Datost (Datost) is an AI data analyst in Slack.
@ycombinator1.7M followers
147,029 views
33 Comments
Comments from the original post on X (Twitter)

@TheMasterTs
Comparing a tool with your full business context pre-loaded to a raw model running cold is not a fair benchmark. Of course it wins. The real number to ask about is how many hours your team spends building and maintaining that semantic layer, and what happens the week your data schema changes.
5 likes

@Prithvi_Jadwani
@DatostApp @MaceoCk Datost better be more than just a fancy Slack bot tho. What's its actual compounding effect on org's data workloads?
4 likes

@bygregorr
@DatostApp @MaceoCk I'm curious how you handle the trust gap when the AI gets it wrong on the other 24.8%. One bad query on production data and teams stop using it. Are you surfacing confidence scores or letting users verify the SQL before it runs?
4 likes

@akkkkiira
@DatostApp @MaceoCk congrats!!
3 likes

@bryce_meow
@DatostApp @MaceoCk Congrats team!
3 likes

@Malik_Muneeb_3
@DatostApp @MaceoCk Exciting news, @datostapp! The breakthrough in text-to-SQL accuracy is impressive. Looking forward to seeing how this transforms data analysis in Slack. Congrats on the launch!
3 likes

@kaushalxcode
@DatostApp @MaceoCk AI data analysts inside Slack feels like the next big shift. Datost building a semantic layer over CRM, docs, and codebase is the real unlock. 75.2% vs 33% on text-to-SQL shows what context can do. We’re moving from querying data to collaborating with it. Congrats to both
2 likes

@manav2modi
@DatostApp @MaceoCk Congrats!
2 likes

@anayflima
@DatostApp @MaceoCk Interesting! Congrats on the launch!
2 likes

@MahitNamburu
@DatostApp @MaceoCk Datost 📈📈
2 likes

@erenzmendi03
@DatostApp @MaceoCk Sick stuff!
2 likes

@connortbot
@DatostApp @MaceoCk i love datost
2 likes

@kirubeltewodro2
@DatostApp @MaceoCk The pattern we've been wrestling with on our own repo: exact string template matching as the dispatch layer. When a paraphrase misses, the fallback is a heuristic that emits a weaker query and the LLM never sees the rewording, so the failure looks like model non determinism.
2 likes

@maradoh22
@DatostApp @MaceoCk Wowwwww would have needed that as a data guy for sure
2 likes

@ceefryingpan
@DatostApp @MaceoCk 🔥
2 likes

@GJarrosson
@DatostApp @MaceoCk Congrats! 🚀
1 like

@PrathamNanekar
@DatostApp @MaceoCk We built something similar for our hackathon, In plain terms, it lets a user manage backend operations by chatting in slack instead of getting their hands dirty with the actual UI :
1 like

@kradisme
@DatostApp @MaceoCk here we gooooo
1 like

@Aqib__786Ai
Big jump — that semantic layer is really the key piece most text-to-SQL tools miss. Datost basically isn’t just translating queries, it’s grounding them in how the business actually defines its data, which is why that benchmark gap vs older models like Claude Opus 4.6 makes sense. This is where AI data tools start becoming reliable enough for real decision-making: • shared business definitions (no ambiguity in metrics) • context across CRM + docs + code • SQL generation that matches intent, not just syntax If this holds in real enterprise setups, it’s less “chat with your data” and more “ask and act on your data” inside Slack.
1 like

@c0delemons
@DatostApp @MaceoCk Good stuff
1 like













