b/ai curious

blockchain/AI infra & dev tooling updates

This newsletter will function as a smörgåsbord of updates and developments across blockchain infra, DePIN middleware, onchain dev tools, smart contract languages and frameworks, account abstraction, farcaster frame tooling, RPC/API providers, shared sequencers, and AI dev stuff. Plus, whatever else I may find fascinating. Curiosity is the guiding factor.

DM/DC me things you think should be included.

I am thinking of publishing either once a month or every two weeks.

Rollups-as-a-Service

Conduit

Conduit is providing organization admins the ability to enable external nodes. This allows anyone to sync and run nodes for your production-grade rollup.

By enabling external nodes, an opportunity emerges for incentivized RPC protocols like Pocket Network or Lava Network to support Conduit rollups. Decentralize your network in one click, earn POKT and/or LAVA for running nodes.

The other week Conduit introduced RPC keys, expanding and updating their RPC offering. With one API key, developers get access to every Conduit chain. They can expect high uptime / performance and low costs.

Reported at the end of last month, Conduit was responsible for sequencing ~20% of Ethereum TXs, which includes transactions from 20+ mainnet rollups on Conduit. I mean look how easy it is to deploy a rollup. No code, just a few clicks. Reducing friction helps. You can even leverage a custom gas ERC20 token in an Arbitrum Orbit L3 rollup.

Worth a listen. Overtime, Conduit plans to reduce costs.

If you want to be on the frontier, they are hiring:

RPC Providers & Protocols

LlamaNodes

v2 released. With it comes: improved MEV protection, better infrastructure via RETH, and revamped premium UX.

Built in Rust, this newer Ethereum client Reth, developed by engineers at Paradigm, allows for much higher RPC concurrent request capacity. It includes a quicker sync time as well as a lower disk usage for archival nodes. Out of this emerges a much lower latency and much faster transaction inclusion.

The llama team plans on adding new reth clients for other chains as those launch. It seems like these llamas have more in store... hint: get paid to use LlamaNodes RPC.

Pocket Network

Earlier this year, Pocket Network announced the launch of Gateway Server, built and maintained by Nodies. It's a lightweight HTTP server (1GB RAM | Storage and 4vCPUs) that integrates the Pocket Network protocol to send/receive RPC requests. You don't have to run resource-intensive nodes anymore, others are incentivized to do so. Just spin up the server. It costs less than 1k USD a month to operate the gateway server across 3 regions.

By leveraging the Gateway Server, onchain applications will be able to tap into any blockchain supported Pocket Network's incentivized pool of nodes. Even Nodies DLB uses the gatway server to power public RPC endpoints, serving ~1k rps through the protocol. It serves as the foundation for every new gateway operator participating in the network. In the spirit of opensource, other developers and organizations can contribute to this simple framework by expanding QoS capabilities and adding blockchain support. For example, Chainstack has contributed to v0.2.0.

Liquify launched a gateway within Pocket Network, making it the 3rd one to go live. It leverages the open-source gateway server maintained by Nodies.

They are already driving ~10M requests over the last 24 HRs according to POKTscan, mostly from Solana and Fuse. Great timing since Liquify will be able to take advantage of the new QoS checks for Solana released in the v0.3.0 Gateway Server update.

Route to POKT coming. When enabled, it will unlock significantly more traffic. Opt-in decentralization.

There are even more gateways coming online soon, some by the end of the month. The Pocket Network Foundation announced three new gateways:

  • Developer DAO

  • Raid Guid

  • Chainstack

I believe Pocket Network is primed to 10x similarly to Farcaster when the demand-side of the marketplace becomes permissionless, especially if the PNF team continues to onboard more and more gateways ahead of the next major protocol upgrade Shannon.

PUP-33: Scale Parameters and Capacity for Evolution (SPACE) updates two parameters to scale and account for more relays, more gateways, and more services (i.e. blockchains, rollups, node clients, and LLMs) prior to the Shannon upgrade. One SPACE parameter doubles blockspace creating more space for onchain sessions to exist, which enables additional gateways and services. The other parameter increases the minimum proofs per session for a node to claim their work, which reduces the usage of existing blockpace.

This is important because it allows for additional participants to drive demand. 

The open-source gateway server will act as an exponential accelerator for growing traffic routed through the network.

The question becomes from which baseline does Pocket Network 10x from: 500M requests, 1B requests, 5B requests, or 10B requests?

Shannon public testnet launches in the next couple of weeks. Documentation to come with it. More gateway alpha in the builder call recording below.

I was curious about how allowlisting chains/services will work in the Shannon upgrade. So, I asked in the discord core-research channel. It's not decided yet BUT the team is leaning towards permissionless with a POKT fee (or something in that direction). This leads me to believe that the fee would contribute to burn, and therefore onchain revenue.

The permissionless model is critically important to enable rollup/rollchain/subnet-as-a-service gateways to sprout on top of the Pocket Network protocol. The whitespace provided by a new parameter pos/RelaysToTokensMultiplierMap allows for flexible reward structures. With custom RTTMs, node operators can be incentivized to host brand new chains or quiet chains with little traffic by adjusting the network parameter higher. A useful mechanism to help bootstrap a pool of nodes for rollups/rollchains/subnets. Non-inflationary models can be composed by setting mint equivalent to burn in these custom RTTM chains.

This presents an opportunity to grow the relay pie. Imagine the burn if DEGEN L3 chain launched through a pokt-native RaaS instead.

It costs 0.85 USD per million relays, a gateway fee service providers must pay, which the DAO burns. As the purchasing power of POKT increases, it will become more cost-effective to run services like RaaS gateways. In order to compete and reduce costs further, the gateway fee parameter could be voted on.

Look what happened with base after EIP-4844 was implemented. Induced demand. TXs spiked once gas costs went significantly down for L2s.

Lava Network

Lava Network released a whitepaper and introduced the LAVA token. The universal work token has a fixed supply of 1B LAVA, no more tokens will be minted.

To attract node operators 6.6% of the supply is set aside for "Provider Drops", a monthly reward distribution. Undistributed validator rewards are burned at the end of month. The range of burn rate will fluctuate between 0%-6.6% of the total LAVA supply. Token holders will be able to delegate and stake LAVA with either validators and providers.

As Lava announced their tokenomics, Rollchains shared that soon every rollchain will be able to quickly bootstrap a pool of nodes by tapping into Lava Network. An incentivized RPC pool, funded with rewards from the native token a rollchain, can be created for each one.

SDKs, Frameworks, & Languages

Scaffold-ETH 2

BuidlGuidl pushes out a base starter kit to help build onchain applications. Scaffold-Base is a loosely maintained fork of Scaffold-ETH 2. It provides native support for Base mainnet and Base Sepolia testnet. You can even fork mainnet Base to leverage as a local hardhat chain, which you can then use to interact with forked Base contracts.

It comes with the coinbase-sdk-wallet beta preconfigured. This allows for account abstraction using passkeys. The Coinbase Smart Wallet will automatically appear when the target network is set to baseSepolia.

You can earn a share of 5k USD if you submit a SE2 onchain application to the BuidlGuidl WaveHack Global online hackathon.

I made a custom gpt for Scaffold-ETH 2 to assist with onchain app development. It's trained on docs crawled from ScaffoldETH, WAGMI, VIEM, Rainbowkit, Hardhat, Foundry, DaisyUI, NextJS, and Solidity by Example.

FROG   

FROG is a developer toolkit for building frames on Farcaster. With a few lines of code, you can create high-performant, lightweight frames.

It comes with a local, feature-rich frame debugger available under localhost:3000/dev. Tightening the feedback loop between development and frame interface.

Frame Interface Guidelines (FIG) is a collection of best practices for building and designing frames. FROG UI is an extension of the frog framework. It provides type-safe layout and styling primitives to assist with building dynamic, extensible, and consistent frame interfaces. The systematic approach gives developers the ability to rapidly compose layouts and manage colors, icons, etc. Wonderful for prototyping frames, especially during weekend hackathons.

Here are a set of frame templates available on figma to help you get started.

I made a Farcaster Frame custom gpt to assist with frame development. It's trained on docs crawled from farcaster, frog, frames.js, hono, airstack, neynar, privy, dynamic, xmtp, base, onchainkit, and pinata.

DePIN x AI

Livepeer

Node operators on Livepeer are earning ETH for performing text-to-image inference on their AI subnet.

Just like that, a new stream of revenue becomes available for GPU owners through image generation. Text-to-Video inference will open another, more lucrative revenue stream. The coupling of generative video with the transcoding needed for distribution presents a synergistic opportunity, especially as the cost of production decreases over time. 

In the meantime, creators can dream up images on a open-source StableStudio platform from Stability AI powered by decentralized p2p infrastructure.

An experimental version of Livepeer Cloud Gateway launches with 3 strategic gateways deployed across the US and EU regions. They're planning to boost transcoding and AI inference demand; as well as, diving deeper into the Livepeer AI subnet.

AI-enabled Developer Tooling

Cursor

Cursor introduced Copilot++ at the end February.

It automates the tedious, low-entropy, predictable steps. Copilot++ has the context of your previous edits so it can infer your intent and predict your next change. Aman predicted that over the short-term Copilot++ will get significantly faster, even smarter, and predict much more than just edits. The Cursor team delivered.

By the end of March, Cursor's Copilot++ became ~2x faster from inference optimizations and their best model. As of early April, Cursor users are now able to use the new gpt-4-turbo model, which the Cursor team observed better reasoning on complex tasks. A few weeks later, the team rolled out a major performance upgrade to Cursor Copilot++. The same model is ~40% faster now overall with additional performance on small edits.

Looking forward to what is next up for Copilot++ (i.e. more intelligence, more capabilities, and even more context).

Similar to v0, you can leverage screenshots when coding alongside AI but with additional context from your codebase and documentation. I can see this become useful when putting together a front-end for a smart contract.

Replit

Replit released Code Repair. It automatically fixes your code in the background.

Replit Teams combines the power of multiplayer mode and AI into a shared developer workspace.

The iterative process continues to accelerate. With the figma to replit plugin, you can go from design to code in seconds.

I've used Replit to fork around and find out with Scaffold-ETH 2 and Scaffold-Base. Advanced port configurations made this possible, particularly localhost port handling and webview port switching. I wanted to be able to spin up SE2 from my phone in case I had a spark of inspiration, so I made it into a replit template for Scaffold-Base. I like the idea of building onchain applications with the assistance of Replit AI while on the go, then picking it back up either on a laptop or tablet later.

v0

Generate UIs with text prompts or screenshots/images.

Edit feature added to v0.

Vercel open-sourced the rendering tech behind v0 with AI SDK. This provides the tools for developers to build AI-enabled applications with relative ease.

Before copying/pasting into your IDE, you get feedback on v0 generations privately by sharing a link with a colleague.

Atlas ZK

Write smart contract code in natural language. Test and deploy to any EVM compatible blockchain.

With a fresh smart contract generated, you'll need a front end. The Atlas team made a dapp starter kit that automatically generates components and hooks from your atlas contracts. It leverages WAGMI, a collection of useful react hooks for interacting with contract functions.

This makes it possible to generate onchain MVPs in minutes. For example:

So

We can create at the speed of thought:

  • Generate smart contracts.

  • Generate front ends.

  • Iterate with AI-enabled dev tools.

  • Generate images (videos soon).

Then, click a few buttons and change a few lines of code:

  • Deploy rollups in minutes.

  • Craft unique in-feed experiences with Farcaster frames.

  • Tap into incentivized onchain infra via Pocket Network or Lava Network.

Loading...
highlight
Collect this post to permanently own it.
weeb3dev logo
Subscribe to weeb3dev and never miss a post.
#dev tooling#rpc#onchain infra#depin#ai#rollups
  • Loading comments...