, so don’t expect it to be life-changing or some super profitable algorithm. But I thought it was a promising start.

Fede: While you share that—I like a lot what Samuel said in the chat, and that’s exactly the way I’m approaching it. I’m adding a process of batching in the middle, because to be more aggressive, my market maker has an onload period and an offload period. I’m doing discrete trading sessions—some with positive outcomes, some with negative outcomes—but it works long term. There’s a balance between how aggressive you are and how you can keep making profits long term.

Mike: Okay, let me share my whole screen. I have Condor running here. We recently updated the installation section of the website—condor.hummingbot.org. David just added a new installation script, so you should be able to download the script, run it in your terminal, and it’ll install everything you need.

The first step is creating the Telegram bot you’re going to use. I’ve named mine Condor—a new bot I created using BotFather. Because I’m already running Claude Code, it’s connected, so I can say “Hi” and it should respond. There’s also the web dashboard. Let me ask Condor what it can do.

As you can see, it’s a harness similar to Open Claw, and what you can do with it is anything the Hummingbot API is capable of—getting portfolio, doing trading, running bots and agents. But the most valuable thing we’ve seen so far is building routines.

Here’s one I created the other day to analyze pools and create an agent. This routine uses the GeckoTerminal API to fetch the top Solana pools—probably the top one or two hundred pools by volume. Then it ranks them by yield, meaning the fees the pool earned divided by the total value locked. I then filter by concentrated-liquidity pools, because I don’t want to buy the memecoin outright—I only want to set single-sided ranges. If the price falls into the range, I’ll buy some, but I’d rather earn fees. So this routine just ranks the top pools over and over again. When I ran it, the top pools were RKC, Troll, and Hanta.

Mike: After I built that routine, I paired it with an agent—called Memecoin LP Yield Hunter. The strategy is defined in the agent.md file. It says: provide single-sided SOL liquidity to the top three trending Solana memecoin/SOL concentrated-liquidity pools.

So it takes the output of that routine—the ranking of top pools—and there are three slots. If the slots are empty, it creates LP executors to put single-sided positions in them. The LP executors have automatic close limits—upper and lower. When an LP executor self-closes, that frees up a slot, the routine triggers again, and the agent deploys that capital into a new slot.

I’ve run this for a little bit. This last session ran for 100 ticks. In the beginning, because it was inheriting from other sessions, all three slots were filled. Then it kept tracking the slots. Sometimes it opened slots—here it actually saved an error: at first it failed, then the next tick it detected the duplicate and did the right thing. Although it needs some iteration, it basically did what it was supposed to do—run the routine, look for open positions, and create LP executors on Meteora.

In terms of actual performance, it did okay. I’ve been tracking it on Meteora, and I had positive PnL in basically all the pools I created. A couple are still in range—they might rug and go down—but so far I’m pretty much in the black in all the pools.

Fede: Seems like there’s a problem with the reporting of the executors—the P&L. If you go down to the executors, do you see that they say zero P&L?

Mike: Yeah. Part of it is the candles aren’t being fetched, because it would need to get candles from GeckoTerminal for these. There were a couple with P&L in there. It’s probably easier to see in the executors list. Part of it is that this is SOL—for instance, this one I made 13% on from a yield perspective, but the volume is in SOL, so the P&L is probably denominated in SOL.

But you’re right, I need to check, because there’s a bunch of stuff with the GeckoTerminal data not displaying correctly. I need to make sure the SOL base displays for executors, as well as general DEX charting.

Fede: Don’t worry about that—I can do it for you. The SOL thing is already solved in my PR. On the top right you’ll have a conversion option that lets you do the conversion.

Mike: Overall—and this is for the audience too—my experience is I’ve found routines to be much more helpful and usable right now. The agents I’m excited about in the long run, but routines are immediately valuable.

For example, here’s another routine I created to analyze the existing LP positions I had created. This is the current price, this is my overall position, and these are the auto-close prices for each position. With this routine I could see whether my positions were in range and earning fees, or close to being closed. In the future I’d also like to have a notification if one closes—then it shows me a list of the top pools and lets me select which one the new slot deploys into. That kind of format I’d feel more comfortable running in the near term.

Fede: I think this is a revolution. Routines and reports are the new reporting era—dynamic visualization. You have an idea, you want to plot it, then you want to reproduce it over and over again. It’s amazing.

Mike: This is interesting, Fede—the first time I ran this routine, I looked at the report and thought, “I actually want to see more details about all these positions.” So I asked the agent to add information about each position to the report. The second time I ran it, it added these liquidity distributions for each pool, showing my position relative to the existing liquidity distribution. I thought it was cool how you can iterate on the reports and make them better over time.

Do you want to share what you’ve been working on? I think I’m even using an outdated version of Condor.

Mike: There’s a question from Raj: are we able to debug some of the LP executor issues? We have been fixing them. I pushed to Condor recently the updated LP executor guide—that’s on main. There’s also a PR open about the API that our QA team is going to review, which fixed a couple of minor issues I found using API plus Gateway.

Overall, the LP executor should be working from an API perspective. We still need to make sure the Condor UI is integrated correctly. For all executors, my lesson is: if you have it configured correctly, it should work. But you may have to do some work—check the API logs and error messages—to make sure it works.

Fede: There’s another question from Samuel—a very good one, because it’s about how the architecture works for routines and agents.

We have the concept of an agent powered by an LLM. All the routines—like the ones Mike was showing—you can build with an agent that’s also in the same product. Inside Condor, you have an agent to build the routine that will help you use Hummingbot to surface all the data you want.

The interesting thing is that once you create the data you want, and you know what you’re looking for, that’s automatically available for the agent. So the trading agent you create will be able to get that data and use it to trade.

That agent has tools for creating and stopping executors. While the agent works, at each tick it logs things in its journal and maintains a learnings.md file with the knowledge it gathered. The next time, it’s aware of the problems it had in the past and is able to face them. So it’s an evolving agent that you work with. Plus you can edit these things from the UI—the learnings, the behaviors, everything.

Mike: Right now, what we’re trying to do is create an open standard—a framework where the agent can operate and process data deterministically using routines. It uses tools as executors, so it’s only using the LLM for the actual decision part, which should be much more token-efficient than throwing a lot of APIs and documentation at a model.

So the Condor harness has this concept of a trading agent, which is hopefully a standard for doing something professional-grade where an LLM is involved in helping you do something over time. We still need to discover what types of agents are best suited for this versus traditional bots—which are more for market making and HFT, deterministic and not really suited to make probabilistic choices all the time. That’s why we’ve designed Condor in a very open way. Routines, as Fede will show, are probably the most flexible primitive we have right now.

Fede: Let me put my screen up. Some of the changes we have for Condor. This is the portfolio page—you already know about it, it didn’t change much. The important thing is you have a central place to see all your tokens, the distribution by exchanges, and the variations over time. What I added is this USDT selector that lets you change between different currencies. Probably we’ll remove some and just leave USDT, USDC, and a few others. This uses the rate oracle to translate the values.

Then the trade interface is still working really well. You can run an executor from here—create a grid, a short grid, a long grid. If you’re getting started with trading and want to run some grids, you can run them across all the CEXs. You configure the parameters and the UI works really nice.

Then we have the bots. Right now I don’t have bots running on this account. If you go to the editor, you can see the configs we have. We also have the backtest, but I don’t recommend using this tool yet, because I’m working on a routine that will do it better. I’m very bullish on routines in general.

For example, for the post-analysis of a bot, the best way will be with a routine. You deploy five controllers on one bot; once the bot finishes, the performance of those controllers is the same as before. So once you finish your run, you can process all the databases together and generate the HTML files. The routine result is an HTML file with all the data you want. Once you process one bot, you don’t need to process it again. So having links to the HTMLs is the best way to store the data of a processed bot, instead of querying all the trades and processing them again every time you want to see an old bot’s performance.

This is the executors view—now we can translate it to USDT. If you have multiple bots... the other day in Botcamp we created a pairs-trading agent that creates grids based on deviations, so you can filter one of those.

This is the agents view—the one Mike showed before. If you go to a specific agent, you can see the strategy MD, the learnings, and the routines filtered to that agent. If you run it from here, it generates this output. You have the two versions to move between.

Mike: Fede, did it take a couple seconds to run?

Fede: Yeah, it takes a few seconds.

Mike: You may want to explore adding a loading indicator—that should help. There should be components you can easily design for that.

Fede: “Can I use Condor directly for a grid bot?” Yes, you can. It will automatically clone the Hummingbot API, and in the trade pane, if you pick grid, you can create a grid directly there.

You can also schedule routines—run them every five minutes, every minute. And I improved the error checking. Here it says “Running, zero runs, routine failed.” This is only for failures—if your routine fails, this is how you see it.

Fede: You can run interesting things. For example, this routine—since Hummingbot has access to the order books—you can analyze the different depths of a specific market. As simple as that, we created a routine that analyzes the order books of four exchanges, compares the same price, and shows the different BPS around the price.

Mike: Is it fair to say there might be an arbitrage opportunity between MEXC and KuCoin here, or maybe cross-exchange market making?

Fede: Cross-exchange might be a good opportunity—I’d place sells here and buys here. This is three, almost four BPS of distance, but there’s no arbitrage. To have arbitrage, these would have to be crossed, and that value would need to be bigger than the fees you’re paying.

But this is the power of routines. I used the agent mode to say I want these order books—use a gather to get them and compare the books—and it was able to do it. Let’s see AAVE. Here there’s an actual opportunity: on Gate we have five BPS to sell, and on Binance it’s 0.5 BPS. So there’d probably be an opportunity to make.

Fede: A good long-term strategy, instead of just doing cross-exchange and making trades only when they arrive and then hedging, would be to sell here and place a limit here—quoting buy here and quoting sell here. That would be a better way to start moving your inventory to one side.

Mike: Right.

Fede: Here we’re talking about mature tokens, but let’s look at a meme coin. This one is only traded on Gate and KuCoin. You can spot opportunities here too.

Mike: I should mention this Protocol coin—I think I was LP-ing for this on Solana. I think it’s a Solana-based meme coin, so there’s probably some opportunity between LP-ing or swapping it on Solana and trading it on a CEX.

Fede: Here you can get 90 BPS of spread. The power is huge—the things you can do are amazing.

Mike: I think I’m going to take this arb-check routine and add a Jupiter leg, and really look at this Troll token for next time. I’m pretty sure the spreads in Jupiter are going to be tighter than this 90 basis points. Since it’s a Solana-made token, the spread on Solana is tighter, so there should be an opportunity to do cross-chain market making—make on these CEXs and take on Solana.

Fede: I’ve now set this to run every one minute, so it’s scheduled.

Mike: I’ll build it off the back of this arb-check routine—I should be able to just tell the agent to modify this one and add the Solana bit.

Fede: There are a lot of back-end improvements, and the Agent tab was improved. I still like to use Telegram a lot, but here you can use the different agents—the agent builder, the Condor agent which is your main agent, and the routine builder, which lets you build routines more effectively.

There are a lot of things to play around with—seeing how the books are moving from here. It’s amazing that we can see this on our live reports now.

You can also see the routines that are running or scheduled in the main UI here. I probably need to simplify this nav bar, because when there’s a lot of data it gets collapsed. I’ll probably remove the date from here.

Mike: You can probably also shorten it—show the name and remove the description, because I think the name is sufficient, as long as you can access the name somewhere. It’s already in the sidebar.

Fede: I need to remove these two things.

Fede: Another thing I added—because I have a friend who uses this a lot and he’s color blind—I added a theme in the portfolio. This theme works for color-blind people.

Mike: Do you do a theme selector, or how does that work?

Fede: It’s just like dark mode, light mode, color-blind mode. You just change this.

Mike: I kind of like that. Basically three modes—that makes sense.

Fede: If not, I was going to do color-blind for light and color-blind for dark, but then I said there are a lot of options and it’s probably not worth it.

Mike: We could build themes, but we probably save that for later. I like this—it makes sense for people who need some visual help.

Fede: I think this was a very good session to demonstrate this. We can catch up next week with more advances.

Mike: I’ll do some homework for next session. After you merge in the PR, I’ll take the arb-check routine and try to incorporate some CEX/DEX arbitrage checking. Next week I’ll come back and report on whether there is or isn’t arbitrage in the Troll token.

I also want to show that routines aren’t just for you to run—you can modify and extend them, and with the agent it’s a lot easier.

Fede: Thank you very much, everyone, for joining. Next week we’ll see more improvements.

Mike: All right. Bye, everyone.

Fede: Bye-bye.