Answer in brief: n8n is a strong candidate if you want self-hosted, visual workflow automation with AI-agent capabilities and hundreds of integrations, but version 2.35.3 is a prerelease published 2026-08-14; the current stable release from the maintainers is 2.34.6. Before production, review the fair-code license, test the connectors you depend on, and track known open issues around AI memory and individual nodes. This n8n 2.35.3 release review is a source and documentation review based on the GitHub repository, releases, and open issues as of 2026-08-16; no runtime testing was performed.

Key facts

What n8n is

The maintainers describe n8n as a “fair-code workflow automation platform with native AI capabilities” that combines visual building with custom code, runs self-hosted or in the cloud, and connects to 400+ integrations. The README goes further, claiming 1,500+ integrations, 9,000+ workflow templates, multi-step AI agents, model flexibility across OpenAI, Anthropic, Google, and open-source models, and JavaScript, Python, and npm support inside workflows.

Note the discrepancy: repository metadata says 400+ integrations while the README says 1,500+. Both come from the same project, so verify the connector list for the integrations you actually depend on rather than trusting either number.

What changed in 2.35.3

[email protected], published 2026-08-14, is marked as a prerelease by the maintainers. The changelog covers five bug fixes and one feature:

• core: allow the details field in continueErrorOutput mode (#36228)
• core: bound workflow publication outbox record processing with an abort deadline (#36286)
• editor: update ready-to-run demo templates to a current OpenAI model (#36201)
• Google Ads Node: migrate from the sunset v21 API to v25 (#36259)
• Microsoft Teams Node: restore the Group.ReadWrite.All OAuth2 scope (#36186)
• feature: skip update approval for workflows created in the same Instance AI session (#36106)

The stable 2.34.6 release published the same day includes a similar set of fixes, notably the Google Ads API migration and the Microsoft Teams scope restoration. The 1.x line is also still maintained; the maintainers published [email protected] on 2026-08-13.

The Google Ads change is the most operationally important item: it is a migration away from a sunset API version, the kind of forced update that breaks workflows if you stay on an old release. If you run Google Ads workflows, plan for this change rather than around it.

Who should consider n8n

Consider n8n if you need workflow automation or AI agents that touch your own systems and you want the workflows to live on infrastructure you control. That fits small businesses connecting CRM, billing, and support tools; developers who want a visual layer with JavaScript and Python escape hatches; and operations teams that must keep sensitive data inside their own environment instead of a managed iPaaS.

It is a weaker fit if you need a conventional open-source license, want zero maintenance burden, or expect every node to behave like a first-party SaaS integration. Adoption is high: 200,808 stars, 60,151 forks, and a push on 2026-08-16. High adoption does not remove the need to test your specific workflows.

What to check before production

License terms

GitHub classifies the repository license as “Other” with SPDX ID NOASSERTION. The README states n8n is fair-code, distributed under the Sustainable Use License and the n8n Enterprise License, with enterprise licenses available for additional features and support. Do not assume this is conventional open source; have the license text reviewed against how you plan to use and distribute n8n, especially if you resell services built on it.

Which release line you run

2.35.3 is a prerelease. If you deploy from the stable tag, you are on 2.34.6 as of 2026-08-14. Choose a line deliberately, test upgrades in staging, and verify migration notes for your data store before touching a production instance.

Connector health

The Google Ads v21-to-v25 migration is a reminder that connector code must track third-party APIs. Before production, check each connector you rely on against the release notes and test the full path, including OAuth scopes. The Microsoft Teams fix in this release is exactly the kind of regression that shows up in connected apps.

AI-specific risks

AI features are central to the README, but there are open issues around them. Issue #14361, open since April 2025 with 56 comments, reports that the AI Agent does not store tool usages in memory; the reporter says this can lead the model to reply as if a tool was called when it was not. If your workflows depend on agent memory, treat this as a risk to test and follow the issue for status.

Deployment and operations

The README quick start uses npx n8n or Docker on port 5678 with a data volume at /home/node/.n8n. That simplicity hides real operational work: backups, secret management, monitoring, and upgrade testing. The maintainers publish SBOM, VEX, and third-party license files with releases, which is a useful supply-chain signal, but you still own patching and monitoring in a self-hosted deployment.

Maintenance signals

The signals are mostly positive: frequent releases, an active default branch (last push 2026-08-16), parallel maintenance of the 1.x line, and SBOM/VEX artifacts on release tags. The counterweights are 1,301 open issues at snapshot time, and issues like #14361 that have stayed open for more than a year, which suggests some AI-adjacent behavior matures slowly.

Limitations

• 2.35.3 is a prerelease, not the stable line.
• The fair-code model is not conventional open source and needs legal review.
• README and repository metadata disagree on integration counts (1,500+ vs 400+).
• Open issues cover AI memory (tool usage not stored) and a Datatables Node reported not working (#20463, reopened with 38 comments).
• This review is source-based; no runtime or performance claims are made.

Evaluation questions and alternatives

Ask these before committing: Which connectors must work on day one, and are they current in the release notes? Do your workflows involve long-running processes, human approvals, or sensitive data that must stay in-house? Does the fair-code license conflict with your business model? Who owns the work when a connector API is deprecated?

Compare n8n against managed iPaaS options if you want someone else to own the platform, and against other self-hosted automation projects if licensing or footprint matters more than the AI feature set. The right comparison depends on your data-location requirements and your tolerance for maintaining a platform.

Verdict

n8n is worth evaluating for self-hosted workflow automation and AI agents, and the 2.35.3 prerelease confirms that the maintainers are actively refining AI workflow behavior and keeping connectors current. For production, stay on a stable release, complete the license review, test the connectors and AI-memory paths you depend on, and budget for ongoing upgrade work. n8n lowers the barrier to building automation, but someone still has to run it.

Sources

• n8n-io/n8n repository: repository metadata, README, and issue data
• [email protected] release: prerelease changelog and assets
• n8n stable release tag: 2.34.6 changelog
• [email protected] release: 1.x maintenance release
• Issue #14361: AI Agent doesn’t store tool usages in memory
• Issue #20463: Datatables Node not working
• n8n documentation: linked from the README

Where I can help

I help teams take automation platforms from evaluation to production: deployment and upgrade planning, workflow design, connector review, and the maintenance checks above. If n8n fits your stack, I can run through these checks with you before you commit to a rollout.