TL;DR
We at Celer are excited to launch AgentPay, a state-channel powered network that allows AI agents to transact at machine speed—delivering millisecond settlement, private payment flows, and sub‑cent effective costs. AgentPay removes blockchain bottlenecks—block‑level latency, $0.01+ per‑transaction fees that translate to 10%–50% on penny‑scale payments, and full transparency that leaks strategy—while staying securely anchored to chains for opening, closing, and disputes. With always‑online agents and ZK coprocessors resolving prior state‑channel difficulties, AgentPay unlocks scalable, cross‑layer payments for inference, data, and robotics, and is advancing interoperability with standards like X402. Celer AgentPay is fully open-sourced (smart contract, off-chain node implementation) and will be deployed on mainnet in the coming weeks with our launch.
A Machine Economy in the Making
Imagine a digital bazaar alive with thousands of autonomous agents—algorithms that don’t just process information but make decisions, strike deals, and settle payments in real time. These agents could be running on decentralized networks, powering trading platforms, managing data markets, or even orchestrating fleets of robots.
This vision is moving from fiction to real deployment. Enterprises and cloud providers are piloting agentic AI that calls APIs, rents compute, and buys data without a human in the loop. Amazon Web Services recently stood up a dedicated agentic AI division to develop independent digital workers. Capgemini projects that such systems could generate “more than $450 billion in business value” by 2028, if they can be trusted to act and transact on their own.
But there’s a catch: this emerging, autonomous agentic economy faces a “payment problem”, and even though people praise cryptocurrency and stablecoins as the AI-native programmable money, today’s blockchains still lack support for this future. Agents need to pay for access to models, compute time, or datasets in milliseconds, and they need to keep those transactions confidential. Public ledgers, with their high gas fees and total transparency, make that nearly impossible.
The Challenge No One Talks About
In theory, blockchains are perfect for autonomous agents: a transparent, verifiable backbone where no middleman is needed. In practice, they move like molasses. Every on-chain transaction must be globally broadcast, validated, and permanently stored—a process that can take seconds to minutes.
For human users, a short delay is tolerable. For AI agents executing high-frequency transactions—say, one query per millisecond across dozens of partners—it’s fatal. High fees, latency, and total visibility means that it leaks valuable signals like pricing or demand levels, undermining competitiveness.
Gartner recently warned that over 40% of corporate agentic AI projects might be scrapped by 2027, citing unclear business models and infrastructure bottlenecks.
More importantly, on-chain costs remain too high. Even on the cheapest L2s, per-transaction fees are impractical for AI streaming micropayments. Sending stablecoins typically costs about $0.01; for transfers of only a few cents, that’s an effective 10%-50% processing fee. If agents pay directly on-chain, those fees become a built-in tax on interactions, constraining viable use cases and slowing adoption.
Privacy can also be essential in AI‑to‑AI payments. Agents may price differently for different counterparties, hide sourcing and arbitrage strategies, or protect demand signals such as volume and timing. Public ledgers expose counterparties, amounts, and flow patterns, enabling front‑running and reverse‑engineering of pricing models. Payment rails must provide confidentiality to non‑participants while preserving verifiability and selective disclosure for audits and compliance.
We need an AI Agent Payment Network that:
- is securely rooted in blockchains
- processes payments in real time, without block‑level delays
- preserves privacy for most payment flows
- minimizes costs without smart‑contract gas overhead
Celer AgentPay is the Solution
We are proud to unveil the Celer Agentic Payment Network (AgentPay) — a state‑channel‑powered network designed to become the financial backbone for the rapidly emerging machine economy. With AgentPay, AI agents can now exchange value as low as a fraction of cents in milliseconds, pay for data, computation, or access to models, and settle it all privately without the latency, exposure, or costs of traditional blockchain transactions. AgentPay is where automation finally meets real‑time economic capability.
As one of the earliest layer‑2 scaling techniques developed for Ethereum, state channels function like private side roads branching off a main blockchain highway. They let participants transact directly with one another off‑chain, achieving near‑instant settlement and zero gas fees, and only post the final, aggregated result once interactions are complete. Every off‑chain transaction is cryptographically signed by all participating parties, ensuring that no one can alter or dispute the agreed state later. If a disagreement occurs, any party can submit the latest signed state to the blockchain for on‑chain verification and resolution. This design maintains the integrity and validity of off‑chain transactions while keeping details private, protecting sensitive information such as transaction histories and pricing strategies, and providing the speed AI agents need to coordinate in milliseconds rather than minutes.
Celer is a pioneer in developing state‑channel scaling solutions. In 2019, we launched the world’s first and only in-production generalized state channel network built on Ethereum. This network features full-fledged real-time streaming payment functionality, and is capable of supporting complex logic using conditional payments.
However, state channel networks faced adoption barriers back then due to two main issues:
- An always-online requirement for parties sending and receiving payments.
- Conditional payment resolution depended on on-chain smart contract modification.
Now, in the new context of AI agent payments and with advances in ZK proof systems, both of these constraints are removed. For AI Agents, they are so obviously always-online and can easily sign messages for payment processing. With solutions such as ZK coprocessors, on-chain historical states can be accessed easily and trustlessly, without implementing additional on-chain logic.
A Deeper Dive into AgentPay

AgentPay evolved from Celer’s battle‑tested State Channel architecture, and introduces a suite of new features tailored for the real‑time AI economy:
Instant Payment Settlement
AgentPay enables AI agents to settle payments in milliseconds by moving transactions off‑chain through Celer’s Generalized State Channels. Once a channel is opened, agents exchange cryptographically signed updates directly, bypassing the blockchain for everyday transfers. Payments flow instantly in both directions using dual-simplex channels, while a sliding‑window design allows multiple updates to occur simultaneously for continuous, high‑throughput communication. Through Hash Time‑Locked Contracts (HTLCs), AgentPay also extends instant settlement beyond direct peers—enabling secure, atomic, and trustless multi‑hop payments across the network. The blockchain is only invoked for channel openings, closures, or disputes, minimizing delay, cost, and congestion.

Cost Efficiency
By aggregating transactions off‑chain, AgentPay removes gas fees during active operations and incurs only minimal on‑chain costs when channels close or settle. HTLC‑based multi‑hop routing allows value to move efficiently across indirect connections without requiring new on‑chain links, further reducing operational overhead while retaining cryptographic guarantees of payment completion or refund.

Scalability and Privacy
AgentPay supports millions of off‑chain transactions per second while keeping the blockchain footprint extremely light. As only the final states of the off-chain transactions are posted on-chain, this privacy‑preserving execution ensures detailed payment data and pricing remain visible only to channel participants. Built on Generalized State Channels capable of executing conditional payments and other complex off‑chain logic, AgentPay couples massive scalability with strong confidentiality—ideal for high‑frequency, AI‑to‑AI economic coordination.

Cross‑Agent Composability
Serving as a universal settlement layer, AgentPay enables AI agents, applications, and ecosystems to interoperate seamlessly across blockchains and execution layers. Through multi‑hop routing, cross‑layer liquidity bridging, agents can transact and coordinate globally without sacrificing reliability, atomicity, or trust.
You can check the AgentPay repo for the contract code and off-chain modules, and read our docs to get started.
The Road Ahead: Use Cases and X402
Serving as the connective layer of the AI economy, AgentPay is invisible, instantaneous, and nearly free. In practice, it’s easy to picture: an AI that labels images pays another for GPU power, which in turn pays a third agent for model inference, with each transaction settled privately over Celer’s mesh network in milliseconds.
The implications stretch from robotics to cloud intelligence. A fleet of autonomous drones could pay ground sensors for real‑time weather data. A decentralized API marketplace might charge users per inference without revealing traffic patterns. Even everyday digital assistants could stream payments for access to premium models as they operate—every microtransaction logged securely in aggregate, never publicly.
This kind of seamless value exchange forms the substrate of the next era of automation, where agents don’t just think but also trade.
Agentic AI is hurtling toward the mainstream, and right now most of its infrastructure still assumes a world of humans pushing buttons and paying manually. Celer’s Agent Payment Network offers a fix—a way for machines to transact autonomously at their operating speed. We will actively work with industry partners to roll out some exciting use cases that align with industry standards such as X402 in the coming months.
With AgentPay, we are laying the groundwork for a true economy of intelligent agents: autonomous systems that earn, pay, and collaborate seamlessly through billions of instantaneous transactions every second.