Jeonlees

Seriously stroke your hair

977Following

1.3Kfollowers

Feed

In the graduating team, there's Termmax. Somehow, I really feel that after this event ends, the TGE is truly coming. Luckily, the puzzle event was completed in time. Can't keep up with MP's pace, so I'm making up for XP. Really, you can't give up halfway when doing events; you have to persist 💪 Yesterday in the group, I even learned from Mushroom Bro that you can actually put the US stocks you buy into Termmax to earn points—two birds with one stone. Recently, US stocks have surged sharply, but unfortunately, I missed this wave, sob sob. Many people, once they see the market start moving, think about leveraging, recycling, and improving capital efficiency. This reaction is normal because when the market is stagnant, no one even wants to open their wallet. But the problem is, when volatility rises, the real danger is often not that you guessed the direction wrong, but that your funding costs suddenly get out of control. This is also why I think TermMax deserves a fresh look now. The core of TermMax is not simply "borrowing" or "lending," but fixed-rate lending. Simply put, users can lock in borrowing costs or yield boundaries in advance, without being dragged around daily by floating rates. TermMax's official site also emphasizes fixed borrowing cost, lock your rate, one-click leverage—meaning you fix your costs before entering a strategy. This feature is very practical in today's market. When the market is stable, floating rates don't seem like a big problem. Whether borrowing costs are a bit higher or lower, it seems acceptable. But once the market starts to fluctuate and capital demand rises, floating rates can suddenly spike. Your well-planned strategy might have its profit margin eaten up due to changes in borrowing costs. Many DeFi users have actually suffered this loss. At first, they focus on yields, but later realize the real killer is the cost. Especially with recycling strategies, leveraged yields, and collateral financing, you can't just look at "how much you can amplify" but also "whether the cost will flip after amplification." This is where TermMax's value lies: it turns leverage and lending from "guessing as you go" into "calculating first, then executing." This doesn't mean fixed rates are always better than floating rates, but it gives users another option. When you expect increased market volatility or need stable capital usage for a period, fixed rates become more meaningful. Because at least you know when the money is borrowed until, what the cost is, and what you face at maturity. I think this is exactly where TermMax fits with current hot topics. The market isn't short of opportunities; it's short of controllable opportunities. Not short of high-yield narratives, but short of funding costs that can be calculated in advance. Not short of leverage tools, but short of structures that keep leverage from getting out of control. Of course, fixed rates aren't a free pass. Collateral prices fluctuate, liquidation risks remain, and pool depth and maturity liquidity must be considered. TermMax can help you lock borrowing rates but can't lock market direction. This must be made clear. So now, I don't see TermMax as just a DeFi protocol chasing trends, but more like a capital management tool during volatile markets. The more chaotic the market, the more you need to calculate clearly. Truly mature DeFi doesn't make users more impulsive to leverage but lets them know their costs, terms, and risk boundaries before leveraging. TermMax's fixed-rate lending precisely meets this demand. The market is moving; fixed-rate lending should be revalued. Because the next phase might not be about who dares to rush in, but who can lock in funding costs before rushing. @TermMaxFi #TermMax

Not sure if it's because I've been taking too much allergy medicine recently, but I've been feeling drained after doing a bit of work lately. Today, I just got up and claimed the Bybit reward, received 131u, and then rested until now. I think the xhunt event is pretty good. I've participated a few times before, but the competition was a bit intense so I stopped. It also started with realgo, and it feels like the threshold has lowered and more spots are available. Everyone can join in when they have time. Today I saw that Termmax's TVL broke $100 million. This is not a small number; it means fixed-rate lending is starting to be validated with real money. In DeFi, user numbers can be boosted by events, interaction volume can be increased by tasks, and hype can be built through narratives, but TVL is a vote with capital. Especially since TermMax is not about Meme or short-term sentiment plays, but a product focused on fixed-rate lending and capital management. So the significance of $100 million TVL is not just about "looking impressive," but it shows that a real batch of funds is willing to enter TermMax to try fixed-rate lending, fixed income, leverage strategies, and maturity management. This is the key. TermMax solves a very straightforward problem: Borrowers want to know their funding costs in advance, and lenders want to lock in their return boundaries ahead of time. Simply put, it moves DeFi lending from "rates jumping wildly with the market" toward "more predictable costs." During bullish sentiment, people might prefer to see high APYs, high leverage returns, and aggressive narratives. But the real capital managers fear not the lack of exciting returns, but sudden cost spikes. Profits today can vanish tomorrow if borrowing rates change, potentially breaking the whole strategy. This is why $100 million TVL deserves special emphasis. Because fixed-rate lending can't just be talked about; it must be validated by capital depth. Without capital accumulation, even the most beautiful interest rate model is just theoretical; but when TVL surpasses $100 million, it at least shows TermMax has moved from "concept understood by some" to "capital willing to come in and try." Of course, $100 million is not the end. Next, we need to see if this capital can stay, if lending demand naturally occurs, and if the maturity pool depth can continue to grow. If the money only came from short-term events, the excitement will fade and the funds will leave; but if users really start using TermMax to manage borrowing costs, fixed income, and leverage strategies, then the value of this $100 million TVL will keep increasing. So now when I look at TermMax, I don't just look at what new features it launches, but at the capital behavior behind this $100 million. Capital willing to stay means fixed-rate lending is not a niche concept. Capital willing to price by maturity means the on-chain interest rate market is truly maturing. Capital willing to be used repeatedly means TermMax has a chance to evolve from a DeFi protocol into a fundamental infrastructure for on-chain capital cost management. $100 million TVL is a very important milestone for TermMax. It is not the final answer, but at least it shows the market has started voting for fixed-rate lending with real money. @TermMaxFi #TermMax

Jeonlees reposted

Everyone is focusing on cloud-based Agents, but I have integrated multi-Agent systems into a desktop application. In the tech community, there are more and more articles about multi-Agent systems, but most revolve around frameworks: ➢ LangChain ➢ AutoGen ➢ CrewAI This time, I want to talk not about frameworks, but a more practical question: Is it possible to build a "living" multi-Agent collaborative system from scratch using only a desktop application without relying on cloud infrastructure? The answer is: Yes. --- And it’s not just a demo. This system is embedded in a real desktop Web3 application and already integrates: - EVM / Solana dual-chain monitoring - SWAP aggregated trading - On-chain new token tracking - Trading dashboard - AI deep analysis Multi-Agent systems are not designed from PPT slides but naturally evolved in production environments. --- Before discussing the technical implementation of multi-Agent systems, I want to answer a directional question: Why did I ultimately choose the desktop over the more mainstream cloud deployment or lighter browser extension solutions? The essence of this choice is not "which is more advanced," but a trade-off among three technical paths. --- Cloud deployment is currently the most mainstream way to implement multi-Agent systems. Its advantages are obvious: - You can add GPUs to the Agent team anytime - Model upgrades don’t require user intervention - The server can maintain global shared memory But the costs are also clear: - User data must pass through the server - On-chain transactions often require the server to have private key access - Continuous deployment and maintenance costs --- Browser extensions are another lightweight route. They can directly inject into pages, read the DOM, simulate user actions, and are very efficient for single automation tasks. But the problems are straightforward: > They run inside the browser sandbox > Lack persistent storage capabilities > Lack long-running background threads > Hard to support complex memory systems > Difficult to support asynchronous interaction between Agents --- Desktop applications occupy a unique position. They have full system resource access: ➢ Can start background threads automatically ➢ Can read/write local file systems ➢ Can establish persistent database connections These capabilities are exactly the underlying infrastructure multi-Agent systems truly need. Of course, there are costs: It depends on local computing power; model inference usually still calls cloud APIs It requires a full Python environment Updates and distribution are more complex than web apps. --- So, choosing desktop to build multi-Agent systems is essentially trading distributed capabilities for data privacy and scheduling efficiency. This is not an absolute advantage but a scenario-driven choice. For systems like cryptocurrency trading, on-chain analysis, and monitoring, data privacy requirements are much higher than typical applications: > Wallet addresses > Transaction history > Position data Having this information on cloud servers is itself a risk surface. --- More importantly, scheduling efficiency. In a monolithic desktop app, the main Agent dispatches tasks to sub-Agents without HTTP/RPC network protocols, but via direct in-process calls. This means: → Network overhead is completely eliminated → Call latency drops from milliseconds to microseconds For high-frequency analysis, on-chain monitoring, and trading assistance, this difference directly impacts system timeliness. --- The first core challenge of multi-Agent systems is not "how to make them chat," but "how to isolate them." Main Agent, contract analysis Agent, security audit Agent, plus the user—if chat records and memories are mixed, identities get confused. Once confused, information loss and erroneous reasoning can happen anytime. --- My approach: Unified code templates, isolated runtime data. All sub-Agents share the same `engine`, but physical data isolation is achieved through dynamic table names: `table_name = f"chat_history_{self.agent_id}"` Then the corresponding table is created automatically. In other words: Trader Agent generates `chat_history_trader` Audit Agent generates its own `chat_history_xxx` Main Agent has its own independent chat table This is not logical isolation but physical isolation at the database level. --- Reflection notes follow the same design. Each sub-Agent records its own reflection key: `reflection_key = f"auto_reflection_{self.agent_id}"` `self.api._agent_remember("master_insight", reflection_key, summary)` So each Agent accumulates only its own long-term reflections, without polluting other Agents’ memories. The essence of this solution is: Designed once, reused for life. --- Adding a third or fourth Agent later requires no core code changes. Just copy the directory structure and add configuration files. The template engine automatically generates for it: → Independent database tables → Independent reflection keys → Independent chat storage areas This makes me increasingly believe: A good architecture isn’t necessarily more complex, but definitely easier to reuse. --- Next is the scheduling problem. In distributed systems, the main Agent dispatches sub-Agents usually relying on: - HTTP / RPC communication - Service discovery - Load balancing But in a monolithic desktop app, I simplified this to a direct function call: `sub = self.sub_agents[agent_id]` `result = sub.process(task, save_history=False)` This is "command, not request." --- This "relay-style scheduling" has two benefits: First, latency drops from milliseconds to microseconds, all data circulates locally Second, the main Agent doesn’t need to know sub-Agents’ internal implementations, only: "What types of tasks it can handle" This is a clear boundary of responsibilities. --- To make the main Agent truly "dispatch work," I dynamically inject all sub-Agents’ capability lists into the main Agent’s system prompt. For example: Contract analysis Agent: available tools `get_contract_market_data`, `run_contract_risk_check` Security audit Agent: available tools `check_token_security`, `check_token_audit_binance` So when the main Agent receives user commands, it can automatically judge the task type and select the appropriate sub-Agent to execute. --- Permission control is one of the core security issues in the entire multi-Agent system. The main Agent holds 26 Web3-specific tools covering: SWAP quotes On-chain analysis Security checks Data queries But each sub-Agent should only use its small subset of tools. So first, I implemented strict whitelist filtering at the code level: `return [t for t in all_tools if t["function"]["name"] in self.allowed_tools]` --- But code filtering alone is not enough. Because large models can hallucinate and may try to call unauthorized tools. So second, I wrote a "tool usage iron rule" at the end of the system prompt: You only have the following tools and must never cross boundaries. If a task requires other tools, you must explicitly tell the boss you don’t have permission. Code layer ensures "can’t see" Prompt layer ensures "won’t cross boundaries" This is the dual-layer protection I implemented for permission isolation. --- There is also a design I personally like but is the least conspicuous: I created a separate "break room" for the entire Agent team. Most multi-Agent systems on the market only do "user -> Agent" interactions. Agents don’t communicate with each other. But I designed a separate `agent_interactions` space to allow asynchronous interaction between Agents. --- Its trigger mechanism is very simple: `selected_id = random.choice(list(api.sub_agents.keys()))` `selected_sub = api.sub_agents[selected_id]` Each trigger randomly selects an Agent, dynamically generates a round of dialogue, writes it back to the database, and the frontend renders it in real time. I also added a background check thread and infinite loop prevention: Every 2-3 minutes, check the last message If the last 3 messages are auto-replies, pause automatically To prevent them from chatting endlessly late at night. --- The value of this "break room" is not in directly creating business revenue but in subtly shaping the system’s personality. It doesn’t emphasize its existence, but quietly maintains the Agent team’s cohesion, character relationships, and health. You barely notice it, but the system becomes more like a "living team" because of it. --- In memory layer design, I ultimately did not introduce vector databases but continued deeply customizing SQLite. Not because of technical conservatism, but because engineering decisions must balance constraints. For desktop apps, every additional dependency adds a failure point, a security risk, and packaging burden. The result is: > Several SQLite tables > Dynamic table names > Structured JSON fields Support independent memory systems for 3 or more Agents. --- This memory system now forms a complete chain: Short-term conversation memory (20 entries) -> Long-term reflection notes (every 6 hours) -> Structured JSON records And I am continuing to advance in 8 directions: ➤ Context continuation ➤ Memory structuring ➤ Memory-driven behavior ➤ Three types of long-term memory in psychology ➤ Team collaboration memory ➤ Dynamic evolutionary memory ➤ Knowledge graph memory ➤ Memory compression and efficient retrieval My goal is not to have Agents just remember what you said, but to evolve from a tool that remembers what you said into a long-term partner who understands, predicts, and collaborates with you. GitHub: Author: Powerpei (Xiao Nan)

Why is it that when I scroll through Twitter today, everyone says that after completing this task, the TGE will happen? Is that true 🥺? Then I need to hurry up and catch up. Yesterday I saw the project team also fixed the related bugs, looking forward to it going live soon. TermMax is not just a simple lending pool; it focuses on fixed-rate lending and one-click leverage. Simply put, users can lock in borrowing costs before entering a leverage strategy, so they don’t have to worry about floating rates suddenly rising while executing the strategy. TermMax’s official website clearly states that it offers fixed-rate borrowing, lending, and one-click leverage, emphasizing predictable rates, fixed borrowing costs, and single-transaction leverage. This feature is quite crucial in the current market environment. When the market is stable, people might think fixed rates aren’t that attractive because with no volatility, floating rates aren’t necessarily painful. But once the market accelerates and capital demand rises, floating rates can become very unstable. You might want to execute a yield strategy, but if borrowing costs change, your profit margin gets eaten up, or you might even have to adjust your strategy. This is where TermMax’s value lies. It’s not about mindlessly increasing leverage, but about making leverage more calculable. When borrowing costs and terms are locked in advance, users can better judge whether a strategy is worth pursuing. For those who truly manage funds, “predictability” is sometimes more important than “seemingly high returns.” Especially the one-click leverage feature—I think it shouldn’t just be understood as “fewer button clicks.” In traditional DeFi, doing recursive leverage often requires multiple steps: collateralizing, borrowing, buying more, and re-collateralizing repeatedly. The more steps, the higher the chance of slippage, gas fees, interest rate changes, and operational errors. TermMax compresses this into a more direct leverage entry, essentially reducing strategy execution costs. Their website also mentions that one-click leverage features single transactions and no manual looping. Of course, this doesn’t mean the risk disappears. Leverage is still leverage. Fixed rates only help lock borrowing costs; they can’t lock collateral prices or guarantee profits. Users still need to watch collateral volatility, liquidation thresholds, term expirations, pool depth, and exit liquidity. Viewing one-click leverage as a “more convenient risk management entry” is fine, but seeing it as a “blindly amplify returns button” is dangerous. So the main point I want to express in this article is simple: After the market becomes active again, what’s worth watching about TermMax isn’t whether it makes people more aggressive, but whether it makes leverage strategies more controllable. DeFi doesn’t lack tools to amplify returns; it lacks tools that clarify costs, terms, and risk boundaries before amplifying returns. If TermMax can continue to improve fixed-rate lending and one-click leverage, its positioning won’t just be “another lending protocol” but more like an on-chain capital efficiency tool. The more volatile the market, the more these tools stand out. After all, truly mature leverage isn’t about daring to rush in, but about being able to calculate the numbers clearly before rushing. @TermMaxFi #TermMax

The lottery results for Ledong Robotics have been announced, and I didn't get in again. But the good news is that there are two new stocks for Hong Kong IPOs. Seeing this, I quickly went to subscribe, and the deadline is 9 AM tomorrow!! To conclude, both are worth subscribing to, but the main funds should be focused on Jitai. #Jitai Technology-P (07666) One lot costs 5302.95 HKD, with 500 shares per lot. 1️⃣ Strong sponsorship background: Jefferies + Deutsche Bank + CITIC Securities. International investment banks combined with Chinese brokers, a strong allocation. 2️⃣ Large issuance scale, low public float. Global offering of 201.229 million H shares, with about 10.0615 million H shares initially available to the public in Hong Kong, only 5%, raising up to approximately 2.11 billion HKD. Once it heats up, the winning rate will likely be low. 3️⃣ Strong cornerstone lineup. Public information shows Jitai Technology has introduced cornerstone investors like BlackRock. Based on the issue price, cornerstone investors will subscribe to about 110 million shares, which is the most important plus for this stock. 4️⃣ Hot theme: AI nano delivery / AI pharmaceutical platform. It talks about the AI-driven nanotechnology innovation system NanoForge, and platforms AiTEM, AiLNP, AiRNA. Simply put, it uses AI and nanotechnology to solve drug delivery efficiency issues. ⚠️ Risks are also clear: Jitai Technology is an 18C specialized technology + uncommercialized company; its valuation mainly relies on the technology platform and future expectations, not current profits. #Inpaier Pharmaceutical-B (07630) One lot costs 4393.88 HKD, with 200 shares per lot. IPO price range 19.75-21.75 HKD, expected to list on May 13. I will subscribe to this one but not heavily. 1️⃣ Strong sponsorship background: Goldman Sachs + CICC. Pharma-B stocks rely heavily on institutional recognition, and this sponsorship combo is a plus. Issuance-wise, global offering of 41.977 million H shares, with 10% initially available to the public in Hong Kong, raising up to about 910 million HKD. 2️⃣ Cornerstone investors are notable. Tencent, LAV, Ruiyuan, etc. participate in cornerstone subscriptions, totaling about 35.87 million USD, approximately 281 million HKD. For an 18A innovative drug company, this background is solid. 3️⃣ Core label: synthetic lethal tumor drug. Its highlight is tumor treatment and synthetic lethality, a rare theme but with a high professional threshold. ⚠️ Risks to consider: Inpaier Pharmaceutical is an 18A unprofitable pharma stock; future depends on pipeline, clinical trials, and commercialization, not a safe bet. Currently, Jitai Technology subscription is 3056.63 times oversubscribed. Inpaier Pharmaceutical is 653.8 times oversubscribed. Hope we all get lucky 🙏 Personal advice: participate rationally and DYOR.

Caught up on it last night, but it suddenly showed 2 days, a bit confusing, but at least it's done, waiting for the XP points to surge after 15 days 😋 Seeing everyone I know having over 300k MP, I'm so jealous. I've been catching up recently too. The projects I was working on also got reversed a few times, but staying at the table always means there's a chance 💪 TermMax recently mentioned an approach in their official updates: earning fixed returns on stablecoins on TermMax, for example, 8% fixed yield in stables, backed by re’s PT-reUSD. I think this point is more worth writing about than just shouting “high APY” because it addresses what the market currently lacks most: not more exaggerated returns, but clearer yield structures. The problem with many DeFi yields in the past wasn’t low returns, but that the returns were too volatile. Today the APY looks good, tomorrow when funds enter the pool, the yield gets diluted; today subsidies are still there, tomorrow the event ends, and rates drop immediately. Users seem to be managing finances, but often they’re just guessing when project subsidies will stop. TermMax’s fixed rate logic hits this pain point exactly. Its core isn’t to tell you “I have the highest yield,” but to let users know in advance the returns or borrowing costs over a period. TermMax’s official positioning is a fixed-rate borrowing & lending marketplace, emphasizing lock predictable rates, amplify yields, and deploy strategies with confidence. Simply put, it aims to solve the most annoying uncertainty in DeFi lending. This approach fits well with the current market environment. Stablecoin yields are no longer in a wild growth phase. More and more users are shifting from “where is the APY highest” to “can this yield be clearly explained.” Especially as RWA, stablecoins, and fixed income start to appear together, the market increasingly cares about term, underlying assets, yield sources, and exit paths. So I think TermMax is more worth watching this time, not because of the 8% figure itself, but because of the product concept behind it: gradually pushing stablecoin yields from floating, short-term, pool-snatching to fixed, predictable, and plannable. Of course, we can’t be blindly optimistic here. Fixed income does not equal risk-free. The asset structure, term, liquidity, redemption path, and counterparty risk behind PT-reUSD all need to be clearly understood by users. TermMax can provide a fixed rate market, but that doesn’t mean all fixed income products can be bought with eyes closed. Especially in DeFi, the biggest problem is mistaking “fixed income” for “principal-protected income.” But in terms of direction, I think this topic is very suitable to write about now. Because the market has passed the stage of only looking at high APY. The next round of stablecoin products that truly retain users will likely not be those shouting the highest, but those who can clearly explain yield sources, term structures, and risk boundaries. TermMax’s fixed-rate lending is exactly filling this gap. @TermMaxFi #TermMax