Mirror

Start from history, not from block zero

A backfill is the expensive part of running a nest: a million RPC credits for the larger Graph Protocol nests, every one of them buying history somebody has already indexed. The public mirror holds that history. nuthatch seed downloads it, checks every file against its content address, and leaves your nest following the chain from the block where the mirror ends.

1 · the nest

Clone the nest and check out the commit it was published from. The mirror is found by the nest's data identity, so any other version finds nothing.

2 · the history

Seed downloads the sealed segments, refuses any file that does not hash to its address, and resumes if interrupted.

3 · the tip

Dev follows the chain from the next block, over your own RPC. Catching up a day costs a few hundred calls.

graph-gns-nest

published 2026-10-05

chain: Arbitrum One
history: 42,008,725 to 511,898,013
rows: 74,694
files: 23 segments, 7 tables

shell

git clone https://github.com/nightswatchhq/graph-gns-nest
git -C graph-gns-nest checkout 33d010ff69b5
nuthatch seed --from https://pub-bc282d5016f242a783a6c28cbcd4401a.r2.dev --dir graph-gns-nest
nuthatch dev --dir graph-gns-nest

nightswatchhq/graph-gns-nest ↗

graph-allocations-nest

published 2026-10-05

chain: Arbitrum One
history: 42,449,227 to 511,913,482
rows: 5,946,669
files: 4,764 segments, 73 tables

shell

git clone https://github.com/nightswatchhq/graph-allocations-nest
git -C graph-allocations-nest checkout 7795a335a910
nuthatch seed --from https://pub-bc282d5016f242a783a6c28cbcd4401a.r2.dev --dir graph-allocations-nest
nuthatch dev --dir graph-allocations-nest --state-rpc <archive RPC URL>

nightswatchhq/graph-allocations-nest ↗

This nest samples contract state, so dev needs an archive endpoint for eth_call.

dips-nest

published 2026-10-05

chain: Arbitrum One
history: 486,895,281 to 511,814,696
rows: 66
files: 26 segments, 22 tables

shell

git clone https://github.com/nightswatchhq/dips-nest
git -C dips-nest checkout 0cf0ec2c7b1a
nuthatch seed --from https://pub-bc282d5016f242a783a6c28cbcd4401a.r2.dev --dir dips-nest
nuthatch dev --dir dips-nest

nightswatchhq/dips-nest ↗

data-services-nest

published 2026-10-05

chain: Arbitrum One
history: 456,917,519 to 495,618,196
rows: 7
files: 4 segments, 4 tables

shell

git clone https://github.com/nightswatchhq/data-services-nest
git -C data-services-nest checkout 9696cb4beb1d
nuthatch seed --from https://pub-bc282d5016f242a783a6c28cbcd4401a.r2.dev --dir data-services-nest
nuthatch dev --dir data-services-nest

nightswatchhq/data-services-nest ↗

What the hashes prove

That every file is the one the mirror's catalogue names, so nothing changed on the way or at rest. They do not prove who wrote that catalogue, or that it is true to the chain. Seed from a mirror run by someone you would trust to run the nest for you, or re-index a sample of it yourself and compare.

Publish your own

Any nest can be a mirror. Run it with --publish-target s3://your-bucket/prefix, or run nuthatch publish sync on a stopped one, and point people at the bucket's public address. The mirror is append-only and costs nothing to read from a bucket that does not charge for downloads. The operator guide has the bucket policy.

Read it without Nuthatch

The mirror is plain Parquet, so DuckDB, Trino or anything else that reads Parquet over HTTPS can query it directly. A public bucket cannot be listed: each dataset's manifest.json names its tables and, for every segment, a hash, and the file is at <dataset>/<table>/<hash>.parquet. The manifest's file field is the name inside a nest directory, not the address here.

duckdb

SELECT count(*), min(block_number), max(block_number)
FROM read_parquet('https://pub-bc282d5016f242a783a6c28cbcd4401a.r2.dev/494ce630fa704ee504e6cdfb8f732c66d585843face32fbbc6b7bf115a9b15c7/gns__subgraph_published/e03958df317ac9e39634669f04c7fc1c6e6caed29e7b5f37630235acd8c9f7f2.parquet');