Fabra: Product analytics on your data warehouse
github.comHey, congrats on the launch. This may be clear for others, but personally I’m missing a bit more details. What precisely is “product analytics on [a] data warehouse”? Are you an Amplitude competitor? Or more of a DIY scripting alternative?
I like that you kept things brief but some screenshots and use cases would be a nice improvement for the landing page.
That's right, we're an Amplitude competitor— but run the queries directly in our customer's own data warehouse.
Appreciate the feedback, will be uploading more screenshots shortly.
Hi everyone! I wanted to share Fabra— product analytics for the modern data stack. Fabra runs directly on your data warehouse, meaning we don't add another data silo or need you to setup costly ETLs.
In addition to self-hosting our open source product, we also offer a hosted version which is WAY cheaper than other tools since we don't store any data.
Let us know if you have any feedback, you can always reach me at nick@fabra.io
It would be good to get a breakdown of this vs existing products. For example, I use Metabase quite heavily but I'm having trouble comparing what your product gives me in comparison to it.
Congrats on launching!
How do you build a data warehouse without ETL?
We aren't a data warehouse, just the query-building and visualization layer. We let product teams easily do things like build funnels, measure trends, and more with the data they've already collected.
From your tagline:
> Build funnels and measure trends directly on your data warehouse, without paying for costly ETL and duplicate storage in outdated analytics tools.
Not being snarky, I just don't understand. The output of an ETL is data in your data warehouse. Then you build reports on top of the DW.
How does Fabra prevent me from needing an ETL?
Reverse ETL would be a better description here— think data warehouse -> Amplitude
Is it similar to https://chartbrew.com ?
Fabra is built for the modern data stack, meaning our customers already send all of their data to a central data warehouse as their single source of truth.
It looks like Chartbrew connects to various third-party tools directly, which can lead to silos/discrepancies in our experience.
Chartbrew founder here, just wanted to clarify for readers that Chartbrew doesn't store any data from the 3rd party tools so there's no silo problem. It queries the data sources, computes the data, and stores only the chart configurations.
The only storage Chartbrew does is to cache the data while building a chart so the data source is not queried too many times and the user can see changes quicker. Afterwards, Chartbrew keeps the charts updated with data coming directly from the other tools.
But allowing for integration with multiple data sources doesn't mean you can't use only one if you prefer so. Does Fabra also store the data, or only connects to an external data warehouse?
Exactly! We don't store any data and natively connect to external data warehouses. This means we're way cheaper— no need to move data or store a copy.
I understand, then what's the advantage compared to ChartBrew? Not having the multi-data-source feature doesn't sound like an advantage.
My take is you should always just store any third-party data sources in your data warehouse, where you can then join it with all the other data. Having a feature like this just enables bad practices, and we'd rather be opinionated about how the data stack should look.
Since this is the "best practice" architecture, we've made the decision to focus on integrations with data warehouses specifically— Snowflake, BigQuery, etc. I don't think Chartbrew has these.
Not the OP, but:
Oftentimes data teams have those data sources in the warehouse already (or a process for ETLing new ones in).
Copying data only once means lower costs (maybe) and fewer inconsistencies.
Plus if they’ve done work to join/clean/transform the data in the warehouse they can take advantage of that for product analytics without having to reinvent the wheel in ChartBrew or Mixpanel.
Cool, I'll give it a spin tomorrow on a fairly hefty dataset and see how it goes.