● data & ai infrastructure

Describe the pipeline.
AI ships it to production.

DexaData delivers data and AI infrastructure as a service. Dexaflow, our first product, is the Apache Airflow experience built with AI: the agent writes the DAG, tests it, deploys it and watches every run.

Join the waitlist View on GitHub

works with the stack you already run

  • dbt
  • DuckDB
  • Snowflake
  • Databricks
  • BigQuery
  • Google Cloud
  • AWS
● dexaflow ai · coming soon

The Airflow experience,
built with AI.

You say what the data needs to do. Dexaflow handles the rest: code, tests, deploy and operations. It is still a Python DAG your team reads, reviews and versions.

  1. 01

    Describe

    In plain language, in the chat or in your editor through MCP.

  2. 02

    Generate

    The agent reads your connections and writes the DAG and the dexaflow.yaml.

  3. 03

    Test

    Every change runs in a sandbox, on sample data, before it reaches production.

  4. 04

    Ship and observe

    Automatic deploys, versions and rollback. The agent follows every run.

● NATIVE DBT

Native dbt, on the warehouse you already use

Point to your dbt project. Each model becomes an Airflow task, with dependencies, tests and lineage on one screen. Dexaflow installs the adapter, builds profiles.yml from your connections and keeps the credentials.

  • Databricks
  • BigQuery
  • Snowflake
  • DuckDB
# dexaflow.yaml
dag_id: sales_daily
dbt:
  project: ./analytics
  target: bigquery   # databricks | snowflake | duckdb
  select: daily_revenue+
  tests: true
● AI BUILDER

DAGs from a single sentence

The agent knows your environment: connections, variables and the DAGs you already have. It writes idiomatic Airflow code with tests, not a draft for you to fix.

● AUTO DEPLOY

Automatic deploys, no pipeline to build

From a prompt, a git push or MCP straight to production. Dexaflow validates, builds the DAG image, ships a version and rolls back in one click.

● SELF-HEALING

Failures explained and fixed

When a task fails, the agent reads the logs, explains the cause in one sentence and proposes the fix as a diff. You approve and it ships.

● MIGRATION

AI-guided Airflow migration

Point it at your DAG repository. The agent reviews each one, adjusts what does not run on Dexaflow and shows you the plan before migrating.

● MCP SERVER

Your agent runs the pipelines

A native Model Context Protocol server. Claude, Cursor or any compatible agent reads the context, creates pipelines, triggers runs and follows the logs.

● DEXAFLOW.YAML

Declarative pipelines, no Dockerfile

One short file describes the DAG, the Python version and the dependencies. It is the format the agent writes and you review.

# dexaflow.yaml
dag_id: sales_daily
python_version: "3.12"
dependencies:
  - pandas==2.2.3
  - google-cloud-bigquery

You stay in control. The agent proposes, you review the diff and decide what goes to production. The code is yours, in Python, in your repository.

● the engine · apache airflow compatible

//Dexaflow

Airflow compatible.
Built for production.

The same UI and REST API as Apache Airflow, with a Go control plane underneath. Your DAGs stay in Python. What changes is what happens when they run.

01_ Airflow compatible

The unmodified Apache Airflow 3.2 UI and the same REST API. DAGs, tasks, variables and connections, in the vocabulary your team already knows.

02_ Go control plane

Scheduling and coordination outside the Python loop. No scheduler stalling between tasks and no re-parsing every DAG on every cycle.

03_ One task, one pod

Every task runs in its own pod. No long-lived workers piling up memory and connections until they die overnight.

04_ Dependencies per DAG

Each DAG gets its own Python environment. pandas 1 in the legacy DAG and pandas 2 in the new one live side by side.

05_ Map-reduce for ML and AI

Fan-out and reduce written as a Python list comprehension. No XCom plumbing, no broker, no special operator.

06_ From laptop to cluster

The same engine runs on a single host, no Kubernetes, or with Helm on your cluster. Start small and rewrite nothing to grow.

● open core

Open core.
Managed cloud.

The engine is open source and runs wherever you want. Dexaflow Cloud is the same engine, run by us, with AI built in.

Dexaflow OSS

Apache 2.0

For teams that want to run it on their own infrastructure.

  • The full engine, compatible with the Airflow UI and API
  • Lite on a single host or Pro with Helm on your Kubernetes
  • An MCP server to use with your own agent
  • You operate it: upgrades, database, security and scale
Code on GitHub Read the docs →

Dexaflow Cloud

coming soon

For teams that want pipelines in production, not a cluster.

  • Everything in OSS, run and upgraded by DexaData
  • Built-in AI builder: generate, test and ship with nothing to set up
  • AI-assisted failure fixes and Airflow migration
  • Corporate login, teams and plans starting at Free
See pricing

Dexaflow OSS is the new name for Leoflow. The code is open at github.com/dexadata/dexaflow.

● dexaflow cloud · coming soon

Managed Dexaflow,
no cluster to babysit.

Capacity measured in DCUs, clear limits per plan and no cluster to look after. Start on Free and move up when your volume asks for it.

● what is a DCU

Capacity measured in DCUs, the Dexa Compute Unit.

Each plan sets how many DCUs run at the same time. You pick the size of each task in dexaflow.yaml; if you say nothing, a task uses 1 DCU. When work goes over the limit, extra tasks wait and run as soon as there is room, without failing.

Example: with 6 DCUs on Small, you run 6 tasks of 1 DCU in parallel, or 3 tasks of 2 DCUs.

Free

Free

To try it out with real pipelines.

Concurrent DCUs
2
Task runs / month
1k
Active DAGs
5
Total tasks
20
Minimum interval
1 h
Users
1
Log retention
7 days
Join the waitlist

Small

for teams getting started

$50/month

For the first data team in production.

Concurrent DCUs
6
Task runs / month
25k
Active DAGs
30
Total tasks
150
Minimum interval
15 min
Users
3
Log retention
14 days
Join the waitlist

Medium

$250/month

For several teams and minute-level schedules.

Concurrent DCUs
20
Task runs / month
250k
Active DAGs
150
Total tasks
1.5k
Minimum interval
1 min
Users
10
Log retention
90 days
Join the waitlist

Large

$900/month

For the whole company's data platform.

Concurrent DCUs
80
Task runs / month
2M
Active DAGs
500
Total tasks
10k
Minimum interval
1 min
Users
Unlimited
Log retention
1 year or more
Join the waitlist
  • Every plan includes the MCP Server, dexaflow.yaml pipelines and the Airflow-compatible UI.
  • On Free, schedules with no dashboard or MCP access for 30 days are paused. We email you first.
  • Prices in US dollars, per month. Prices and limits may change before launch.

Enterprise

A dedicated environment, custom limits and support with an SLA. Or run Dexaflow OSS on your own cluster.

Talk to us →
● about dexadata

DexaData builds the data and AI infrastructure teams use as a service, without assembling a platform. Dexaflow is the first product. The next ones follow the same rule: an open core, AI in the workflow and operations on us.

● get in touch

Let's get your
pipelines running.

Want early access to Dexaflow Cloud, to migrate DAGs from Airflow or to run Dexaflow OSS at your company? Tell us about your setup and we will get back quickly.

email [email protected]
● OPEN SOURCE

Open source under the Apache 2.0 license. No lock-in.

● AIRFLOW COMPATIBLE

Your DAGs and the Airflow vocabulary still apply.

● AI NATIVE

AI proposes, you approve. The code stays yours.