I had never opened rotki. Now it reads my history from my own indexer
By this week I had written an integration between nuthatch and rotki, opened the pull request, and signed the CLA. I had not, at any point, opened rotki. This is the afternoon I put that right, using my own wallet, and wrote down everything that happened, including the parts that went wrong.
What rotki is, and why it has to ask a stranger
rotki is an open-source portfolio tracker and crypto accounting tool. It runs on your machine, keeps its database encrypted on your disk, and turns your addresses and exchange accounts into balances, decoded history and profit-and-loss reports. It is the privacy-minded alternative to the hosted trackers, and it is quite good at it.
It has one awkward dependency. To learn an account’s history it asks a block explorer: Etherscan, Blockscout or Routescan, for every transaction, transfer and withdrawal of every address you track. The explorer therefore learns every address you own, which is an odd thing for a privacy tool to have to do. It also became an expensive thing to do this year. Etherscan’s free API now refuses Base, Optimism, BSC and Gnosis outright (“Free API access is not supported for this chain”), and rotki’s own issue tracker has been finding out what that does to its users.
A nuthatch nest in address-history mode answers the same questions locally. It watches only the
accounts it is given, reads the chain over your own RPC, and serves the answers at /api in the
shapes rotki already reads. The rotki side is
rotki/rotki#13342, which is awaiting review as I write.
Getting rotki running took five minutes, one of them on Node versions
I ran rotki from the pull request’s branch, because the integration is not in a released version yet.
rotki pins Python to 3.14 and Node to >=24 <25. My Mac has Node 26, and rotki’s preinstall check
said so in red. brew install node@24 installs alongside without replacing the default, and after
that it was uneventful: uv sync installed 152 Python packages in a fifth of a second, pnpm install
took 18.7 s, and pnpm dev:web compiled rotki’s Rust supervisor in 28.7 s and had the interface up on
localhost:8080 about 40 seconds later. It downloads an Electron binary even in web mode, which I
mention only because it is the kind of thing one notices.
I created a local user, pointed rotki’s Nuthatch setting at the nest, and set Ethereum’s indexer order to Nuthatch alone. Etherscan was not a fallback. It was not in the list.
The first sync failed, and it was my clone’s fault
rotki asks for history from block 0. My nest’s history starts at block 25,000,000, and it says so: NUTHATCH_INCOMPLETE: history starts at block 25000000.
The integration is meant to retry from the block the nest names. It did not, and the sync failed.
The reason was mundane. My local rotki checkout was 42 commits behind the pull request, and the version I was running predated the retry. On the pull request’s actual head, rotki tried block 0, was told where history starts, asked again from 25,000,000, and got its answer, for transactions, internal transactions and token transfers alike. The integration was fine; my copy of it was 42 commits behind.
My own wallet found two defects in a release I had called ready
Before rotki ever asked it anything, I indexed my wallet with the released nuthatch 5.3.1, and it was about to take twelve hours.
A backfill works in windows, and the window is sized so that a request arriving mid-backfill never waits long: a quarter of the 8 s request deadline. Over a hosted RPC, each window’s checks cost about 2.3 s whatever its size, so every window ran long, the pacer kept halving, and it settled on its floor of 64 blocks per 2.3 s. For a wallet with 1.19 million blocks of history, that is roughly twelve hours. Every test had passed, because the one that covered this ran with a 20 s deadline, under which a 2 s window looks perfectly healthy.
The second was quieter. Once the account was covered, the nest started reading every block body, about a million of them, to look for staking withdrawals and produced blocks. rotki only ever asks for those about a tracked validator’s withdrawal address. My wallet has never staked anything.
nuthatch 5.3.2 fixes both. The pacer now separates a window’s fixed cost from its per-block cost and lets windows grow past it, never beyond the request deadline. Block bodies are read only for an account somebody has asked about. On the default deadline the same wallet now backfills in about 20 minutes and 7,324 RPC calls, and the index it leaves behind is 2.5 MB.
Nine transactions, nineteen events, and nothing asked of Etherscan
I checked the nest against Etherscan first, hash by hash, block by block, value by value: 8 normal transactions, 4 token transfers and 1 internal transaction, identical on both. (The same comparison for vitalik.eth over blocks 18 to 20 million matched exactly, field by field, which is the larger test and the one the integration was built against.)
Then rotki took over. It pulled the history from the nest in 11 s and decoded 9 transactions into 19
events, swaps and approvals among them, each with its gas. Its log shows 9 indexer queries, all of them to 127.0.0.1:8394, and no request to Etherscan
at all. It is genuinely boring, which is the highest compliment one can pay a data source.
What still leaves the machine
I would like to say it was all local. It was not, and the log says so.
rotki talks to Ethereum nodes as well as to indexers, for receipts and balances, and by default it uses a rota of public ones. During this one sync it contacted about ten of them, among them 0xrpc, nodies, publicnode, drpc, sentio, blockpi, blockrazor and pocket, and Stake DAO’s API besides. The nest replaces the block explorer. It does not replace rotki’s node connections, and those providers see your addresses in the requests they serve.
The honest claim today is narrower than the headline I wanted: no block explorer learns your addresses. Getting to “nothing leaves the machine” means pointing rotki at your own node as well, which rotki supports.
Two other limits. The integration covers Ethereum mainnet only. Base, where rotki’s users lost the most, needs a different discovery path, because hosted Base RPCs cap trace queries at 100 blocks and re-execute them; that is the next piece of design. And none of this is in a released rotki yet.
Try it
The setup lives at nuthatch-indexer.com/rotki: one nuthatch.toml naming your account, your
RPC and where history starts, then nuthatch dev. Use nuthatch 5.3.2 or later. To use it from rotki
before the pull request is merged, run rotki from that branch, as above, with Node 24 to hand.