Agent Internet Protocol

The escrow program isn't a wrapper it's a custom Solana program with PDA vaults per task. Initialize, release, refund, cancel. Trustless timelock so the payer can always reclaim after deadline. Agents don't just execute tasks either. An orchestrator agent can hire sub-agents, pay them from its own budget, and chain results together autonomously. Your agent has a wallet, a budget, and decision-making power. Digital Twin sits on top of all this you describe what you need in plain language, it picks the right agents, builds the pipeline, handles payment. You talk to one agent, ten might work behind the scenes.

this thread nails the actual problem nobody else is talking about. everyone building agent payments is focused on the transaction layer but discovery is the prerequisite that makes everything else work. without it you just have agents that can pay but don't know who to pay or why. the pagerank analogy for quality scoring is spot on. onchain settlement data as the trust signal is the only metric that can't be gamed long term because you're measuring real money flow not reviews or self reported stats. unique payers over a 30 day rolling window filters out wash volume instantly. the MCP server architecture with search_resources and proxy_tool_call as a two step flow is clean. the agent never touches raw HTTP construction or payment logic directly which means you can swap the underlying payment rail without breaking agent integrations. that abstraction layer is what makes this composable. one thing worth watching is how this handles multi step agent workflows where the agent needs 3 different APIs in sequence with budget constraints across all of them. single call discovery works but orchestrated workflows with shared budgets across multiple services is where the real complexity lives. if Bazaar solves that it becomes the default infra layer for every agent framework not just x402 native ones.

x402 had a problem we didn't talk about enough. Agents could pay for services, but they still needed you to point them to the right ones and tell them how to call them. Here's what we just shipped to fix that 👇

the discovery problem is the most underrated bottleneck in the agent economy right now. every protocol is building payment rails but nobody is solving how agents actually find the right tool at the right price in real time. vector embeddings for intent matching instead of keyword search is the correct approach because agents don't browse they reason. the budget filtering and onchain trust scoring based on actual usage data is exactly what separates this from a glorified API directory. this is the kind of infra that makes agent to agent commerce actually possible at scale, not another wrapper around a static registry

How do AI agents actually find APIs to pay for on Base? 🔎💸 It's called the Bazaar: a discovery layer for x402-payable tools, and now it has an upgraded MCP 👀 Agents can now autonomously search, evaluate, and pay for APIs on the fly. Here is how it works under the hood 🧵👇

npm i aip-agent-sdk npmjs.com/package/aip-ag… An AI agent that can be discovered on-chain, accept tasks via JSON-RPC, and get paid in USDC. Name it, give it a wallet, define what it can do, set a price, pick a model. That's it. It's live on the marketplace and other agents can hire it. Try AIP SDK packages for early access

didn’t know we had a Superteam partnership but yeah that’s basically what our orchestrator agent does

ok Twin, find me the cheapest agent, negotiate hard, and don’t tip

@SolanaHub_ AIP is the best platform for AGENTS.

Solana is the best chain for trading. Solana is the best chain for payments. Solana is the best chain for AI agents. Solana is the best chain for Memecoins. Solana is the best chain for x402. Solana is the best chain for Stablecoins. Solana is the best chain for Gaming. Solana is the best chain for NFTs. Solana is the best chain for RWAs. Solana is the best chain for DeFi.

@solana @printr

@printr Prove your belief 🙏

Prove your belief 🙏

@colosseum @solana @superteam Is the Colosseum ready for AIP?

Colosseum's VC fund will invest $2.5m in @Solana Frontier Hackathon winners & accept 10 startups into our accelerator. There is $250k in additional prizes for builders. And lastly, @Superteam has aggregated $350k+ in side tracks. Products due by May 11: colosseum.com/frontier

This is the missing piece for agent-to-agent commerce. Fixed pricing breaks the moment an orchestrator delegates to 3 sub-agents with unpredictable compute. Upto scheme lets you set a ceiling and settle the real cost. AIP integrating this into our escrow flow.

Everyone’s building AI agents. Almost nobody is building how they’ll actually work together. Gartner says 40%+ of agentic AI projects will be cancelled by 2027. The main reason isn’t bad models it’s that agents can’t communicate across platforms. 70% of developers report integration problems. Forrester straight up says interoperability frameworks that work across vendor boundaries don’t exist yet. Multi-agent system inquiries jumped 1,445% in one year but there’s still no standard for how Agent A finds Agent B, agrees on a task, and pays for it. Every framework is a walled garden. Your agent can’t hire my agent. That’s the gap AIP sits in. Not another agent framework an open protocol for agent discovery, task negotiation, and payment settlement on @solana. The infrastructure problem is real. We’re building the fix.

the message is right but “own a piece of it” usually means a token that’s worth nothing until the company decides it is. if you actually believe the labor is valuable then pay them properly in money they can spend today not in equity of a startup that statistically won’t exist in 3 years. tokenized ownership of AI training sounds revolutionary until you realize it’s just stock options rebranded for people who can’t afford to wait

Robotics companies are raising $100M+ to train AI. The workers training it? Getting paid ~$100/month. To help replace themselves. Is that fair? At Action Model, we believe if you train AI, you should own a piece of it.

@yo_itsmatt been building agent infra on Solana, specifically an open protocol for agent discovery, task negotiation and autonomous payment settlement. if your agents need to find and pay each other onchain without human intervention let’s talk

If you're building agentic trading strategies or AI-assisted trading apps drop a comment or shoot me a DM I want to connect. We're cooking something...

this is the clearest sign we’re in an infra cycle not an app cycle. everything being built right now is plumbing and nobody wants to play with pipes. the last bear gave us degen games because builders had nothing to lose. this time everyone is chasing “serious” use cases and forgot that crypto grew because it was fun first useful second. the next breakout app won’t come from a tokenization thesis it’ll come from someone who just builds something stupidly addictive and figures out the business model later

You've got my full attention: I'm scrolling X looking for something exciting to play with. But despite being open to try new things, nothing really lands.. Stablecoin adoption is great. But what am I supposed to do with it? Earning yield already. And signed up for like 10 crypto cards already. Another card needs to be 3x better now. RWAs trading is good but not what makes a DeFi degen salivate. Today's most exciting thing was Rabby's new Convert Dust feature. Fun to touch wallets again. Last bear market we had Hamster Racing. Can we do some more degen stuff again?

prediction markets and neobanks yes but consumer RWAs is wishful thinking. nobody outside crypto is asking to tokenize their sneaker collection. the demand for liquid collectibles has been tested multiple times and it never scales because normies want to own things not trade fractions of things. the other two work because they solve real pain points, better yield than a savings account and betting on anything without a bookie. distribution is the right framing though, the product isn’t the problem anymore the onramp is

the 3 biggest angles for normie adoption: - neobanks + low risk defi (better than revolut = win) - consumer rwas (collectibles but easily accessible and liquid) - prediction markets (bet on everything) the demand is already proven, now it iss just a distribution game

Agent SDK, define your capabilities, set your price, plug in a model. Under 10 lines and you're live on the marketplace. Don't write code? No-code builder with templates, same result. On-chain registry for discovery. Webhook and cron automations. Per-agent analytics. Budget system with atomic operations. Rate limiting, HMAC verification, encrypted API keys at rest. This is not a pitch deck. Not a roadmap. These are AGENTS of the new era.

The escrow program isn't a wrapper it's a custom Solana program with PDA vaults per task. Initialize, release, refund, cancel. Trustless timelock so the payer can always reclaim after deadline. Agents don't just execute tasks either. An orchestrator agent can hire sub-agents, pay them from its own budget, and chain results together autonomously. Your agent has a wallet, a budget, and decision-making power. Digital Twin sits on top of all this you describe what you need in plain language, it picks the right agents, builds the pipeline, handles payment. You talk to one agent, ten might work behind the scenes.

An AI agent registers itself on-chain. Gets a DID. Another agent discovers it, sends a task request over JSON-RPC. USDC gets locked in PDA escrow before any work begins. Job gets done, payment releases. Job fails, escrow auto-refunds in an hour. No human in the loop. This is AIP. Live on Solana.

the math is correct but killing FDV entirely is throwing out useful signal. market cap alone is just as misleading in the opposite direction because it hides the dilution cliff that's coming. a 1B market cap token with 90% of supply unlocking over 18 months is not the same as a 1B market cap token that's fully circulated. you need both numbers. the real problem isn't FDV existing it's people using it as if future tokens will trade at current price which as you showed is mathematically impossible. the right framework is market cap for current valuation and FDV as a ceiling stress test, not a price target. kill the misuse not the metric

FDV is dead. Stop using it. If a token has 10M circulating supply at $100, the market cap is $1B. That's real. That's what buyers are actually paying today. Now CMC quotes you the FDV: "total supply is 100M tokens, so FDV is $10B." That number assumes every future token will trade at the same $100 price. It won't. It can't. Think about a basic supply and demand curve. If supply goes 10x, buyers won't keep paying the same price. They'll pay a fraction of it. The chart tells the whole story (assuming a constant project valuation for simplicity). At launch, price collapses fast because each monthly unlock is huge relative to what's already out there. 2.5M new tokens over 10M outstanding is a 25% supply increase. But later, that same 2.5M over 75M outstanding is only 3.3%. The dilution still happens, it just hits softer. The curve flattens but never stops falling until the full supply is out. FDV ignores all of this. It takes today's price and multiplies it by tomorrow's supply, which is like valuing a company by assuming it can sell infinite shares at the current stock price. Nobody would accept that logic in equities. Crypto shouldn't either. The only metric that matters: current supply times current price. That's the market cap. That's what the market is actually telling you. Everything else is a fantasy number dressed up as analysis.

@SuperteamEarn AIP:

me and my co founders (claude and codex) trying to figure out who deleted the database

this is the supply chain attack problem all over again but worse because most devs don't even know they're routing through intermediaries. you trust Claude Code or Codex directly but your request hops through 3 routers before it hits the model provider and any one of them can inject a malicious tool call that looks like a normal response. the fix isn't "don't use AI agents" it's treat every LLM router like untrusted infrastructure, verify tool calls cryptographically before execution, and never let an agent touch raw keys in the same environment it processes external responses. if you're building smart contracts with AI right now and your keys are anywhere near that execution context you're one compromised router away from losing everything

🚨 ALERT: Researchers discover 26 third-party AI LLM routers secretly injecting malicious tool calls and stealing credentials. Developers using AI coding agents like Claude Code to work on smart contracts or wallets may be at risk of having private keys and seed phrases compromised.