What Shipped at Dwellir in October 2026: Elysium RPC, Measured Hyperliquid Latency, L1 Gateway 1.12 and a Rebuilt Site

In October Dwellir added Elysium testnet RPC with archive state, debug tracing and WebSocket, published measured latency for Hyperliquid fills, order books and the mempool stream, shipped L1 gateway 1.12 behind the Hyperliquid streams, put yearly billing on the pricing page at 14 to 17% off, and rebuilt the website and the public endpoint directory. The Elysium mainnet endpoint followed on 8 November. This is the first monthly summary; every item links to its page, post or docs.

New Chains and Endpoints

Elysium was the only chain added in October. Its testnet RPC went live on 22 October. Elysium is Kinetiq's Arbitrum Orbit chain; it settles to HyperEVM and uses HYPE for gas, and What Is Elysium? covers the architecture. The public RPC rejects WebSocket upgrades, has no debug namespace and caps eth_getLogs at 2,000 blocks. The Dwellir endpoint runs Arbitrum Nitro in archive mode with the eth, net, web3, debug and arb namespaces over HTTPS and WebSocket. The launch post has the viem, Foundry and Hardhat quickstart, and Top Elysium RPC Providers compares six providers on archive, debug and price.

Mainnet followed on 8 November, ahead of Kinetiq's 9 November launch. The two additions before Elysium, Etherlink and Taiko, went live in the last days of September.

Both answer eth_chainId, debug_traceBlockByNumber and a WebSocket newHeads subscription, checked on 9 November. trace_* is not available on either, because Nitro does not implement it.

The public RPC endpoints page was rebuilt on 7 October. It used to list 27 Polkadot-family URLs. It now has one row per network, 136 networks on 96 chains, with the chain ID where one exists (60 networks, testnets included), the HTTPS and WSS URLs, whether a key is needed and whether the node is archive, plus a search box and a JSON twin. Arc got debug and trace docs pages on 13 October, the retired Moonbeam, Moonriver and Moonbase docs were removed, and Tempo replaced Gnosis among the popular chains in the Chains menu on 20 October.

Existing endpoints got more capacity. Two Base full nodes in OVH US East joined the Base pool on 1 October; they serve recent state and send older-state calls to the archive nodes. Two Robinhood Chain archive nodes in the same OVH US East region went into rotation on 30 September, where US East traffic had been served from the West archive. A fifth Tokyo Hyperliquid node, on Latitude bare metal, came up on 19 October as dedicated capacity.

Hyperliquid Data Products

October's Hyperliquid work was measurement and gateway releases. Every latency figure on the site now comes from a dated run on an AWS Tokyo (ap-northeast-1) machine.

The order book figures are from a dedicated node; shared endpoints were within 15 ms at the median. The public l2Book feed sent about one update every 5 seconds against about 14 per second from Dwellir. The latency benchmark puts Hydromancer and GoldRush in the same tables, and the site's order book claim changed from "24% faster" to "over 30% lower median latency" to match it. The mempool post explains why a 357 ms lead matters when cancels run before orders inside a block.

The tooling lives in Dwellir's hyperliquid-benchmarks monorepo. Its surfaces-bench tool measures every Hyperliquid surface from one AWS Tokyo (ap-northeast-1) machine: HyperEVM HTTPS and WSS, the HyperCore /info endpoint, L2 and L4 order books, gRPC StreamBlocks and StreamFills, the raw WebSocket trade tape, HIP-3 builder dexes and HIP-4 outcome markets, and writes one self-contained HTML report. Every benchmark in the repo records the host it ran from. Tokyo VM-only probes were added on 5 October, and on 6 October an acceptance runner that puts each newly deployed gateway endpoint through the smoke suites and a data-parity and latency check against the shared endpoint, with a pass, warn or fail summary.

L1 Gateway Releases 1.11 to 1.12

The L1 gateway, the service behind the Hyperliquid gRPC and WebSocket endpoints, went from 1.11.0 on 30 September to 1.12.0 on 20 October.

1.11.0 is where the mempool stream comes from. StreamMempoolTxs and GetMempoolTxs serve the node's pre-consensus mempool_txs lines verbatim, positioned by timestamp, and Position.timestamp_ns resumes exactly. The same release trades memory for latency: order-statuses blocks decode 84% faster and l2Book fan-out is 65 to 81% faster at 5 to 40 subscribers, in microbenchmarks against 1.10.1. tpslUpdates and StreamTpslUpdates take a market selector, so spot and HIP-4 outcome trigger orders are covered once the L1 emits them. Patches 1.11.1 to 1.11.3 on 5 October fixed filtered StreamMiscEvents failing on the CreateSubAccount events mainnet began emitting that day.

1.12.0 brings the WebSocket market channels to gRPC parity: a new l4BookUpdates channel carries order-level diffs over the resting book, users and coins filters work on trades, tpslUpdates, l4BookUpdates and allFills, and a dexes filter narrows the fills, order-status, book-diff and TP/SL streams to one HIP-3 dex. Progress frames are opt-in on the six block-positioned v3 raw feeds (20 second default, 10 to 60 seconds), so a filtered stream that goes quiet still advances its resume cursor and stays open through idle-closing proxies. The transport now measures backpressure, a lagging l4Book subscriber receives a fresh snapshot on the same connection instead of a close, a node pause no longer closes subscriptions (staleness-mode: hold), and POST /info caches meta, spotMeta and the other rarely-changing metadata types for 10 seconds. A raw stream that falls behind the live feed now ends with DATA_LOSS instead of silently skipping records, so resuming from the last cursor re-reads what was lost.

A peering slot now allowlists your node on three Dwellir peers instead of two, in two Tokyo datacenters, at the same $500 a month.

On the archive side, the historical data docs show today's date as the end of coverage for the eight live S3 prefixes, since the uploader adds a day every 24 hours, and gained a section on one-off windows and filtered exports, which are quoted by email. Each data catalog page now carries an "Observed fields" tree sampled from a real object, so you can check that a field exists before you buy a dataset. The homepage shows the raw HyperCore archive as its own product: 119.16 TB of node files since 25 January 2025, with L4 book reconstruction possible from 14 April 2026.

Docs and Guides

The Agent Cookbook went up on 12 October with 12 copy-ready prompts for AI coding agents working against Dwellir endpoints. The CLI docs moved to 0.3.0, which never prints a full API key; keys copy and keys env replace reading the key out of --json output. The support docs gained RPC and streaming troubleshooting pages and an archive sizing estimator.

Six production method guides landed: eth_getBlockReceipts, eth_simulateV1, debug_traceCall, eth_createAccessList, trace_filter and trace_block and eth_getProof. Base's 200 ms blocks works through what 10x the blocks does to a per-block workload: a 29-call tip-of-chain stack costs $112 to $977 a month on Dwellir, and the post prices the same stack on two other providers. Also new: Hyperliquid fees explained, Best Bittensor RPC Providers, an onchain research agent with Claude Code, building financial apps on Arc and a guide to finding the wallet behind a Hyperliquid order. The provider comparisons were re-checked against live pricing pages: free-tier facts for dRPC and one other provider across 34 posts, and Chainstack's Business plan at $499 for 200M requests.

Pricing and Plans

Prices did not change. Developer is $49, Growth $299 and Scale $999 a month; one response is one credit, trace and debug included. What changed is how the pricing page shows them.

• A Monthly/Yearly toggle on the Shared Nodes plans. Yearly billing is $500, $3,049 and $9,900, 14 to 17% less depending on the plan, and until 15 October it was only visible inside the dashboard after signup.
• The provider cost calculator moved to /rpc-provider-pricing-calculator, with a fix to its overage maths.
• The plan comparison table renders on the server, so all 23 facts in it (prices, allowances, overages, RPS and burst, eth_getLogs range, trace and debug) are in the HTML for crawlers and AI assistants. Before the change, 7 were.
• Developer plan support is listed as Telegram DMs, which is what it was in practice. The cards said "chat support".

The Free plan stays at 20 responses a second and 100,000 a day. Unlimited Nodes (25 to 1,000 RPS, $119 to $4,799 a month, flat) and the dedicated cluster list prices (archive clusters from $2,400 a month) date from September and did not change either.

Website and Tooling

The website moved to a new design system over the month. The design was locked in on 15 October; pricing, the homepage, the network directory, the docs and the blog moved onto Tailwind and shadcn parts between 16 and 24 October, and the full rebuild merged on 1 November. Network pages show uptime over the past 90 days and link the status page. Fonts are self-hosted, the site runs Next.js 16.4.2, a rune icon set replaced lucide-react, and the brand book is public.

For search engines and AI readers: an IndexNow key is live, 15 network page titles now say what the endpoint serves ("Polygon RPC Endpoint: Archive & WebSocket"), and "chain rpc" links across the provider posts and docs point at the network pages. Docs AI classifies prompt-injection attempts and answers archive export and quote questions from the docs. Envio joined the customer stories.

Early November

Besides Elysium mainnet: the pricing table states the free plan's 5 million monthly responses; three more method guides (debug_traceBlockByNumber, debug_traceBlockByHash, eth_sendRawTransaction); exchange deposit detection across 18 chains; high-throughput RPC providers ranked; and a Hyperliquid vaults and leaderboard guide.

On the L1 gateway, 1.13.0 (2 November) serves every gateway-native stream and read over WebSocket as well, on a new /ws/raw route with the same positions, filters and replay as gRPC v3. 1.13.1 and 1.13.2 (3 November) fixed L4 queue order: the gateway had ignored the node's insertBefore on priority ALO orders since July, so about 6% of new orders sat at the back of their price level on StreamL4Book and L4 snapshots when the node had them in front of the anchor. 1.14.0 (4 November) adds per-wallet orderUpdates on WebSocket and gRPC v3, metaAndAssetCtxs and allMids for HIP-3 builder dexes and HIP-4 outcome mids, and a graceful drain on shutdown, so clients are routed elsewhere before they are disconnected.

Try the Elysium Endpoint

Create a free account, pick Elysium mainnet or testnet in the endpoint list, and run the eth_chainId call from the launch post. You should get 0x1863d on mainnet and 0x185d9 on testnet. For a dedicated Elysium node or the Hyperliquid gRPC add-on, contact the Dwellir team.