n8n has published the beta of version 3.0, and the company says it plainly: there are no new features to celebrate, there is a deep cleanup that breaks things. Legacy nodes disappear, in-memory binary data mode is removed, the maximum time for the Code node drops from five minutes to one, and installation via npm or npx is no longer supported. The stable version will arrive in early December, and anyone managing an instance has until then to review what will break.
The reason for the version number jump is semantic versioning. Over the past year n8n has been accumulating incompatible changes it could not publish without bumping the major version, and now it bundles them all into one release. Since 2.0 in December 2025, more than 45 versions have shipped, roughly one per week, with the new features included. In other words: the product has kept moving forward, and this 3.0 deals with the foundations.
That approach has a practical reading worth understanding before upgrading. An automation tool is not a text editor: its workflows connect systems, move data and sometimes touch production. A behavior change in a node does not stay at home, it spreads to everything that depends on that automation. The company itself has made available to customers a report that does the audit work, and that is the part any technical lead should read before the list of new features.
What breaks and why?
The first block is about reliability. Manual executions now always run on workers, so the main instance stops doing that work. In-memory binary data disappears and everything moves to the file system or external storage such as S3, which behaves more predictably under load.
The second block is about limits. The Code node had a maximum time of five minutes, now reduced to one and configurable. Decompression limits also drop, with an explanation that does not hide the problem: protecting the instance from malicious compressed files, the well-known zip bombs.
The third block is the calendar. The Cron and Interval nodes are replaced by the Schedule Trigger, which relies on the platform's more robust scheduler. Anyone with workflows using the old ones will have to migrate them or accept that they stop working.
| Change |
Before |
In 3.0 |
| Manual executions |
On the main instance |
Always on workers |
| Binary data |
In-memory option |
Files or S3 only |
| Code node time |
5 minutes |
1 minute, configurable |
| Scheduling |
Cron and Interval nodes |
Schedule Trigger |
| Installation |
npm, npx or Docker |
Docker only |
| Community packages |
Available |
Disabled if not verified |
Security is no longer optional
The most relevant block for anyone managing third-party instances is the one on defaults. Encryption key rotation is now always active. Slack credentials require a signing secret, so no one can send fake requests impersonating the service. The SSRF blocklist grows when protection is enabled, and unverified community packages are disabled by default.
That last point will spark conversations in teams that relied on unofficial integrations. The decision is correct from a risk standpoint and annoying from an operational one, which is the usual balance in security.
Dropping support for npm and npx follows the same logic. The project leads explain that some new features depend on a container environment, and anyone who valued the quick one-line install has a script that reproduces the result. Kubernetes and Helm are not affected.
Any instance on version 2.39 or later has a Migration Report in the settings menu. The report lists the affected workflows and configurations, automatically fixes many of the problems by saving each change as a new version of the workflow, suggests an owner for each case, notifies that person by email and allows exporting a summary in CSV.
Two details make it genuinely useful. The first is that published flows keep working in version 2 while the migration is being reviewed, so nothing has to be switched off to get started.
The second is that it filters out flows that are not published or have not run for a long time, so they can be marked as discarded and you can focus on what matters.
The company's advice is to upgrade first to the latest 2.x to get the most recent version of the report. It is a one-hour step that saves a weekend.
Removing nodes, with caveats
The list of removed nodes is long: Function, Function Item, Item Lists, Cron, Interval, the old OpenAI node, earlier modes of AI agents, old versions of image editing and nodes for services that no longer exist. Most have a direct replacement and many migrate on their own.
Function becomes the Code node, Item Lists is split across Split Out, Aggregate, Sort, Limit, Remove Duplicates and Summarize, and that fragmentation is good news in the medium term: a node that did five different things becomes pieces that can be tested separately.
There are also retired features with no exact equivalent. Chat Hub is disabled by default, importing from URL disappears from the editor and the AI query tab is gone from the Code node. Anyone who relied on them will have to look for an alternative outside the platform.
This kind of cleanup is similar to what the Python community did when it removed its experimental compiler: shedding weight in order to move forward, as we reported when we analyzed the removal of CPython's JITIn both cases the message is the same: keeping old code has a cost that someone ends up paying.