Enterprises don’t just need agents that perform. They need agents they can shape, govern and improve as their business evolves.
That’s the idea behind our work with LangChain.
Read more: https://t.co/3hu5eARj5N
V2 of the Sei Giga Whitepaper is live.
Sei Labs can now confirm that Giga will have substantially better performance and an improved featureset over what was initially described.
Giga will have:
- Sub-250ms finality
- Pre-execution privacy
- Protocol-level MEV resistance.
Here's what's new and why it matters:
Finality: sub-250ms, down from 400ms.
Autobahn's pipelined two-phase BFT is designed to reach consensus in 1.5 round trips versus Tendermint's 3.
In the design, ordering finality is decoupled from state attestation: every trade will settle at consensus speed, while execution happens asynchronously so heavy computation never delays block production.
State commitments will use lattice hash divergence digests over write logs, eliminating Merkle tree computation from the hot path.
The biggest addition in this new version is Sedna.
On single-leader chains, the block producer controls ordering. On multi-proposer chains, any proposer that sees your full transaction can duplicate it and race you in the merge order. The whitepaper identifies three MCP-specific MEV channels.
Sedna is designed to address all of them. Senders would encode transactions into rateless fragments distributed across proposer lanes, and no proposer would see the full payload until the finalization threshold is crossed.
Pre-execution privacy without encrypted mempools, threshold decryption or adding latency to the critical path.
Under Giga, transaction ordering will be fully deterministic.
The tip priority merge rule orders lanes by descending maximum tip, preserves intra-lane position, and deduplicates by hash. The output is a pure function of the committed cut.
Every correct executor will derive the same ordered transaction sequence from the same finalized data.
Storage and execution were also substantially upgraded in v2.
Storage: flat LSM-based key-value store will replace Merkle Patricia Tries. Tiered hot/warm/cold architecture. Divergence commitments over write logs introduced.
Execution: Block-STM parallel execution with optimistic concurrency control. Pipelined parsing, address recovery, and signature verification.
The paper also clearly defines:
- Production Sedna with PIVOT-K incentive layer
- Autobahn consensus upgrades
Forthcoming research:
- Full transaction fee mechanism
- Full-scale PQC alternatives beyond ML-DSA
- Giga tokenomics