Can I build a Substreams package without knowing Rust?

Updated August 2026

Yes. You can reuse a ready-made package from the Substreams registry, or install the open-source Substreams agent skills and have an AI coding assistant build a custom package from a plain-English description.

Rust is still what a Substreams module compiles to. It is no longer what you necessarily have to write.

What are my options, from least effort to most?

	What you do	Good for
Reuse a package	Pick from the registry, point it at a sink	Common cases — popular contracts, standard token transfers
Prompt an AI assistant	Install the skills, describe the data you want	Custom logic, specific contracts, your own schema
Write it yourself	Rust modules and a manifest	Full control, performance tuning, unusual requirements

Most people who think they need option three actually need option one or two.

How well does the AI route actually work?

Better than you would probably assume, and it has been measured rather than asserted. StreamingFast published an evaluation in which an agent was given plain-English prompts for 14 tasks across Ethereum and Solana — block statistics, USDC transfers, NFT mints, Uniswap V2 and V3 swaps, Raydium CLMM swaps, Pump.fun launches, a Postgres sink.

Every one of the 14 built and ran successfully. Twelve of the 14 produced output matching a golden reference byte for byte. Two were completed with no ABI or IDL available at all — the agent derived what it needed from Solidity and Rust source.

The evaluation is also explicit about limits. Vague prompts produce confident guesses rather than clarifying questions, so specify your contract addresses, block ranges, and output schema. Non-standard ERC-20 tokens can break field decoding. Read it before you rely on it.

What does the workflow look like in practice?

Install the skills into Claude Code, Cursor, or VS Code. Describe the dataset you want in ordinary language — which chain, which contract, which events, what shape the output should be, where it should go. The assistant generates the manifest, the protobuf schema, and the Rust modules, then builds and runs them so you can inspect the output.

You review the result rather than author it. If you can read SQL and reason about what a table should contain, you can evaluate whether the output is right.

Do I still need to understand what's happening?

Yes — and this is the honest caveat. You need to know what data you want, roughly how the chain represents it, and how to tell whether the output is correct. The skills remove the syntax barrier, not the thinking.

That is a meaningful distinction, but it is also a much lower bar than learning Rust.

Frequently asked questions

Which AI assistants are supported? Claude Code, Cursor, and VS Code 1.107+. The skills are open source under Apache 2.0, so they can be adapted to other assistants.

Is the generated code production quality? It builds and runs, and in the published evaluation it matched golden references on 12 of 14 tasks. Treat it as you would any generated code — review it, test it, and be specific in your prompt.

Do I need to be a developer at all? You need to understand your data requirements and be able to judge whether output is correct. You do not need to write Rust. Non-engineers have shipped working packages this way.

Get an API key at thegraph.market — no personal information required — and install the skills from github.com/streamingfast/substreams-skills.