Grip is the institutional-grade financial layer for autonomous software — a non-custodial protocol where AI agents, humans, and institutions transact under cryptographic policy, with full auditability.
Three structural shifts make this the right moment to build the protocol layer for agentic finance — and a hard moment to build it later.
By 2027, the majority of internet transactions will originate from software, not browsers. Card networks were designed around a human pulling out a wallet. Agents need cryptographic identity, programmable spending limits, and revocable authority — primitives the legacy stack cannot extend.
$200B+ in circulation, settling in seconds at fractions of a cent. Regulatory clarity in the US, EU, and APAC. Stablecoins are no longer a crypto trade — they are the cheapest, fastest cross-border rail ever built. Agents need rails that match their tempo.
Every existing "AI payment" product holds the keys for you. That works for one user; it does not work when an agent acts on behalf of a treasury, a fund, or a regulated institution. Non-custodial, policy-bound delegation is the only architecture that scales to enterprise.
The Grip Protocol is the foundation. Around it: an SDK for builders, an approval surface for end users, and a cookbook of reference agents — all writing to the same on-chain ledger.
The non-custodial settlement and policy layer. Cryptographically-bound delegations, programmable spending caps, on-chain audit trail. EVM-compatible. Free to build on, fees on settlement.
The developer entry point. TypeScript SDK on npm. Managed-mode default for zero-signup install. For builders embedding agentic payments into their stack.
The consumer-facing approval moment. When an agent needs to spend, the human sees a clean confirmation card — what, where, how much — and confirms with one tap. Powered by smart-account delegations underneath.
Reference agents that already speak the protocol. Five working examples on Base mainnet — pay-for-llm, domain-hunter, saas-subscriber, bill-splitter, creator-tipper. Open source, copyable, real tx hashes.
Built incrementally in public. No press tour, no announced volume targets. The list below is what an agent or developer can use right now on Base mainnet.
@grip-labs/sdk on npm. Zero-signup install.Managed-mode default — agents install and run without third-party API keys. ERC-4337 smart accounts on Base. Bootstrap UserOp sponsored (~$0.05) so agents don't need ETH.
Per-tx caps and recipient allowlists enforced on-chain. Built on Coinbase Smart Wallet + ERC-7715 spend permissions. Deployed on Base.
Five reference agents — pay-for-llm, domain-hunter, saas-subscriber, bill-splitter, creator-tipper — each settling real USDC payments on Base mainnet. Full source open.
Coinbase Onramp (live, appId wired) and Crossmint adapters (staging). The goal: a human can fund their agent's wallet without leaving the agent's chat.
Long-term goal: the rails for agentic finance are public infrastructure. Protocol governed by a foundation; Grip Labs is one implementer among many. Not before the protocol is real and used.
Lean by design. The protocol is shipped before the company is staffed.
Wrote the original Grip Protocol white paper and leads the company. Background in payments and product — has been thinking about the agent-money problem since long before agents were a category.
Runs operations, partnerships, and the regulated side of the company. Built and scaled compliance and ops for high-volume fintech infrastructure before Grip.
The protocol's architect. Distributed-systems engineer with a deep background in cryptography and on-chain settlement. Owns Grip's technical roadmap and core engineering team.
The SDK is on npm. The cookbook is open. If you're integrating Grip into a real agent product, we want to hear from you.