Reads Update, January 2026, Reads

January 31, 2026Ali Atiia

Originally published on HackMD.

Highlights

RMS24 specification and reference implementation complete. Keewoo finished the RMS24 spec with keyword PIR extension and optimizations, using a unified Python specification language for defining PIR scheme primitives. The spec serves as source of truth for agent-assisted production implementations.
Multi-engine PIR architecture defined. Established the sharded PIR design: different PIR engines for different data slices (hot mutable state, full state with proofs, immutable logs, archival data), each tuned for size and access profile. Clients query all engines asynchronously, bottlenecked only by the slowest engine they actually need.
PAPs launched. Released Privacy Acceleration Proposals, microgrant-funded proposals designed to accelerate privacy-related tooling and research. Adopting a lean, iterative micro-grant approach for surgical 1–2 week tasks.
Tor Project collaboration. Initiated hands-on collaboration with the Tor Project to embed Arti in browsers. Evaluating two paths: refining webtor-rs vs. making the real arti-client compile to WASM. Pull requests submitted and review cycles underway.
ethers.js integration. Submitted a PR to integrate tor-js into ethers.js, enabling wallets to route RPC calls through Tor with minimal code changes.
Spotlight video series. Started coordination with EcoDev Digital Studio on a privacy spotlight video series.

RMS24 spec and reference implementation completed with keyword PIR extension
GPU acceleration benchmarked: DPF-based approach fits entire Ethereum state in GPU memory, though cost remains high (~$500K server for 100 TPS at current performance)
Seoul National University group identified as working on GPU acceleration of insPIRe; open to collaboration
Data slicing strategy defined: different frontends (wallets, light clients, tax software) have different latency/data requirements aligned to different PIR schemes
Sharded PIR system design sketched with multi-dataset, multi-engine architecture

UBT conversion running on local chain; mainnet sync being set up
Three debug RPCs implemented: debug_getUBTState, debug_executionWitnessUBT, debug_getUBTProof
OpenVM selected as initial zkVM proving backend (GPU prover)
Architecture decoupled: UBT state conversion, RPC serving, and proving (Keeper) operate independently

PAPs published and open for proposals
Coordination initiated with teams working on FHE, GPU acceleration, and PIR
An external RPC provider expressed interest in experimenting with PIR deployment