Solana Yellowstone gRPC Streaming — Kaldera

Kaldera is constant-k's enhanced Yellowstone gRPC service for Solana: account, transaction, slot, block and entry streams delivered over gRPC from dedicated bare-metal RPC nodes in New York and Frankfurt. It speaks the standard Yellowstone (Geyser) protocol your tooling already understands — then adds leader shred injection, a per-subscriber fanout layer and an optional pre-execution fast path on top.

What Kaldera Adds to Yellowstone gRPC

Shred-Accelerated Updates

Standard Yellowstone gRPC emits updates after the validator executes a block. Kaldera additionally ingests leader shreds as they arrive, so hot data reaches your stream ahead of the normal Turbine propagation path. When you are racing for a landing slot or watching a pool for the first state change, those milliseconds are the edge.

Standard Filters, Generous Caps

Filter your stream by account owners, addresses, transaction inclusion, commitment level and data slices using stock SubscribeRequest filters — no proprietary query language. Reconnections are painless too: the geyser keeps roughly twenty minutes of slot replay, so a dropped connection does not mean a gap in your data, and zstd compression is available on every stream.

A Dedicated Lane for Every Subscriber

constant-k's custom gRPC fanout gives each subscription its own isolated stream from Yellowstone. A slow consumer elsewhere can never stall your feed, and your bandwidth allocation is yours alone: 12.5 MB/s on Blade Runner and 75 MB/s on Nexus, with burst headroom above both.

Turbo Mode: Pre-Execution Streaming

For workloads where seeing the pipeline early beats seeing it enriched, Kaldera can serve your regular Subscribe request from a shred-reassembly path that surfaces transactions before execution — two stages earlier in the transaction lifecycle. It is raw and unconfirmed by design, and it is there when you want it. Meet Turbo Kaldera (aRPC) →

Run by Operators, Not Resellers

Kaldera runs on bare-metal servers we build, own and tune ourselves in New York and Frankfurt — the same nodes whose latency we publish live on our benchmarks page. No white-labelled capacity, no shared firehose, no mystery middle-boxes between you and the chain.

Connect in Minutes

Kaldera speaks the stock Yellowstone protocol, so any Yellowstone client works unchanged. A minimal TypeScript example with @triton-one/yellowstone-grpc:
import Client, { CommitmentLevel } from "@triton-one/yellowstone-grpc";

const client = new Client(
  "https://kaldera-newyork.constant-k.com", // or kaldera-frankfurt
  process.env.CK_API_KEY,                   // your constant-k API key (x-token)
  undefined
);

const stream = await client.subscribe();

stream.write({
  commitment: CommitmentLevel.PROCESSED,
  accounts: {
    pools: {
      account: [],
      owner: ["675kPX9MHTjS2zt1qfr1NYHuzeLXfQM9H24wFSUt1Mp8"], // Raydium AMM v4
      filters: [],
    },
  },
  transactions: {},
  slots: {},
  blocks: {},
  blocksMeta: {},
  entry: {},
  accountsDataSlice: [],
});

stream.on("data", (update) => {
  // updates arrive as they happen — shred-accelerated
  console.log(update.filters, update.account?.account?.pubkey);
});
Kaldera is included on the Blade Runner and Nexus plans. Compare plans on the pricing page, check per-method caps on plan limits, and watch the live p50 latency of the exact nodes serving you on our Solana RPC benchmarks. Need raw shreds instead of decoded updates? That is ShredPrism.