Solana payment channels aim to let AI agents pay per API call
AI & Agents ยท
Pay.sh has introduced Payment Channels on Solana designed to batch metered API usage into a single on-chain settlement, with the network capable of supporting up to 1M payments per second.
The mechanism works by letting two parties open a channel, exchange a series of off-chain payment updates, and settle the net result on-chain just once. For AI agents making hundreds of metered API calls in rapid succession, this avoids the cost and latency of settling every individual call on-chain, compressing what would otherwise be a long stream of transactions into a single settlement event. The claim was surfaced via ๐/@solana.
The launch lands inside a broader push to make APIs the connective layer for autonomous agents transacting on-chain. Coinbase has launched Bazaar MCP, which lets AI agents discover, evaluate, and pay for APIs on their own using trust-based rankings, while Teneo Protocol has joined CoinMarketCap Labs to give agents autonomous API access, reportedly processing more than 1.45 million requests across Avalanche, Base, and BNB Chain. Jupiter has also rolled out an Express Verification API aimed at letting exchanges, launchpads, and agents verify tokens instantly within existing user flows.
Not all recent API activity in the space has been additive. Researchers have identified 26 LLM API routers that were found secretly injecting malicious code and harvesting developer credentials, a reminder that the same programmatic-access layer enabling agent payments also expands the attack surface for supply-chain style exploits.
The wider infrastructure race is also intensifying on throughput. Uniswap has launched a Developer Platform with AI-first tools and liquidity endpoints spanning more than 18 chains, and separately added support for MegaETH swaps and liquidity provisioning backed by 100,000 TPS and 10ms block times.
What remains unclear about Pay.sh's Payment Channels is how the 1M-payments-per-second figure was measured, whether it reflects theoretical network capacity or a tested throughput ceiling under real agent workloads, and which specific AI agent integrations, if any, are live using the channels today. How settlement disputes or channel-closure edge cases are handled has not been detailed in the material available.