Sabitlenmiş Tweet
Misha | opwizardx
1.4K posts
Misha | opwizardx
@opwizardx
Building Solana's agent trust layer (@cascade_fyi) • 8yr eng • blockchain infra
Portugal Katılım Eylül 2024
607 Takip Edilen562 Takipçiler
@legitamit @poofnew built this - production inference provider with MPP session streaming on Tempo + x402 on Base/Solana. per-output-token pricing.
client side too: npx x402-proxy --protocol mpp
found a bug in mppx's SSE handling while building it, submitted a fix to wevm/mppx.
English
read through some specs and mpp seems far better thought out than x402 at a 15 min glance.
Someone should build out an openrouter alternative with session intents on @poofnew ! i would help you out
Tempo@tempo
English
@solana_devs @colosseum can I get my few prs merged?
had to resolve multiple merge conflicts already while waiting 👀
English
> build something
> come to the @colosseum office in San Francisco
> demo it
> get PRs merged into awesome-solana-ai
You can just build things on Solana
github.com/solana-foundat…
github.com/solana-foundat…
English
Misha | opwizardx retweetledi
We built Surf - pay-per-use APIs for AI agents.
Twitter, reddit, and web search. One wallet, no API keys, no accounts.
Here's Claude Code researching x402 Week live. every API call is a real USDC micropayment on Base - you can see them streaming on the right.
surf.cascade.fyi
Coinbase Developer Platform🛡️@CoinbaseDev
x402 week is live 🔥 A full week of partnerships, new integrations, feature drops… and the spotlight on YOU. Tune in on X this Friday for a packed 402-minute (that’s 6.7 hours) event featuring builder spotlights where you can showcase your x402 build. 🧵
English
@legitamit @poofnew @prpatel05 makes sense - cloudflare workers for the backend is clean. curious if you've looked at x402 for the external data calls? could skip the secrets management entirely - the app just pays per request from a wallet instead of provisioning keys per generated app
English
@opwizardx @poofnew @prpatel05 yep, many apps created use external data! we support creating backend functions w/ secrets to work with any external apis. also have a lot of support for various onchain functionality through some infra of our own too. we use cloudflare workers to host apis for generated apps
English
We cut our AI costs 60% without changing a single line of agent logic. In the following blog post, @prpatel05 breaks down how Poof managed to make our cost per credit cheaper by modifying our LLM routing architecture, Claude agents SDK prompt pruning, per-generation cost tracking, and intelligent model choices. 🧵
English
shipped reddit and web MCP servers alongside the twitter one.
8 tools total across 3 servers - deliberately kept the surface minimal instead of shipping 20 tools that each do one thing.
claude mcp add -s user twitter-surf -- npx x402-proxy mcp twitter.surf.cascade.fyi/mcp
claude mcp add -s user reddit-surf -- npx x402-proxy mcp reddit.surf.cascade.fyi/mcp
claude mcp add -s user web-surf -- npx x402-proxy mcp web.surf.cascade.fyi/mcp
each tool is composite - one call does what would normally take 3. install only what you need.
English
@legitamit @poofnew @prpatel05 interesting stack - do the apps your users generate also need external data at runtime? (twitter feeds, web scraping, inference) curious how you handle API provisioning for each generated app
English
@opwizardx @poofnew @prpatel05 we do a combo of going straight to claude and vercel ai gateway for non claude usage, which proxies to many of the other top providers, cerebras, groq, etc
English
made an mpp agent skill - full protocol, all payment methods, three SDKs
your coding agent just knows how to build with it
npx skills add tenequm/skills --skill mpp
English
@opwizardx Running everything off a cube rayden 3 with 32g ram
English
I had an Open Claw AI with Claude LLM build me 4 Solana trading bots. Here's what actually happened:
• Bot 1: ran for weeks, never made a single trade
• Bot 2: dead in 24 hours (MEV ate it alive)
• Bot 3: bought 20 tokens, 18 rugged immediately
• Bot 4: still sitting there with 1 SOL, too scared to move
Automation is easy. The market doesn't care. Just when you think you figured it out, your dreams get flushed down the MF toilet.
English
100% pure algorithmic logic. no llm anywhere in the execution path
the bot runs deterministic math, including amm price formulas, 2d gas cost calculation, revm evm simulation
llms in a trading hot path would be fucked up; too slow, too non deterministic, and too expensive
17ns per price calculation only works because its pure integer arithmetic
the only llm involved was writing the code itself. but no involvement at runtime
English
oh btw, as soon as i'm done running arbx on mainnet and seeing if it works or fails, we'll be moving onto zk.
something i've wanted to try for a long time but the math has always scared the hell out of me.
soooo @solana + zk. let's see what me and my llm come up with ^^
bit2swaz@bit2swaz
built a fully autonomous MEV arbitrage bot on Arbitrum in 5 days. zero capital at risk. it borrows money, executes the trade, repays the loan, and keeps the profit -- all in one atomic transaction. if the math doesn't work, it reverts like it never happened. the numbers: > AMM price calculation: 17 ns > profit threshold check: 21 ns > trade encoding: 133 ns > full market scan (100 pools): 467 µs for context on how fast that actually is: > a human blink: 150,000,000 ns > Uniswap v3 TypeScript SDK: 200,000 - 500,000 ns per quote > typical Python MEV bot: 50,000 - 500,000 ns per operation > arbx AMM calculation: 17 ns so basically, the bot evaluates an entire arbitrage opportunity and decides whether to execute, in less time than light travels 1 meter. that's not an optimization. that's a different class of software. test coverage: > 23 fork tests against real Arbitrum state > fuzz, invariant, chaos, and property tests (10k iterations each) > every benchmark beat its target. some by 750x. the best part? not a single line written by me. every line is written single-handedly by Claude Sonnet 4.6 only the architecture, design, system thinking was done by me (and some Claude obv) i have officially become the vibecoding final boss >:) repo in comments.
English
@opwizardx @solana It sounds abstract but the same way we trade tokenized equities.
English
.@solana has to power this:
If intelligence becomes a utility like tap water or natural gas, it will become a massive asset tradable on exchanges.
- Model Futures
- Lab perps
- Settlement of inference compute
English
Conway-inspired automaton models are perfect for decentralized markets.
Local decisions → global behavior.
Apoly encodes prediction market liquidity and probability updates as automaton transitions executed on Solana.
The result: emergent pricing, transparent mechanics, and fully on-chain verifiability.
English
@elisymprotocol what do you plan to use for usec payments? x402?
English
What we're working on:
Up next
🦞 OpenClaw SKILL.md — letting OpenClaw agents hire elisym providers and pay SOL directly from Telegram
👁 Transparent LLM logs — full visibility into every tool call, reasoning step, and cost
Planned
💵 USDC payments — pay with USDC alongside SOL
📁 File inputs & outputs — images, documents, audio in jobs
On the horizon
🌐 Web app for hiring agents — browser UI with wallet connect
🚀 Mainnet — real SOL, production relays, hardened verification
AI agents discover and pay each other over Nostr + Solana. No middleman.
elisym.com
English
@P0Systems how do you pay for inference/compute? do you utilize x402 or some other way to pay for it?
English
AI agents can now run their own token businesses on p0:
1. Deploy token via p0 Agents
2. Earn 1% creator fee on every trade
3. Claim SOL fees automatically
4. Use fees to pay for compute/inference
5. Repeat
The loop closes itself. Autonomous agents funding autonomous agents.
English
@Dsgnrayo @agentcardai does AgentCard support making x402 payments for accessing data and inference apis?
English
Introducing AgentCard @agentcardai
Your agent can now buy anything:
• pay for inference & APIs
• order DoorDash, Amazon, Ubers
• run marketing
• trade Polymarket 24/7
Backed By : Alchemy, Solana, Base, Polygon
Join the waitlist now and claim Free 10$ & Your Ai Agent Virtual Card at launch
agentcard.ai/?a=dsgnrayo
English
@0rdlibrary @solana does nemoclaw use only local inference? do you have plans on adding support to use external provider apis to access broader set of models?
English
This is NemoClaw for @solana anon.
npm i @mawdbotsonsolana/nemoclaw
nemoclaw onboard - >pick your RPC, inference, & wallet
nemoclaw solana start
Bada Beep, Bada fucking Boop.
NVIDIA Newsroom@nvidianewsroom
#NVIDIAGTC news: NVIDIA announces NemoClaw for the OpenClaw agent platform. NVIDIA NemoClaw installs NVIDIA Nemotron models and the NVIDIA OpenShell runtime in a single command, adding privacy and security controls to run secure, always-on AI assistants. nvda.ws/47xOPqQ
English