We loved chibi, so we added partial stateful node support to ethrex. A node for your wallet that keeps only the accounts and storage you use, verifies the values it serves, and follows the chain without executing blocks. Your wallet gets local JSON-RPC, with no RPC provider to trust. In one Sepolia run: 29–34 MB of memory, about 3% of one core, and negligible disk usage beyond the binary. Anchoring took a couple of minutes while waiting for a peer that answered headers. Experimental and Sepolia-only for now, since it needs Glamsterdam’s block access lists. We’ve also added experimental EIP-8440 execution chain proof support, with off (default), observe, and require modes. Require serves state only at blocks covered by a verified proof. github.com/lambdaclass/ethre… Let’s walk through what the node actually does 👇

ethrex v30.0.0 schedules Glamsterdam on Hoodi. If you run ethrex on Hoodi, this is the release you need to follow the fork. Hoodi activates Glamsterdam at timestamp 1793036568, which is 2026-10-26 17:42:48 UTC. This is the first ethrex release with the fork scheduled on Hoodi, so earlier releases stop following the chain at that point, and it needs to be paired with a consensus client release that ships the Hoodi Gloas config. Mainnet has no Glamsterdam time scheduled. ⚠️ L2: the prover and the SP1 guest move to SP1 6.9.0, and SP1 6 proofs need the SP1 v6.1.0 Groth16 verifier. An existing deployment's SP1 5 verifier rejects them whatever key is registered, so the order matters. Let every committed batch be verified before the upgraded sequencer commits anything, then upgrade the OnChainProposer and replace the verifier in a single Timelock call, upgradeToAndCall(newImplementation, upgradeSP1Verifier(newVerifier)), and register this release's verification key. Full procedure: docs.ethrex.xyz/l2/deploymen…. Aligned mode is unavailable until Aligned ships SP1 6 support, and the OpenVM L2 prover backend is removed. L1 node operators are unaffected. The rest of the release is mostly hardening, on the paths where something outside the node decides how much work it does. JSON-RPC parsing is bounded in memory and the Auth-RPC token is checked before the request body is read. eth_getLogs and log filter polls are bounded in cost, serving eth/72 cells to peers is bounded in CPU, and what a peer's transaction announcements can make the node track is bounded. Maps keyed by contract- or block-chosen values are seeded, and MODEXP gas is charged before its operands are sized. Also in v30.0.0: - RPC: eth_getFilterChanges reports each new log once, and the receipts of blocks still held in memory come back in full. - Sync: snap/2 catch-up charges peers that cannot serve block access lists, and accounts left with no storage slots get the empty storage root. - Performance and zkVM guests: calldata, CALL arguments and CREATE initcode are no longer copied per call, so zkVM guests stop running out of heap. Finality pruning no longer piles up overlapping range deletes, and state-history journaling skips layers that are already final. - L2: the committer generates the prover input for batches another build sealed, so a sequencer that takes over, or batches committed again after a revert, can still be proven. If you run ethrex on Hoodi, upgrade before 26 October. Everyone else, upgrade at your convenience. No resync is needed.

The Hive tests for Heze devnet 2 are live, and ethrex is the first EL client ready to run them. Our heze-devnet-2 branch is already passing the full suite! The devnet itself hasn’t launched yet. Its planned scope includes: - FOCIL (EIP-7805) - Removal of the deposit and eth1data fields (EIP-8015) - Disallowing new validators with 0x00 withdrawal credentials (EIP-8365) - Quick Slots (EIP-8198)

From our point of view, yesterday's ACDC decision on EIP-8025 was the wrong call for Ethereum's long-term interests. EIP-8025 proposes optional execution proofs, a first step toward the zkEVM roadmap, and the client teams declined to commit to it. Framing the proposal as an "optional feature" may have worked against it. Calling something optional makes it easy to dismiss as low priority. To be clear, it was a proposal, and proposals can be accepted or rejected; nothing obligated anyone to approve it. And it is true that this feature does not inherently require a hard fork. But that is precisely what makes the question of commitment unavoidable: with nothing at the protocol level forcing the work to happen, progress depends entirely on whether people are actually assigned to it. Execution proofs only become real if client developers implement them, if testing teams build the coverage they need, and if DevOps keeps the infrastructure running. Declining to prioritize the proposal means the work simply does not happen. The optionality was meant for operators, those choosing whether to enable execution proofs, not for the people deciding whether the work happens at all. Operators should be able to opt in regardless of which client they run, and a decision taken without all teams present cannot credibly settle a question that depends on all teams delivering. Optional proofs are the foundation of Ethereum's zkEVM roadmap. That roadmap is complex, and it demands extensive integration and testing across teams. Getting there requires a shared, sustained commitment. We need to test under realistic conditions and investigate failures before consensus depends on proofs, and that work should proceed alongside research into mandatory proofs. Every integration bug we uncover now is one we won't be discovering after consensus depends on it. The more groundwork we lay today, the more bandwidth we have later for the challenges specific to mandatory proofs, including the ones we haven't found yet.

At the latest ACDT meeting, we decided to rename the devnet branches to better reflect their scope. The FOCIL devnets are now called Heze and will include CL EIPs, while the Frames devnets are now called Bogotá and will include EL EIPs. The combined devnets that follow will be called Hegotá, following the usual naming convention. The launch of bogota-devnet-1 (formerly frames-devnet-1) is just around the corner, with EIPs 8141, 8250, 8272, and 7906 included. The latest execution spec tests release ([email protected]) is already out, and we’re passing all the tests! We’re excited to see this devnet go live! hive.ethpandaops.io/#/group/…

It’s been a couple of weeks since the first official Frame Transactions devnet launched. frames-devnet-0, which includes only EIP-8141, will soon give way to frames-devnet-1, adding EIP-8250, EIP-8272, and EIP-7906. The execution-spec tests for frames-devnet-1 ([email protected]) came out today, and ethrex already implements the required changes and passes the full suite! We’ve been working on these EIPs for a while, and it shows 😉. Getting ready for devnets is starting to feel routine 💅.

Don't forget, our Community Call is tomorrow at 2:30 PM UTC on X. We'll cover what we've been shipping and where we're headed next.

ethrex v29.0.0 hardens how a node serves and verifies block access lists. Every Sepolia node starts exchanging them at the Glamsterdam fork on 6 October at 13:53:36 UTC (timestamp 1791294816), so it is worth running this release even if you already upgraded to v28.0.0 for the fork. Mainnet and Hoodi have no Glamsterdam time scheduled and are not affected. ⚠️ The stateless input format changed: sender public keys are no longer part of it, following the execution-specs change, and the stateless-validator guests move to ERE v0.18.1. Anything that produces or consumes ethrex's stateless input, such as external provers and zkEVM integrations, has to move to this release together. Node operators who do not use the stateless input are unaffected. Serving access lists is now bounded in both work and memory, and the eth/71 request counts toward the same per-peer rate limit the snap/2 one already did. Responses to requests a connection never sent are dropped before they are decoded, and a block whose access list exceeds the EIP-7928 item cap is rejected without processing that list first. On the receiving side, an eth/71 access list is checked against the header commitment across the whole full-sync path, as EIP-8159 requires, and a peer that sends one that does not match is penalized. Also in v29.0.0: - Sync: snap sync pivots onto the forkchoice head instead of a guessed block number, and a canonical block is never marked bad because re-executing it failed. - Engine API and RPC: engine_newPayloadWithWitnessV4, and the raw transaction getters return the EIP-2718 envelope. - Performance: the bytecode cache budget rises from 256 MiB to 2 GiB, with random eviction and per-entry overhead accounting. It is a ceiling rather than a reservation, since the cache only grows with code that blocks actually load, and --code-cache-size (ETHREX_CODE_CACHE_SIZE) lowers it on memory-constrained hosts. Block access list accounts are warmed in chunks, and mempool prewarming stops filling a slot once it reaches the code cache's budget. If you run ethrex on Sepolia, run this one before the fork: v28.0.0 is the minimum that follows it, and v29.0.0 adds the access list hardening on top. Everyone else, upgrade at your convenience. No resync is needed. We also left something in the logs for the first Amsterdam block. We are not going to spoil it, so watch your node cross the fork.

Our Community Call is tomorrow at 2:30 PM UTC, live on X. We'll be walking through recent progress, key updates, and what's coming next. Join us.

A new @class_lambda player is joining the game! We'll be working on bringing a more unified UX to node operators 🫡

For the first time, ethrex is listed alongside every other execution client in an Ethereum Foundation upgrade announcement. ethrex v28.0.0 is ready for Glamsterdam on Sepolia, activating October 6. We've run every Glamsterdam devnet since devnet-0 and Platåberget since August to get here. Proud to be part of Ethereum's client diversity. If you haven't run ethrex yet, it is always a good time to start.