Manifesto

An endpoint you own

Every application that reads a blockchain needs an index. Raw nodes answer questions about blocks and transactions; applications ask questions about balances, positions, holders, and history. The layer that turns one into the other is not a feature you bolt on at the end. It is infrastructure, and it sits underneath everything the product does.

For most teams that layer arrives as a bill. You send your queries to someone else's service, you pay per request or per month, and the meter never stops. It works, and on day one it is the fastest path. But you have handed a load-bearing part of your application to a vendor, priced it as a subscription, and accepted that if the vendor changes terms, degrades, or disappears, your product goes with it. That is a choice. It is worth noticing that it is a choice, because it is usually made by default.

Nuthatch exists because the default should be the other way round. An index of a public chain is derived data. The inputs are public, the transformation is code, and the output is something you can hold on your own hardware. There is no physical reason it has to be rented. A single machine, often a small one, can follow a chain's tip and serve a product's queries. The hard part was never the compute. It was the operational weight of the tools, and that is an engineering problem, not a law of nature.

So Nuthatch is one static binary. Point it at a contract address, and minutes later you have a live indexed API on your box: entities, read-only SQL, balances, and MCP, over an embedded store with no Postgres, no Docker, and no third-party data service in the path. It runs against any RPC today, and — by design — your own node next. Nothing phones home, because there is nothing it needs to phone home about.

Owning the endpoint only matters if you can trust it, so the core is deterministic by construction. Decoding, reorg handling, and entity derivation are reproducible: the same blocks in always produce the same state out. Entities are declarative incremental views, so a reorg is a retraction rather than a hand-written rollback, and a backfill is the same logic run as a batch. Where you need imperative code, it runs as a capability-isolated WASM component: one granted no capabilities is a pure function, and a pure function can be re-executed and checked by anyone. Verification here means running the code again and getting the same answer, not trusting a claim about it.

This is why Nuthatch is free and licensed AGPL-3.0. Infrastructure you depend on should be infrastructure you can read, run, fork, and keep. There is no hosted tier, no token, and no pricing page, because the point is not to sell you back the endpoint you could run yourself. Use it, change it, and if you run a modified version as a service, share the changes. That is the whole arrangement.

Be your own indexer. Install it, or read the source.