n8n 2.46.6: Agent Platform with Visual Builder and Self-Hosting
Release 2.46.6 of the n8n platform came out on November 4, 2026. It is a visual builder for AI agents and workflows that can be deployed on your own server or in the cloud. Six releases have accumulated since the last review — from 2.45.5 to 2.46.6.
What's new
- npm installation fixed on all platforms — after the failure in previous versions, installation works again everywhere Node.js is available.
- Improved credential resolution when loading node parameters — fewer errors when the agent calls external services.
- Expressions now read named array keys — you can access data by name, not just by index.
- Fixed handling of Error objects in expressions — errors keep their name and fields when they cross the expression runtime boundary.
- Faster migration report — Slack users and credentials are loaded once per scan instead of for every record.
- Archived workflows hidden in the migration report — only active ones are shown by default, which makes navigation easier.
What the platform does
n8n combines a visual canvas with the ability to write your own code in JavaScript and Python. The platform supports multi-step AI workflows with logic, tool calls, human approval and full observability. Self-hosted deployment with role-based access and auditing is available.
Your own model instead of the cloud
You can connect your own model via the N8N_INSTANCE_AI_MODEL_URL environment variable by specifying the base URL of an OpenAI-compatible endpoint — for example, LM Studio. After that, requests to the model go to the specified address instead of the built-in provider. The platform is distributed under the Sustainable Use License: internal use and non-commercial distribution are permitted, commercial distribution — only free of charge and for non-commercial purposes.
How it differs from its neighbors
Compared with Sim, FastGPT and Activepieces, n8n has no features that only it offers. MCP support is claimed by most of its neighbors but is not mentioned in the n8n documentation. A visual editor and Docker deployment are available in all four platforms. For connecting your own model, n8n offers LM Studio and an OpenAI-compatible API — Activepieces has a wider list thanks to Ollama and LiteLLM.
| n8n | Sim | FastGPT | Activepieces | |
|---|---|---|---|---|
| Type | builder | builder | builder | builder |
| Language | TypeScript | TypeScript | TypeScript | TypeScript |
| License | Sustainable Use Licen… | Apache-2.0 | FastGPT Open Source L… | MIT Expat |
| GitHub stars | 209,583 | 29,886 | 29,886 | 25,179 |
| Releases in 90 days | 58 | 85 | 13 | 14 |
| Custom tools and functions | — | yes | — | — |
| MCP support | — | yes | — | yes |
| Agent memory | — | yes | — | — |
| Knowledge base (RAG) | — | yes | — | — |
| Sandboxed code execution | — | yes | — | — |
| Human approval | yes | — | — | yes |
| Graphs and state persistence | yes | yes | yes | — |
| Visual editor | yes | yes | yes | yes |
| Tracing and debugging | yes | yes | — | — |
| Own server and Docker | yes | yes | yes | yes |
| Images and documents | — | — | yes | — |
| Own model: Ollama | — | yes | yes | yes |
| Own model: vLLM | — | yes | — | — |
| Own model: LM Studio | yes | — | — | yes |
| Own model: Xinference | — | — | yes | — |
| Own model: LiteLLM | — | — | — | yes |
| Own model: OpenAI-compat. API | yes | — | yes | yes |
Comparison based on the projects' READMEs and documentation as of 11/05/2026: "yes" means the feature is claimed there, a dash means it is not mentioned (which does not mean the feature is absent). Stars and releases are per GitHub data.
Updating is worthwhile for those who installed n8n via npm and ran into installation errors, as well as for those who actively use expressions in workflows. The platform is worth a look if you need a self-hosted agent builder with a visual editor and the ability to connect your own model via an OpenAI-compatible API. You can follow news about neural networks on your own hardware in the CPU3D news feed.
