Thank you for your submission. You can access it here:
Last week we publicly released Kai, our in-platform AI assistant, a data engineering agent that can build integrations, write transformations, debug failures, and document your entire project. I'm extremely proud of the team and what we've delivered.
Yes, everyone has an AI assistant now. But most are chat wrappers that look great in scripted demos and fall apart with real work. Building something that actually helps our users day-to-day is a different challenge entirely - and that's what we set out to solve.
In this post, I'll share the challenge, our approach, and what Kai can actually do for you today.
From the beginning, our goal was to build a data engineering agent capable of managing the entire lifecycle of data processing within Keboola.
Keboola is a flexible platform. There are usually multiple ways to solve any problem, and experienced users can build sophisticated pipelines quickly. But that flexibility comes with complexity. Projects grow organically, team members change, and even experienced engineers can get lost when inheriting someone else's work or debugging a pipeline they didn't build.
We wanted to give every user an AI Keboola expert - one that understands the platform deeply and can guide you through any use case, just like having our most experienced consultant by your side.
Building a useful AI agent with such comprehensive capabilities is difficult. We realized early that an agent is only as good as its tooling and the context it has to achieve its goal. Without the right abstractions, even the smartest model will struggle to do anything useful.
So we started by focusing on the tooling itself. We were early adopters of MCP (Model Context Protocol) - an emerging standard that lets AI models interact with external tools and data - and invested significant effort in building the right abstractions for working with Keboola. After months of internal use and private beta testing, our MCP server was officially released in June 2025, and for many of us, it transformed the way we work with data.
With that foundation in place, we moved on to embedding our own agent directly within Keboola: context-aware and helpful to any user, with the MCP tooling at its core.
Kai is an embedded AI Assistant a data engineering agent that understands your data, transformations, flows, and configurations. It can help you explore, build, and debug, always with your approval.
Unlike generic AI assistants, Kai is context-aware. It knows where you are in the platform and what's happening in your project, so it can proactively suggest relevant actions rather than waiting for you to explain everything.
We built Kai with three principles in mind: guidance, safety, and control.
Your job failed? Open Kai, type "Fix it," and it will guide you through the process. With "follow mode" enabled (the default), Kai navigates you through the pages it's working on so you can see changes in real-time. Disable it if you prefer Kai to work in the background while you do something else.
Kai can work in development branches, ensuring it doesn't affect production. Review all changes before merging them back. Just create a development branch and start as many Kai sessions as you need.
Kai asks before performing any action that modifies your project and displays a diff with proposed changes. You always maintain control over what it does.
Since Kai became available, it's changed how we work in Keboola. I personally use it for feature adoption tracking and product analysis. I don't write code anymore, and honestly, I've even moved away from our internal BI system. When I need something, I just ask Kai to create a Keboola Data App that covers exactly what I need, then share it with whoever needs it.
Here are the top use cases from our private beta users:
When joining an existing project or inheriting someone else's work, Kai helps you understand what's already built. Explaining the data model, tracing lineage, and showing how transformations connect.
"It really does save time. Things that would have taken me two days I had done in three hours."
-- Analytics Lead, Private Beta user
Need a quick dashboard or report? Ask Kai to create a Keboola Data App in Streamlit. It handles the code, you review and deploy, then share it directly with business users who want the insights.
Need to connect a new data source? Kai can set up extractors, configure destinations, and even create custom Python components when a native integration doesn't exist.
"When I have a new API and API documentation, I say 'make me a Python extractor' and I'm done. It's absolutely great for that."
-- Data Engineer, Private Beta user
Kai reads the API documentation, builds the extractor, debugs it, and provides end user documentation - all in one conversation.
From simple SQL queries to complex multi-step transformations, Kai can write, optimize, and debug your transformation logic.
"When I need some bigger, more complex script... that's what I use it for."
-- Analytics Lead, Private Beta user
It understands your existing data model and coding patterns, creating transformations that fit seamlessly into your project.
When something breaks, Kai helps you fix it fast. It can analyze job failures, resolve configuration errors, investigate data quality issues, and debug API authentication problems.
Kai can generate comprehensive project documentation, explaining flows, transformations, and data dependencies, ready for onboarding or knowledge sharing.
"I wanted to create a mermaid diagram showing exactly where data flows from first table to last... I use it to generate markdown with the process description and the SQL script underneath."
-- Analytics Lead, Private Beta
It can also enrich your metadata layer: updating table and column descriptions, identifying foreign keys, and adding business-friendly context. Point it at a flow and ask it to document everything.
As one beta user put it:
"It's not just a chat window outside the platform, but it actually understands where I am and what I'm doing."
Kai is now available in public beta for all Keboola users. Whether you're exploring data, building integrations, debugging pipelines, or generating documentation - Kai is ready to help.
We made Kai free during public beta: Contracted customers: 150 messages per month per user, Pay-As-You-Go customers: 50 messages per month per user.
Already a Keboola user? Open Keboola, click on Kai, and start your first conversation.
New to Keboola? Start a free trial and see what Kai can do for your data workflows.
Thank you for your submission. You can access it here:
.png)
