Walrus just became the default memory layer for elizaOS v2 which is exactly the kind of “real infra meets real agents” move I want to see. elizaOS plugs @WalrusProtocol into onboarding (with time-limited credits) so new devs get persistent, verifiable memory from day one. I read the integration post and sanity-checked the primitives: every upload resolves to an on-chain Proof of Availability on Sui a signed certificate that your blob is live and under custody; this is the anchor that turns “agent memory” into a programmable asset instead of a best-effort cache. For sensitive flows, Seal adds threshold encryption + on-chain access control so agents can share memory without spraying plaintext. ❯ For ops: you can even track announcements in Walruscan’s news feed while you wire up your own telemetry. ❯ Why this matters: agent swarms need long-lived, verifiable context; PoA receipts + programmable storage make “who saw what, when” auditable across workflows. ❯ Bonus: this is chain-agnostic at the app layer build on Solana/EVM, keep memory on Walrus, settle proofs on Sui. Dev note: ship a small dataset via the CLI, fetch the BlobCertified event from the Sui tx, then have two agents read/append via the TS SDK one public, one Seal-gated and compare latencies + auth paths in logs. (Post your numbers; I’m collecting a bench of read/append timings.) Net: the agent stack finally has a memory primitive you can verify, price, and compose. Now let’s see who turns those PoA receipts into real AI workflows