Alan Sapède

What a week at @EFDevcon in Buenos Aires 🇦🇷 We kicked off with @ffarall giving a talk on “Beyond the Runtime – Unlocking the Substrate Node Potential" at @Polkadotsub0. Then @BridgetHearst moderated "The Age of Autonomous Economies: How AI Agents Will Redefine On-Chain Payments" at @openagisummit Huge congrats to Moonsong's very own @tomimorenox and the @zksync team for winning the @UniswapFND Cup 🏆⚡️ And we closed out the week at our Living Knowledge Café y Conexiones with @aleph_im @miranetwork @Gaianet_AI @ZKamigos. Thanks to all that attended! Buenos Aires delivered. On to the next. 🌙💫

@dani_avila7 Something I've been experimenting in github.com/Moonsong-Labs/… (fully vibe coded) is to include documentation about the "why", which is always missing in projects

Claude Code documentation agent keeps project docs up to date with Docusaurus. Created a Docusaurus site: npx create-docusaurus@latest my-docs classic Added the agent: npx claude-code-templates@latest --agent=documentation/docusaurus-expert --yes Then prompt: "Use docusaurus-expert to review that the documentation is correct and up-to-date" Now the docs stay synced with code changes. That’s how I set up docs.aitmpl.com, as shown in the video.

@66777512 @moonsonglabs_ Since all our projects using it are open source, we don't have a fine-grained access as the privacy part is low and the access is restricted to the company. We could add it however, since few of you asked about it

@moonsonglabs_ THIS COULD TURN DEV WORKFLOWS INTO A TEAM SPORT - HOW DO YOU HANDLE LOG ACCESS?

🚂 Introducing Agent Prompt Train 🚂 An open-sourced high-performance proxy that works with Claude Code to help turn it from a single-developer tool into a collaborative development platform. Claude Code changed the game. Now, Agent Prompt Train is here to help teams: ✅ Log every conversation for review ✅ Analyze & optimize prompts with AI ✅ Turn best practices into reusable team commands ✨ The payoff: Smarter prompts, faster learning, and shared knowledge across your whole team. 📖 Read more on the blog & test it out 👇 moonsonglabs.com/blog/agent-pro…

@tagvlk @moonsonglabs_ The initial target for the development was our internal teams, which only work on open source so the risk of privacy is low. But that gives me the idea... We could also have the PromptTrain checking with an LLM if a prompt is "safe" or not. That wouldn't be 100% but a good start

@moonsonglabs_ agent prompts get smarter how will you guard privacy

@ili_boo1 @moonsonglabs_ Those would be cool features yes, I'll see if I can add them. For now I mostly spent time on productivity features

@moonsonglabs_ team prompts need governance: add role-based access to logs

@0xFelixHFT @moonsonglabs_ It doesn't run things automatically, but you can plug it to your automated process. Think of it as a proxy

@moonsonglabs_ can it run prompt CI so PRs break the brain before prod

@petnicken @moonsonglabs_ In addition to prompt libraries we also have knowledge sharing meetings to consolidate our prompts. But it requires some effort to maintain them.

@moonsonglabs_ impressive, curious how you scale prompt libraries across teams without log fatigue

@useblacksmith Cool feature. That'll prevent having to do reverse ssh from the machine directly.

Today, we’re launching an SSH feature for GitHub Actions. Here’s how we built it: - iptables to route network traffic to the right VM - CoreDNS and Redis plugin to have real-time DNS record updates - Active SSH sessions detection to give users a grace period For the full story, check out our deep dive below.

‼️PROMO CODE ALERT‼️ 🍩Glazing is for donuts, not your AI. 👀With open-weight models, what you see is what you run. 👉Use code KLUSTERGEMMA by Thursday 5/15 at midnight to get $20 in credits. ⏳Once activated, you’ve got 7 days to use them.

🚨M3-Embeddings just hit kluster.ai - and it dunks on OpenAI’s models. 👌Fully compatible with OpenAI’s API. Just better results: 🔹 +30.3% (multilingual) 🔹 +19.6% (long docs) 🔹 +8.6% (cross-lingual) ➡️Try it here → kluster.ai

@GeminiApp Any date for including it in Gemini Code Assist ?

Gemini 2.5 Pro is taking off 🚀🚀🚀 The team is sprinting, TPUs are running hot, and we want to get our most intelligent model into more people’s hands asap. Which is why we decided to roll out Gemini 2.5 Pro (experimental) to all Gemini users, beginning today. Try it at no cost at gemini.google.com

🤖 What are AI Agents? At their core, they’re autonomous software systems that can perceive, decide, and act in pursuit of a goal—without constant human input. At Moonsong, we’re exploring a few distinct ways to think about agents—and those perspectives are actively shaping how we build. Read more in our latest blog post ⬇️ moonsonglabs.com/blog/what-are-…

New chains, new ecosystems, and more tokens demand better infrastructure. That’s why we’re excited to be part of @lifiprotocol next chapter with Catalyst and Glacis. Catalyst will push aggregation to a new level with intent-based protocols, while Glacis makes it possible to support every major interop token standard, starting first with OFT, NTT, and CCT. Warp Tokens, ITS, SupERC20, and more will have support added right after launch. This means better pricing, zero slippage, and a frictionless user experience across any chain, in any size.

Amazing improvement coming to Moonbeam for next release (next week?) Stay tuned... :)

💫It's here! We’re opening Early Access for kluster.ai – the platform that empowers teams to run large scale batch inference models at a fraction of the cost. Join to unlock $500 in credits! 💰 Learn more and sign up here👇 kluster.ai/early-access/ #ArtificialIntelligence

You’ve probably heard a lot about AI agents recently. With advances in LLMs and automation, we’re seeing software programs capable of setting and achieving goals independently—taking automation to a whole new level. So, what exactly are AI agents 🤖 and why are they interesting🤔? AI agents are autonomous programs powered by LLMs, designed to perform tasks on their own. They’re given goals and use AI-based reasoning to figure out how to achieve those goals—without ongoing human intervention.💡 These agents don’t just think—they act. They can use tools like external API calls or execute actions on Web3 platforms. By observing the effects of their actions, they create feedback loops, learning and adapting as they go. An AI agent with its own X account, posting with specific objectives… check out @truth_terminal and @luna_virtuals—both generating posts autonomously and working toward their own goals.✨ AI agents are catching on in enterprises  💼, too. Companies are deploying them for complex tasks like customer support, data analysis, and back-office operations—tasks that traditionally needed human oversight. In the coming years, we expect a proliferation of AI agents across different domains. Many will operate in controlled enterprise environments, but others will interact openly on social media and blockchain networks. 🌐 This shift opens up enormous potential to build agent-centric infrastructure. These agents are becoming a unique type of ‘user,’ and as their numbers grow, so will the need for protocols designed specifically for them. At Moonsong, we’re focused on this agentic future, building the tools and infrastructure needed to support it. Follow us to stay updated as we explore the possibilities of an agent-driven world.🌍

In our latest blog by @jayfarei, we present Unravel—a prototype to decode Ethereum transactions and improve user clarity. You can check it out at unravel.wtf!

@itsbirdo_ @derekyoo That makes sense. I'll check if we can add it. One challenge is that some steps are sequential and might be already cached making it a bit unpredictable

@derekyoo this is super cool! A UX improvement could be to add the estimated wait time for an unravel. If I know they normally take 70s to process, I'll stop watching it count up and come back later ;)

1/ unravel.wtf is one of our AI experiments in “learning by doing”. We wanted a real use case we could dig into - user intent interpretation - in a domain we were knowledgeable about - blockchain transactions. Read on for key insights ⬇️