Routing model
If a lookup in Goldsky data fails or is incomplete, Boost forwards the request to your RPC provider. A forwarded read can add proxy latency, but Boost does not substitute a different result.
Methods eligible for indexed data
This table lists what Boost serves. Every other method is forwarded and still works.
Eligibility does not guarantee that Boost serves the read. If Goldsky does not have the requested data, Boost forwards the request.
Tip of chain
eth_blockNumber and eth_getBlockByNumber with latest are answered from Goldsky’s indexed data, which follows your source chain at the tip and through reorgs.
Other tags forward: pending, safe, finalized, and earliest on any method, and latest on methods other than eth_getBlockByNumber. The same rule applies to eth_getLogs when either range bound is a tag. Use a concrete hexadecimal height when you want those reads served by Boost.
Everything else forwards
Writes, traces, and state reads such aseth_call and eth_getBalance go to your RPC provider. Boost also forwards methods it does not recognize as eligible.
For a forwarded call, Boost preserves the JSON-RPC method, parameters, and ID. The result body comes from your RPC provider. HTTP headers are handled separately; see request and response forwarding.
Cache-only mode
Addcache-only=true and Boost never contacts your RPC provider. Calls it can answer from Goldsky data are answered; everything that would otherwise forward returns JSON-RPC error -32099 instead.
Use it when the forward is the cost you are avoiding. Boost is free during early access, but a forwarded call still reaches the provider you pay. A backfill, a replay, or any job with a spending ceiling can run against indexed data alone and retry the gaps later.
Send it as a query parameter or a header. The two are equivalent:
null already carries meaning here: eth_getTransactionReceipt returns null for a transaction that does not exist. So answering a miss with null would report an absence that Goldsky never verified. Repeat the call without the flag to forward it normally.
eth_chainId and net_version still answer under cache-only, because Boost synthesizes them and no provider is involved either way. Refused calls report x-edge-source: none: not endpoint, since nothing reached your RPC provider.
In a batch, the flag applies per item. Eligible items are served and the rest are refused individually, in one response array.
Skip-cache
The inverse:skip-cache=true, or the x-boost-skip-cache: true header, forwards the call to your RPC provider without reading Goldsky data. Use it to time your provider directly or to fetch a value Boost would otherwise serve. Statics still answer at the edge. The next call without the flag is served by Boost as usual; nothing is invalidated.
Write either value as exactly
true. 1, TRUE, yes, and a bare flag all read as “not requested”. Sending both flags on one request returns -32600; they contradict.Batch requests
Boost evaluates JSON-RPC batches per item. Eligible items can come from Goldsky data while the remaining items are sent to your RPC provider together in one onward batch.- Give every item a unique
id. JSON-RPC does not guarantee response order, so match results by ID. - Use
x-edge-billableas the served-item count.x-cacheisHITonly when Boost serves every item in the batch. - Keep arrays at 100 items or fewer. Larger arrays are forwarded whole without cache lookups.
x-cache: MISS and x-edge-billable: 44.