The data layer for AI
Give your AI one connection
to all your data and every system you run.
See what you get One MCP server you control
Connect your CRMs, databases, warehouses, apps and files once. They all sit behind one MCP server, so Claude, ChatGPT Business, your own agent or a self-hosted model can ask one question across every system you connect, and the platform combines what they return. Connect a new source and every AI you use can ask about it, with nothing new to set up.
One federated MCP server
One connection for your AI, not one per app
An MCP server per app hands your AI that app's raw records in its own field names and leaves the combining to the model. Here every source you connect sits behind one MCP server, so one question can reach your CRM, your databases and your files, and the platform combines what they return. Connect a new source and every AI you use can ask about it, with nothing new to set up in the AI client.
Source changes
It keeps up when your systems change
When a source changes, renamed columns and moved tables the platform recognizes are followed, and a removed column's setup is kept in case it comes back. Your AI keeps asking in the same business terms.
Universal schema: the same understanding across every system
Each system is understood in the same common terms. You review what the platform is unsure of, and if two tables have no link, your AI is told rather than left guessing.
Universal schema and data relationships
The same customer, matched across every system
Get one row and one total for each customer, supplier or product, wherever the platform has matched its spellings across your systems, without building the matching yourself. Map each system once into shared business terms, which harmonizes your data end to end, and review what the platform is unsure of. It also looks for how your tables relate and links what it finds, so one question can span them. If two tables have no link, your AI is told rather than left guessing.
Reduced AI cost and improved performance
An AI model sounds just as sure when it is wrong. Here the matching and the arithmetic are done before your AI reads a result, and when a figure leaves something out, or a question cannot be answered, the result says so and your AI is told to pass that on.
Lower LLM cost
Lower your LLM bill
Your AI works from matched, added-up results instead of raw tables, so you can use a low-cost model on a low reasoning setting instead of paying for a more expensive one. Most of each prompt can also come from the model provider's cache. And a question your team asks every week can become a dashboard, built once: opening it calls no AI model of yours, so it adds nothing to your LLM bill however often it is opened.
Your choice of model
Works with frontier, open-source and self-hosted models
Use a frontier model such as Claude or ChatGPT Business, an open-source model, or one you host on your own hardware: any model that can call tools, through an app that supports MCP. The matching and the arithmetic are done before a model reads a result, so you can choose a lower-cost or self-hosted model where it meets your needs. Switch models when you like: nothing ties you to one AI vendor.
A model needs a context window large enough for the results it reads: see "Which AI models work? Can I run my own?"
Dashboards for routine questions
Save the answer to a question your team asks every week as a dashboard. Opening it again calls no AI model of yours, so it adds nothing to your LLM bill, and view-only seats let the rest of your team open it.
Built-in dashboards
Routine questions: build a dashboard once, and pay no AI cost to view it
A question your team asks every week does not need an AI model every time. Save the answer as a chart once, built in the app or saved by your AI if you allow it. Opening it again runs the query against your data with no call to your AI model, so it adds nothing to your LLM bill, and the platform's own AI work is included in your fees. Charts use the same business terms and matched names your AI answers from, and view-only seats let the rest of your team open them.
A saved chart combines a few sources at a time, one series each. For one total across every source, ask your AI.
Data migration
Move data once, without a standing pipeline
When you do need a copy, send it to another database or download it as a file, mapped or not. You pay per job, with no pipeline to keep running.
A connection whose mapping is largely paid for has no job fee.
Coming next
Connect your own BI tool
Point Power BI, Tableau or Excel at the same layer your AI and dashboards read. In development.
Bring dashboards over from other tools
Rebuild Power BI, Tableau and Looker dashboards here. In development. Today the platform can translate the formulas in Tableau, Looker and SSRS files for you to review.
Federated data you own, protect and secure
Security comes from pass-through data: each question reads what it needs from your systems when it is asked, and the result passes back through us to you or your AI, with no pipeline and no warehouse copy to protect. Records in a connected database, warehouse or app stay there, a read-only login is all a database needs, credentials are kept encrypted, and you decide who, and which AI, can read each source.
Security
No warehouse copy of your data to secure
Records in a connected database, warehouse or app stay there. What a question reads passes through us to the answer instead of being loaded into a warehouse, so there is no second copy to protect, back up and put through every security review. Answers are cached for a few minutes at most, a read-only login is all a database needs, and credentials are kept encrypted.
We do keep a limited set of reference data, which can include values from your data, along with files you upload. What we keep.
Access control
You decide who, and which AI, can read each source
Admins can limit each person to chosen sources, and the limit holds for that person's AI as well. AI access reads only: no AI tool writes to your systems, and an AI creates or deletes dashboards only if you allow it. Each AI connection has its own token, which can be revoked at any time.
A dashboard is shared by group, so one built on another source stays readable to that group.
No pipeline to build
Start asking without a pipeline project
Connect a source, let the platform map it, and ask. You approve what it holds for review and can check the rest at any time. With no row syncs to schedule or watch, your team spends its time using the answers.
Already use Fivetran or Airbyte? Keep them for what they load, and skip another connector, sync and row-based bill for the systems you never loaded: here the bill follows the columns you map, not the rows you read.
Data freshness
Answers from today's data, not the last sync
Each question reads your systems when you ask it, so an answer is at most a few minutes old, and often seconds old.
Uploaded files are as current as the last upload, and a warehouse as its last load.
Building AI assistants or agents on company data?
Give your AI agents one MCP server for the systems you connect, with names matched and numbers added up before they read a result.
Want AI on company data without a data project?
Connect your systems and ask from the AI you choose, on a low-cost model if you like.
Already run a pipeline and a warehouse?
Keep them, and ask across the warehouse and the systems you never loaded, in one set of business terms.