Ascend.io Migration: Moving Your Pipelines and Python

4 min read

By August 2026, Ascend.io had replaced its homepage with a short note, signed by Otto, its AI agent, and the team: after 10 years, the company is winding down operations. On 30 October, the list of acquisitions that the European Commission publishes under the Digital Markets Act showed where part of it is going. Apple has the right to hire certain Ascend employees, a non-exclusive licence to its intellectual property, and some of its patents.

The note thanks "the customers who ran production on us". As of 18 November 2026, neither the homepage nor the documentation gives them an end-of-service date or an export guide. Every Ascend customer now has a migration to plan without knowing the deadline. This guide is for them.

 

What happened to Ascend.io

Ascend was founded in 2015 by Sean Knapp and raised a $31 million Series B led by Tiger Global in April 2022. In May 2025 it launched what it called the first Agentic Data Engineering platform, and named Mattel, News Corp and UnitedHealth Group among its customers. The product kept shipping until at least 21 April 2026, the date of the last entry in its changelog.

Apple informed the Commission of the deal in June. The notice says nothing about Ascend's service or its customers. Read plainly, Apple acquired people and rights, and the platform stays behind.

 

Your Ascend.io migration options

An Ascend customer can take its pipelines to 3 places:

  • Another pipeline platform, including the declarative pipeline tools that the warehouse vendors now offer. The logic is rewritten in that platform's format, and the next vendor decision lands on the same pipelines.
  • The warehouse alone, with scheduled queries and stored procedures. It holds for simple flows, and the logic ends up scattered across the warehouse with little testing around it.
  • Code you own: Python pipelines in your own repository that process the data at ingestion, before it lands in the warehouse. They come with their tests and run under an orchestrator such as Airflow, Dagster or Prefect.

Ascend customers are closer to the third route than most teams leaving a platform. Ascend ran its queries on the customer's own data plane, whether Snowflake, BigQuery, Databricks or MotherDuck, so their data already sits in their warehouse, and their project already lives in a Git repository, usually their own. The output of every flow sits in that warehouse too, and it is the reference each rebuilt flow will be checked against.

 

Ascend to Python, component by component

In Ascend Where it goes What to watch
SQL and Python transforms Transformations applied at ingestion, generated from a sample of the source data Their code is not carried over. Review each suggested transformation, then check the result against the output the old transform produced
Read components Python ingestion code in the repository Each connection is rebuilt and tested against a real sample
Write components The pipeline's load step Keep the table names, and decide whether downstream queries still need the batch column Ascend added
Tests The pipeline's own test suite They run in CI, before every merge
Automations Airflow, Dagster or Prefect Schedules and event triggers are declared again
Processing state, run history, lineage A full reprocessing, then the orchestrator's run history and the pipeline's documentation Smart transforms lose their partition metadata
Otto and the other agents Code review and CI checks The monitoring work goes back to the team

 

First, secure what exists

Before the rebuild, 4 things cannot wait:

  • Ask in writing for the end-of-service date, and for an export of run history and lineage.
  • Freeze the repository: Clone it, keep a copy outside the working branches, and export run logs while the interface still works.
  • Save the documentation: Ascend's docs are still online and describe how each kind of component behaves. They are the best migration guide customers have, for as long as they stay up.
  • Rotate every credential Ascend held: Git deploy keys and warehouse service accounts stay valid after a vendor disappears.

 

How to migrate from Ascend.io without changing the data

Then work flow by flow, in 4 steps, the same method as for a Talend migration:

  1. Rank the flows by their users: Start with the outputs someone reads every day, and leave for last the ones nobody can name. A flow without a user is a candidate for switching off.
  2. Take a sample of what each flow ingests: A representative extract of the source data, edge cases included. The flow's current output, already in your warehouse, stays the reference.
  3. Rebuild it from that sample as a Python pipeline in your repository that processes the data at ingestion, with tests, documentation and an output contract that states what the flow produces.
  4. Run old and new in parallel: Point the new pipeline at separate tables, and compare row counts, keys, totals and rejected rows with the Ascend output for a few cycles. Switch once they match and someone signs for the result against a written bar for production-ready.

Ascend's design makes the last step cheaper than in most migrations. Both versions can load their output into the same warehouse, side by side, which keeps the comparison simple. Expect differences where an old transform joined other tables or applied a business rule that the source data does not show: the comparison is where they surface. Plan a full reprocessing for each smart transform, since its partition metadata stays behind. For a plain incremental flow, the latest timestamp already loaded gives the new pipeline its starting point. Count the sources during qualification, before anyone fixes a price: each one needs its own sample and its own review.

 

After Ascend: the question to ask every vendor

The useful question for any platform is easy to ask: if the service stopped tomorrow, which of our pipelines would still run? Go through it piece by piece, the code, the schedules, the state and the connections, and write down where each one lives. Whatever lives only in the vendor's runtime is the part you will rebuild under pressure. That is how lock-in builds up, one connector at a time, and the consolidation of the data stack will keep testing it, through acquisitions as much as through shutdowns.

For consultancies, the question applies to every project delivered on a platform. A client's pipelines inherit the fate of the tool they were built with, and the consultancy that chose the tool will be asked to fix it when it goes. Deliver code the client owns, run by an orchestrator it controls, and the next shutdown becomes a change of supplier.

Ascend spent 10 years on declarative pipelines and on automation that knows what has changed. Those ideas will outlive the company. The pipelines its customers rebuild this winter should keep them, in code their team owns.

 


PrettyWhale.AI is an Engineering AI for data pipelines. It generates the ingestion code for each source from a sample of its data, as Python in your repository, with tests, documentation and an output contract. See how we modernise pipelines.

 

You have any questions on PrettyWhale.ai ?

Contact us