Skip to content
[ docs · beta ]

Known limitations of the beta

Dexaflow Cloud documentation · Updated: November 18, 2026

  1. Getting started
  2. Connect your AI agent (MCP)
  3. Plans and limits
  4. Known limitations

Dexaflow Cloud is in a closed, invite-only beta. It is free, and it comes with the limits below. We list them so you can plan around them; most of them go away before general availability.

Service

  • No SLA. The beta is offered as is. We watch it and fix problems quickly, but there is no uptime or response-time commitment yet.
  • A limited number of organizations. We open places in small waves. If you need another organization, write to us.
  • One region. Everything runs in Google Cloud us-east1. Your data and runs stay there.
  • Maintenance without notice. We may restart the engine to ship fixes; a run that is in progress may take longer than usual.

Plan

  • Free plan only: 5 active DAGs, 20 tasks, 1,000 task runs a month, 2 tasks running at once, and no schedule more frequent than once an hour. See Plans and limits.
  • Small tasks. A task gets 0.5 vCPU and 1 GiB by default, 1 vCPU and 2 GiB at most, and runs for 30 minutes at most. Heavy jobs belong in your warehouse or another service that the task calls.

Building and running DAGs

  • The first Check after a quiet period is slow. The code checker starts on demand, so the first Check after some idle time can take about two minutes. The next ones take seconds.
  • A fixed runtime. Tasks run on our Python 3.11 data runtime: the Airflow providers for Postgres, MySQL, AWS, Google, Databricks, Snowflake and HTTP, their drivers, dbt, pandas, numpy and pyarrow. You cannot add packages or bring your own image yet; the Check tells you when an import is not available.
  • Outbound traffic on ports 80 and 443, and the usual database ports. Your tasks cannot reach other ports on the internet, nor private networks.
  • After an engine upgrade, deploy older DAGs again. A DAG deployed before an upgrade keeps the old runtime until you run Check and Deploy again.
  • New DAGs start paused. Unpause a DAG in the engine to start its schedule.

Security

  • Connections and variables are shared inside your organization. Every DAG of the organization can read every connection and variable of that organization, as in Airflow. They are encrypted at rest and never visible to another organization. Per-DAG scoping comes later.
  • Work email only. Sign-in needs a corporate email address; personal providers are not accepted.

AI agents (MCP)

  • No list of connected AI clients yet. You can create and revoke your own MCP tokens, but the console does not yet list the AI apps connected through sign-in (OAuth). To cut one off, remove its access in the AI app.

Help

Write to [email protected] with your organization's name.

© 2026
Dexaflow documentation GitHub Beta terms Privacy Beta limits

DexaData · CNPJ 28.918.992/0001-50

Apache, Apache Airflow and the Airflow logo are registered trademarks of the Apache Software Foundation. Dexaflow is not affiliated with or endorsed by the ASF.