(Ryadh from ClickHouse here)
Your comment is spot-on.
This the main challenge with Agentic Analytics and there are known limitations. It is also where we are orienting our investments atm.
Our own experience running internal agents taught us that the best remediation comes from providing the LLMs with the maximum and most accurate context possible. Robust evaluations are also critical to measure accuracy, detect regressions, and improve. But there is no silver bullet.
SOTA LLMs are increasingly better at generating SQL and notoriously bad with math and numbers in general. Combining them with powerful querying capabilities bridges that gap and makes the overall experience an useful one.
IMO, we'll always have to deal with the stochastic nature of these models and hallucinations, which calls for caution and requires raising awareness within the user base. What I found watching our users internally is that, while it's not magical, it allows users to request data more often, and compounds in data-driven decision-making, assuming the users are trained to interpret the interactions
I'll freely admit you have more data (experience) to work with on this than I did in the tests I ran almost a year ago. I spent a lot of time documenting my schemas, feeding the LLM sample rows, etc and the final results were not useful enough even as a starting point for a static query that a developer would improve on and "hard code" into a UI. I approached it as both:
- Wouldn't it be cool to let my users chat with their data? ("How many new users signed up today/this event/this month/etc?" or "How much did we make yesterday?")
- An internal tool to use as a starting point for analytics dashboards
I still use LLMs to help write queries if it's something I know can be done but can't remember the syntax but I scrapped the project to try and accomplish both the above goals due to too many mistakes. Maybe my data is just too "dirty" (but honestly, I've never _not_ seen dirty data) and/or I should have cleaned up deprecated columns in my tables that confused the models (even with strict instructions to ignore them, I should have filtered them completely) but I spent way too much time repeating myself, talking in all caps, and generally fighting with the SOTA models to try to get them to understand my data so that they could generate queries that actually worked (worked as in returned valid data, not just valid SQL). I wasn't doing any training/fine-tuning (which may be the magic needed) but I felt like it was a dead end (given current models). I'll also stress that I haven't re-tested those theories on newer models and my results are at least a year out of date (a lifetime in LLM/AIs) but the fundamental issues I ran into didn't seem to be "on the cusp" of being solved or anything like that.
I wish you all the best of luck in improving on this kind of thing.
Keep in mind that it's not fully "fair", since these public dataset are often documented in the internet so already present in pre-training of the models underneath (Claude Sonnet 4.5 in this case)
It's a fair concern, and I understand where you are coming from. What I can say is that it's not our first rodeo incorporating another OSS product in our family. I tried to summarize it in the post:
> "This proven playbook is the same one that we applied when joining forces with PeerDB to provide our ClickPipes CDC capabilities, and HyperDX, which became the UX of our observability product, ClickStack."
If you research both instances above, the result is that these projects got more traction and adoption overall.
I hope this helps! and thank you for using LibreChat
Ryadh from ClickHouse here, happy to answers questions if folks have any.
So, why this move ?
Basically, we noticed that the existing "agentic" open-source ecosystem is primarily focused on developer tools and SDKs, as developers are the early adopters who build the foundation for emerging technologies. Current projects provide frameworks, orchestration, and integrations The idea behind the Agentic Data Stack is a higher-level integration to provide a composable software stack for agentic analytics that users can setup quicky, with room for customization.
Cool so your bet is that chat is essentially the new interface for BI... or ad hoc analytical inquiry... which opens up more dynamic BI... instead of asking an analyst in a slack conversation who then goes and runs a bunch of data pulls and munging, it's all handled agentically and a response is brought back to the user.. one thing is for sure: Tableau needs to be disrupted so happy to watch this one play out!
It actually comes from our own experience at ClickHouse.
We deployed this stack internally 8 months ago, and since very few people here have touched our legacy BI systems :)
I have never seen an adoption curve like this one tbh. It's obviously not perfect and can hallucinate sometimes, which can be tricky, but with the right approach and awareness in place, the value it delivers is massive. What really happens is that more users get access to data instantly, and as a result, we make better, data-driven, decisions overall.
My favourite use-case: our sales and support folks systematically ask DWAINE (our dwh agent) to produce a report before important meetings with customers, something along the lines of: "I'm meeting with <customer_name> for a QBR, what do I need to know?". This will pull usage data, support interactions, billing, and many other dimensions, and you can guess that the quality of the conversation is greatly improved.
This is really cool; does this mean Danny gets a salary to work on his open source project; would you consider this a "sponsorship" or would he have other jobs within ClickHouse / have a manager etc?
> The idea behind the Agentic Data Stack is a higher-level integration to provide a composable software stack for agentic analytics that users can setup quicky, with room for customization.
I agree with this. For those who have been programming with LLM, the difference between something working and not working can be a simple "sentence" conveying the required context. I strongly believe data enrichment will be one of the main ways we can make agents more effective and efficient. Data enrichment is the foundation for my personal assistant feature https://github.com/gitsense/chat/blob/main/packages/chat/wid...
Basically instead of having agents blindly grep for things, you would provide them with analyzers that they can use to search with. By making it dead simple for domain experts to extract 'business logic' from their codebase/data, we can solve a lot of problems, much more efficiently. Since data is the key, I can see why ClickHouse will make this move since they probably want to become the storage for all business logic.
Note: I will be dropping a massive update to how my tool generates and analyzes metadata this week, so don't read too much into the demo or if you decide to play with it. I haven't really been promoting it because the flow hasn't been right, but it should be this week.
My biggest question and concern is whether or not LibreChat will end up introducing the SSO tax or other "enterprise tier" features. Is this something you can speak on?
Interestingly, LibreChat has a broad range of applications already and we'll continue to support them. The investment area we want to tackle in priority is around the analytics use-case specifically.In that space, I don't see an SSO-tax scheme unfolding tbh, it's really about better visualizations, semantic layers and anything that can improve the quality of the insights produced on top of analytics data
That's good to hear! I'll be honest, I was a bit concerned about how LibreChat was going to support long term development and definitely see that this could be a good thing.
The reason we don't is that we currently use Drizzle for schema introspection and query building and Drizzle doesn't have an adapter for ClickHouse yet.
There's an active issue on the Drizzle repo requesting Clickhouse support that has some interest and the possibility of using the Postgres interface that ClickHouse exposes was discussed there.
Would be great to talk about this in more detail with you, shoot me an email ([email protected])
Congrats on Tablespace! it's good to see innovation in that space and I always love to see ClickBench mentioned in the wild :) (I work for ClickHouse).
I'm curious to hear your take about the tradeoffs that the HTAP model introduces? any impact on ingest times or query throughput for example?
The readme contains a lot of the implementation details:
"We use three different tables with different levels of detail: planes_mercator contains 100% of the data, planes_mercator_sample10 contains 10% of the data, and planes_mercator_sample100 contains 1% of the data. The loading starts with a 1% sample to provide instant response even while rendering the whole world. After loading the first level of detail, it continues to the next level of 10%, and then it continues with 100% of the data. This gives a nice effect of progressive loading."
