gamegoblin

Datafusion, Polars, and us are all based on Arrow. Datafusion is more targeting database users, Polars is targeting Python programmers/dataframe users, and we are targeting spreadsheet users.

They are all ultimately just different UIs on top of Arrow.

They're complimentary tools. We can take data out of the spreadsheet and put it into Polars instantly (you can do this in the Python code window if you want), etc.

Regarding why not implement our spreadsheet built on top of those: spreadsheet allow for heterogenous types in columns, so that requires a lot of extra infrastructure to manage, whereas Datafusion and Polars require homogenous column types.

No SIMD in the parser path... yet... :)

The main win from writing our own custom parser is writing directly into the final in-memory format from the parse stream without any intermediate allocations or data movement.

Awesome project! Adding it to my list of things we may plug into in the future.

Not yet, but we could add that feature. Feel free to contact us at contact[at]rowzero.io so we can learn more about your use case. We could also maybe do some kind of webhook situation? Lots of options.

We have definitely gotten requests to embed spreadsheets in webpages (e.g. as dashboards, etc), it's on the roadmap

We have native dataframe support that encourages homogenous types, but we also want to meet Excel users where they are! Probably half the world's economy runs on Excel, it's the world's most popular programming language :)

Not yet, but we can add it if customers want it.

You can technically use the Python to write back to Postgres if you are comfortable putting your creds into the code window. The downside of this is your creds would be viewable to anyone you share the workbook with.

We have gotten a lot of requests for write-back-to-DB (Snowflake, Postgres, etc) so will be adding first-class support for this feature soon that will use the same connection creds which are stored encrypted in KMS and are not viewable to people you share the workbook with.

Would love to chat about your use-case if you want to reach out to us at contact[at]rowzero.io

We have not tried clickhouse, but are probably comparable to DuckDB. DuckDB is also backed by Apache Arrow the same as us, but is targeting database users, whereas we are targeting spreadsheet users. They are very complementary.

We will add R2 support soon!

I will post it next week! It's very fun to play with. Also a good demo of Row Zero's templating capability — I can make a template so when you click the link, you get your own copy you can mutate and play with.

All the compute happens in the cloud, you're exactly right about the websocket pushing commands.

We use Apache Arrow data format, so it's fairly columnar. We have built some custom layers on top of Arrow as well to handle some fancier data types.

For storage and orchestration we use S3 and Dynamo.

Yes exactly, we use CRDTs for multiplayer stuff.

As the guy who wrote most of the Row Zero backend engine, I pray to god I am never asked to implement anything like this :)

Excel is really great if you don't have too much data. We love Excel! But we just had too much data for it, so we built the solution we wanted.

We can handle 100x more data than Excel (online or desktop), Google Sheets, etc.

I want to hear more about this self-rewriting VBA macro

All computation happens in the cloud. This is really nice because we can scale up and down to fit your data size.

If you need a supercomputer for 30 minutes, we can get you one.

It's very high on the list for post-launch feature additions

Copilot for formulas, auto-write python, etc, it'll all land soon!

All our backend engineers were on the S3 filesystem team.

I'm Grant Slatton, founding engineer at Row Zero, and before this I designed and led the team that built S3's new filesystem, ShardStore (check out this paper https://www.cs.utexas.edu/~bornholt/papers/shardstore-sosp21...). Our other Row Zero backend engineers (Breck, Erich, Greg) were all on the team.

Our frontend guy is ex-Airtable.

Google Sheets is banned in our households, we planned our neighborhood block party in Row Zero!

Tech stack:

Everything runs on AWS

Frontend: TypeScript + some Rust compiled to WASM, our own virtualized custom canvas magic

Backend: Rust, Apache Arrow

The biggest tech challenges... where to start.

1. As with any editor application, Undo/Redo is devilishly hard

2. A spreadsheet is modeled as a Directed Acyclic Graph where cells depend on other cells down to some root cells. There are a TON of fascinating graph theory algorithms you can bring to bear here. Some of the hardest problems I have ever worked on. My whiteboard looks like an insane person's with crazy arrows between cells and stuff.

3. Sometimes there really is no magic bullet and you just have to run benchmark, get flamegraph, chip away 3%, repeat, do that 100 times and you are very fast. Gotta grind.

I have a demo I will post soon that runs MNIST in Row Zero, we should be able to handle all of GPT2 without too much trouble, probably a lot faster than Excel, we may have to try it!

We haven't benchmarked against gnumeric, but will make a perf blog post in the next week or two (will post on HN too) and will see if we can add it to the comparison mix

We try really hard to be Excel-compatible with all our formulas and hotkeys, etc. If there is a formula or hotkey we are missing, let us know and we will add it (usually with < 24 hour turnaround time!)

The use-case you describe is exactly what we built it for. There are a lot of people who have data that is no longer human-scale and won't fit into Excel, but they still want to use their Excel skills.

We love Excel and spreadsheets, but when working at AWS we found we just had too much data to fit into it.

It varies depending on what you are doing, but is probably 100x faster on average.

Some things are literally 10000x faster because we are internally using algorithms with better asymptotic complexity than Excel, but there are a few edge cases where we are maybe only 10x faster

(but we are always working to improve that — half my job is just analyzing flame graphs and grinding out perf improvements)

Hello all, I'm Grant, founding engineer at Row Zero.

Working on this spreadsheet engine has been one of the most exciting, complex, and stimulating engineering experiences of my career.

Feel free to ask any technical questions about the product. We're really proud of what we've build and what's on the roadmap!

I just switched back from a decade of FF use to Chrome, sadly.

Whenever I would alt-tab back to FF, it would hang for a half second, just long enough be annoying. Chrome is instant.

Only thing I miss is middle click mouse drag feature.

Lots of those are various experiments with different prompting strategies, models, etc, so not representative. It regularly beats Stockfish Level 4 even without opening blunders.

source: I am OP

Only for Playstation consoles, but might have been one of the Twisted Metal games.

The real value of dictation, IMO, is it's so much easier to ramble verbally than via text, so you can just bang out a draft without getting blocked on picking the exact correct words.

I follow ~200 extremely high quality people on twitter, never use the algorithmic timeline -- just the "Following" tab, and it is easily 10x more signal to noise than HN for me, I'm almost upset I didn't start using it sooner.

Only works if you can resist the temptation to follow national news orgs, politics, etc. In my little niche, though, I essentially never see any of that, even less than one sees on HN. It's all just tech, AL/ML, space/rockets, startups, etc.

AWS has written parts of major services (EC2, Lambda, S3, etc) in Rust.