Numbers Station Labs releases Meadow—the simplest agentic framework built for complete data workflows.

How Numbers Station Built a Modern Data Pipeline via a Multi-Agent Architecture

Born from cutting-edge research, built for real-world problems.


AI HAS changed the world.

Start with the LLM.

Bring in your own or use ours. Numbers Station uses LLMs to build a user-oriented data pipeline instead of an IT-focused pipeline.
But LLMs are becoming commoditized and it takes more than just an LLM to get a high degree of accuracy and usability, especially when the users aren’t technical.


KNOWLEDGE specific to your business.

Add business knowledge.

The semantic catalog is populated during initial implementation and automatically captures key metadata about your business — metrics, dimensions, filters, acronyms, common join paths, and more.
Numbers Station continuously learns with more questions — making it smarter, more usable, and more accurate as your business changes and grows.
Leverage this knowledge across the data pipeline to supercharge your data strategy.



Orchestrate a multi-agent framework.

Multiple, specially-trained agents are much more accurate and effective than one monolithic model.
Add functionality over time without impacting the accuracy and performance of other agents. Build agents specific to your business. Numbers Station grows as you do. Integrate AI seamlessly with existing tools in your stack.
It’s all possible with Numbers Station.



Protect users with guardrails.

Users should be empowered with data, but with power comes risk.
Guardrails protect users from imprecise questions and invalid metrics, directing them to what they need when they need it while clearly identifying AI-generated metrics from validated metrics.


new metrics based on your behavior.

Build common data workflows.

We add powerful workflows to keep analysts in the loop, letting you align with your governance strategy. Promote metrics and formulas, deprecate them, and give continuous feedback to the AI.
These “AI-generated” metrics will be flagged as such until a human analyst verifies its validity.


FULL-STACK SOLUTION BEYOND conversational analytics.

Expand to the entire data stack.

As Numbers Station gains knowledge about your business, it can be applied to other parts of your data stack like data quality, schema creation, and entity matching.
As Numbers Station improves its knowledge of your business and data, everything else benefits too.




Numbers Station understands that every business is different, and we provide flexible deployment options.
Whether in our cloud or yours, using our interface or designing our capability into your product (via our API) — it’s all possible with Numbers Station.


Still have questions?

LLMs enable the Numbers Station experience, but our research has found that better LLMs aren’t the key to accuracy in analytics. Numbers Station is LLM agnostic, and can deploy with public models like GPT, Llama-3 or our private foundation model for local deployment.

Implementation time depends on the complexity of the use case and which functionality is being implemented, but Numbers Station can usually be implemented in a couple of weeks including the initial training time.

For conversational analytics, Numbers Station is used most often by non-technical end users who need data to make decisions.
Other solutions like data quality are used by people in multiple roles, including both business users who need to understand a new dataset or IT workers trying to streamline the data pipeline.

Numbers Station is a horizontal solution with customers in all major verticals. Each implementation involves custom training, so the AI will learn your business in the same way a new employee would. The foundation model at the heart of the platform already has deep cross-industry knowledge, and it will continue to learn as it’s used.

In some cases fine tuning helps achieve better results. Fine tuning is available to all Enterprise customers.

Numbers Station offers custom agents as part of an Enterprise license. Often these are tool agents to enable Numbers Station to access external platforms for tighter, AI-driven data integration.

While our semantic catalog is one of the keys to high accuracy and usability, Numbers Station is not a semantic catalog company. You can leverage your existing investment in a semantic catalog solution to jump-start Numbers Station’s training.

Numbers Station requires very little administration. During the first week of use it’s typical to set aside time to give the platform feedback and show it how to answer some questions. After the initial training period, administration is generally limited to an hour or two a month to give continuous feedback.

Hear from our researchers.

With a decade of experience in machine learning research from institutions Stanford and the University of Washington, our team is committed to unraveling the mysteries of data and leveraging artificial intelligence to extract valuable insights.