Starknet has made two significant moves this week, each reinforcing its push toward practical, user-controlled infrastructure. On July 15, the network officially opened its privacy stack to external developers with the launch of the Privacy SDK and the Privacy Wallet API. A day earlier, a community draft proposed a user-owned memory protocol for AI agents, putting the concept of data sovereignty for artificial intelligence on the crypto agenda.
The privacy launch is built on STRK20, a layer already running on mainnet that allows any ERC-20 token to exist with encrypted balances and private transfers. Transactions are shielded through ZK-STARK proofs that confirm validity without exposing sender, recipient, or amount. Only the deposit and withdrawal steps remain publicly visible. Starknet frames this not as an ideological stance but as practical functionality—most past privacy products failed because they were slow, expensive, or segregated from existing liquidity. STRK20 integrates directly into the assets, wallets, and applications users already rely on.
Developers have two access routes: the Privacy SDK (Apache 2.0 license), designed for wallets and advanced integrators that need direct control over viewing keys and proof generation; and the Privacy Wallet API, which delegates proof and note management to the user’s wallet via starknet.js, keeping viewing keys out of the dapp’s reach. For complex DeFi flows, anonymizer contracts enable atomic private operations—swaps, loans, or staking—with full reversion if any step fails. Projects like avnu, Troves, ForgeYields, Ekubo, and Endur are already building on this infrastructure.
Meanwhile, the memory protocol draft proposes a system where AI agents operate with scoped, temporary, and auditable access to user data, enforced by capability tokens. It is still an early design, but it reflects a growing push to give users more control over how AI agents interact with their information. The proposal opens a conversation about whether blockchain networks can enforce rules that make agent-driven systems safer and more transparent.
Both developments underscore Starknet’s evolution from speculative hype toward operational utility. The privacy stack adds immediate, concrete tools for confidential transactions, while the memory draft introduces a research avenue that could shape the next generation of on-chain AI. Together, they give builders and users a clearer picture of what the network is capable of—and what it may become.
