Your data warehouse, built by an AI agent in days
Tell an agent what you need. It connects the systems you already run, organizes the data and keeps it current in your own cloud. You approve every step.

You lead. The agent builds, and Core tells it what's next.
Core keeps a checklist of what the project needs next. The agent works through it, checks back after every step and stops for you whenever a decision or a password is needed. The same loop keeps the platform running after launch.
Building it
You
Say what you want
“Build us a warehouse and connect HubSpot.” The agent asks three quick questions: which cloud, which warehouse, what to call it.
Vecten Core
Lists the next step
A checklist of what the project needs now, in order, so the agent never has to guess.
AI agent
Builds and tests it
Connects the source, organizes the data and runs the tests.
You
Review and approve
You type the passwords and approve anything that touches the cloud. Every change can be read before it ships.
Your cloud
It's live
Data loads on a schedule and stays current.
Repeats for every new source and metric, back to “Lists the next step”.
Keeping it running
Vecten Core
Spots the problem
A source renamed a field or a sync failed overnight. Core lists the issue with its likely cause and fix.
AI agent
Prepares the fix
One fix per issue, tested before anyone looks at it.
You
Approve it
You read the change and approve it.
Your cloud
Back to normal
The fix ships and the issue clears.
Repeats whenever something changes, back to “Spots the problem”.
Say what you want
Point your agent at an empty directory and ask for a warehouse. It asks three questions (cloud, warehouse, name) and sets up a running platform, on your laptop first and then in your cloud account.
Ask for a source
Name a system and the agent connects it. When no ready-made connector exists, it writes one in the same loop, with the same schedule and test as every other source, and then models the data.
Review and approve
You type the credentials and approve anything that touches the cloud. The rest is the agent's work, and every change lands as a pull request you can read before it ships.
A warehouse you own. Built the same way every time.
Any system, even without a ready-made connector
Built in the same loop
When a system has no off-the-shelf connection, the agent writes one, with the same schedule and tests as every other source. Passwords are stored as secrets, never in the code.
Connectors (opens in a new tab)Know-how that ships with it
Measured in public
The agent works from built-in guides for each kind of job, from connecting a source to fixing a failure. We tested the agent with and without them and published the results.
Skill evaluations (opens in a new tab)Proven tools underneath
Nothing exotic
Core is built on widely used open-source tools that data teams already know, so your own engineers or any partner can work on it.
What it's built onOne way of organizing every source
Readable without a tour
Every source is organized the same way in every project, and tests check it. A new engineer or a new agent can find their way around a project they've never seen.
The schema convention (opens in a new tab)Your warehouse, in your cloud
Hosting cost shown up front
Core deploys to your own account with major cloud providers like AWS, on a database it sets up for you or the Snowflake you already have. Before anything is deployed, you see an estimate of the monthly hosting cost.
The data warehouse (opens in a new tab)Updates keep your changes
Yours to reshape
New releases merge into your project instead of overwriting it, so your edits stay. If an update and one of your edits collide, it stops and shows you where.
Updating a project (opens in a new tab)The everyday jobs. Without the queue.
Data work never ends: sources change, numbers turn out wrong, syncs fail. Here is what three everyday jobs look like with an agent doing the work.
Adding a source
Before
- Raise a ticket
- Wait in the queue
- Write the code
- Model the data
- Test and release
- Raise a ticket
- Wait in the queue
- Write the code
- Model the data
- Test and release
Days to weeks
With Vecten Core
- Ask the agent
- Check what arrived
- Approve
- Ask the agent
- Check what arrived
- Approve
Same day
Fixing a wrong number
Before
- Spot it
- Raise a ticket
- Change the definition
- Release
- Nobody tells the asker
- Spot it
- Raise a ticket
- Change the definition
- Release
- Nobody tells the asker
Weeks
With Vecten Core
- Flag it in your assistant
- Agent proposes a fix
- Approve
- Flag it in your assistant
- Agent proposes a fix
- Approve
Same day
Recovering a broken sync
Before
- Dig through logs
- Find the runbook
- Make the fixes
- Tell the team
- Dig through logs
- Find the runbook
- Make the fixes
- Tell the team
Hours of digging
With Vecten Core
- Core flags the cause
- Agent prepares the fix
- Approve
- Core flags the cause
- Agent prepares the fix
- Approve
Minutes of your time
Proven open-source tools. Try it on your laptop first.
You need Python 3.14 and Docker. No cloud account, domain or GitHub account is required. The commands below create a project with a HubSpot source and start Apache Airflow and the MCP on your machine.
uv tool install vecten
vecten init my-project --name my-project --users 5 --source hubspot
cd my-project
vecten upWith an agent, point Claude Code at an empty directory and say what you want. It asks the CLI what the project needs next and follows the items. When it is time for the cloud, see deploying to AWS (opens in a new tab).
- Warehouses
- Postgres 17, managed in your cloud, or your own Snowflake account. One warehouse per deployment.
- Clouds
- Major cloud providers like AWS, deployed with Terraform and run by GitHub Actions.
- Stack
- dlt loads each source, Apache Airflow runs it daily and rebuilds the models once new data has landed, and dbt turns raw tables into clean ones. Everything they run comes from your repository, not the Airflow interface.
- Not included
- An analyst dashboard, Kubernetes, built-in model inference and a connector catalogue.
- In the repository
- Core is the
platformcomponent and thevectencommand line.

Run it yourself, or let us run it for you.
Vecten Open is free and complete. It runs in your own cloud account, nothing passes through Vecten, and you don't need a partner or a paid plan. If you stop working with us, you keep everything: the code, the data, the definitions and the record of every question. When you'd rather not run it yourself, or need it built, connected and compliant, that's the work Vecten does.