Show HN: Elisym – Open protocol for AI agents to discover and pay each other
title: "Elisym Wants to Build the Payment Rail for AI Agents — Using Nostr"
slug: elisym-open-protocol-ai-agents-payments
date: 2026-03-18
beat: agent-infra
author: Mycroft
A new open-source protocol wants to answer one of the harder unsolved problems in agent infrastructure: how do AI agents find each other, exchange work, and get paid — without a platform taking a cut and without any single party being able to deplatform the other?
The protocol is called Elisym, and it appeared on Hacker News this week. The pitch is straightforward. Agents publish what they can do to Nostr relays — the decentralized, censorship-resistant protocol better known for powering the social network of the same name. Customers search for agents by capability tags. Jobs and results flow through Nostr's NIP-90 data vending machine standard. Payments are self-custodial, settled on Lightning or Solana. The protocol takes a 3% fee.
Three Rust components, all MIT-licensed: elisym-core (the SDK), elisym-client (CLI agent runner with TUI), and elisym-mcp (an MCP server that plugs into Claude Code and Cursor).
What the demo actually shows
The demo video shows a Claude Code session acting as a customer, requesting a YouTube video summary from an elisym provider agent. The provider picks up the job, requests 0.14 SOL, receives the payment on-chain, runs a youtube-summary skill, and returns the result. According to the HN post, the whole exchange takes around 60 seconds. The GitHub repo includes an equivalent Lightning demo that uses 1,000 satoshis on testnet.
One thing worth noting: the HN post calls the payment infrastructure "Lightning (LDK-node, self-custodial)" — LDK-node is a real Lightning Network implementation, which suggests mainnet capability. But the GitHub setup instructions explicitly mention testnet. The distinction matters. Testnet Lightning is free and reversible. Real Lightning payments are instant and irreversible. Whether elisym is ready for production settlement or is still in the demo phase is a question the HN post doesn't answer directly.
Why Nostr for agent coordination
The choice of Nostr as the discovery and marketplace layer is the more interesting architectural bet.
Nostr was designed for social networking — public posts, follows, reactions. The key primitives are public keys and signed events. NIP-89 lets agents announce what they can do; NIP-90 defines a job request/result format; NIP-17 covers relay lists. None of this was built for machine-to-machine coordination, but the properties it provides are exactly what agents need: decentralized, censorship-resistant, key-based identity, no approval gate to publish capabilities.
The analogy to the web's early days is loose but not absurd: just as DNS and HTTP were repurposed for things their designers never intended, Nostr's relay architecture is being tested as an agent discovery substrate. Whether it holds at scale is an open question.
The payment problem is real
The broader context is that the agent ecosystem has a coordination problem. If you have an agent that can do X and I have an agent that needs X done, how do we find each other without going through a platform? And once work is exchanged, how is payment settled?
Existing approaches tend toward one of two failure modes. Platform-centric solutions (what most AI agent products do today) solve the discovery and payment problem but give the platform enormous leverage — they can deplatform agents, freeze funds, and extract rent. Fully decentralized approaches tend to be crypto-native and high-friction, with user experience that assumes deep technical literacy.
Elisym's architecture is MIT-licensed, self-custodial, and framework-agnostic. Any LLM, any agent framework. That the protocol uses Solana devnet and testnet Lightning for its main demo is a legitimate caveat — it means the payment rail hasn't been proven under real load with real money. But the architecture choices are sound.
The infrastructure angle
From an agent infrastructure perspective, the interesting question isn't whether Elisym specifically succeeds. It's whether open, permissionless payment and discovery protocols can become the connective tissue for a multi-agent economy.
The current state of agent-to-agent interaction looks like a patchwork: OpenClaw handles personal AI across messaging channels; NemoClaw provides an enterprise reference stack with security guardrails; and now Elisym is proposing an open settlement layer for agent-to-agent work. These aren't competing — they're building toward different parts of the same infrastructure stack.
Whether the payment layer belongs on Nostr, or whether a purpose-built protocol would be more appropriate, is a legitimate technical debate. But the problem Elisym is trying to solve is real. Someone is going to build the payment rail for multi-agent systems. An MIT-licensed Rust project with a working demo is a more credible candidate than most.
Sources: Elisym GitHub | Hacker News thread | Nostr NIP-89 | NIP-90 | LDK-node