n8n has more stars than Dify.

[foreveryuya]

Self Checks

• I have searched for existing issues search for existing issues, including closed ones.
• I confirm that I am using English to submit this report (我已阅读并同意 Language Policy).
• [FOR CHINESE USERS] 请务必使用英文提交 Issue，否则会被关闭。谢谢！:)
• Please do not modify this template :) and fill in all the required fields.

Content

I've only used n8n for now and just installed Dify. Is there anyone who has used both? How do they compare?

[DavideDelbianco]

I use both and this is only my personal opinion:

The good in Dify

Dify: has observability with 5-6 different tracing tools and Open Telemetry data
n8n: does not have observability and to get access to Open Telemetry data you need to pay enterprise
Notes: An essential feature to help you debug and monitor LLM that are usually black boxes

---

Dify: has streaming output from LLM nodes
n8n: does not support streaming output
Notes: An essential feature for chatbots where the user is used to the typewriter effect of ChatGPT

---

Dify: has logs of every call request
n8n: has logs only for the last 24h and only if you register the free license, otherwise you need to pay Enterprise
Notes: An essential feature for developers, because you may miss important bugs if the data is just not there when you have to inspect it

---

Dify: has a workflow board can automatically organize nodes, and can be dragged by mouse click and middle mouse click
n8n: does not have an automatic nodes organization
Notes: The nodes organizer is not essential, but when you tinker with the workflow you may move nodes around quite a lot, it's a nice to have feature

---

Dify: has a nice looking final-user front-end chat that can be embedded too
n8n: has an horrible final-user chat UI (to be fair they promised this year will be centered on UI improvements, I hope for them this includes the chat too and not just their main site dark theme)

---

Dify: has easy integration with lots of vector databases, especially weaviate which is small, fast, and incredibly scalable
n8n: has integration with few vector databases and often relies on third party nodes, after 2 years they still depends on the weaviate team to develop for them the custom node to talk to that famous vector DB, the other few free-choices are Qdrant and Supabase that are not very scalable. (they may have added more since last time I've checked)

---

Dify: has dedicated RAG section that, hopefully, will be expanded to support graph and other chunking methods (they are developing semantic chunking right now), the downside is that to fully manage the KB you have to use API that are poorly documented.
It does not have a node that allows you to easily index documents from workflows (there's a third party plugin written by a dify developer, but it lacks features that would be achievable only with a native node)
n8n: RAG is handled only by nodes in workflow, it is "external" to n8n, you have to setup a Vector DB and interact with it using nodes, the chunking options are better though..
Notes: RAG is essential to any LLM application, I strongly prefer to have it as a native feature as it is in Dify

---

Dify: has LLM key load balancing, but only if you have a paid license
n8n: does not have key load balancing
Notes: None is a clear winner when using Community Edition, and you have to rely on third party tools like LiteLLM for both if you want key load-balancing without paying a license.

---

Dify: has more LLM connector
n8n: has fewer LLM connectors
Notes: None is a winner, since most connectors in dify could be used as OpenAI compatible ones, something you can do in both applications, it just makes it easier to setup LLM (to be honest 90% of users rely on Open AI, Google, Deepseek, Qwen, it doesnt really matter much)

---

Dify: has an open-source license for the Community Edition, Enterprise license does not have "crazy" prices
n8n: has a custom license invented by them, that essentially does not allow you to use n8n for anything that is not internal automation tools, even using it for chatbots based on RAG is a "gray topic", also their Enterprise license for commercial use is "just" 50k / year
Without that license anything you do can potentially infringe their custom-made license, be careful.

---

The good in n8n

Dify: the code node, supports the addition of third party libraries only for python, NodeJS packages are not supported
n8n: the code node supports the addition of packages from npm for nodejs too
Notes: I have a preference for NodeJS because the editor support code formatting with alt+shift+F, and the code execution is not based on code indentation like python, It would be nice to be able to use npm packages too

---

Dify: has fewer native "logic" nodes
n8n: has nodes like Merge, Split, Switch, etc..
Notes: The "variable aggregator" node in Dify is not a merge node, to achieve simple merge, split etc.. you have to use a code node and in a low-code / no-tool like those, this should be supported with native nodes.

---

Dify: code node executes javascript code way slower than pyhon code
n8n: code node executes javascript fast
Notes: In Dify I am often forced to prefer python in code nodes to speed-up workflow execution.
Sometimes I have to prefer the template node to other nodes because executing an IF/ELSE in template node is faster than using the IF/ELSE node, workflows will generally run a little bit faster in n8n IMHO, the dify-sandbox could be improved

---

Dify: workflows can be triggered only by API or chat FrontEnd, to trigger them with webhooks, or on a schedule you have to rely on partially developed third party plugins
n8n: workflows can have multiple "entrypoints" and you can take different workflow branches based on the entrypoint
Notes: Dify is developing right now a native "trigger" other than the third party "cron" plugin and it will probably catch-up quite soon, lacks webhook invocation that can be worked around with plugins and "endpoints" i would give the point to n8n until we have "scheduled triggers"

---

Dify: nodes cannot link backward and create a cycle
n8n: allow nodes to link to ancestors and create loops (with exit condition)
Notes: I think this is useful to take advantage of LLM thinking capabilities, and to work on paginated outputs. You can partially achieve the same result using the Loop node, but it's not the exact same thing and partially complicates the workflow

---

Dify: from 1.5.0 you can use "last-run" values while debugging workflow
n8n: you can do the same, with a different UI experience
Notes: None is a winner, just something to know if you self host and are behind

---

Dify: does not support dynamic credentials linking in LLM nodes, you have to setup a key in LLM configuration and that's it
n8n: keys and credentials are managed separately from LLM configuration, in Enterprise you can also use external secret storage, allows you to choose the credentials to use in workflow LLM node.
Notes: To be honest in Dify, you may add a specific model multiple times with different keys, but because of the lack of model name alias, and a bug in the model dropdown selector it is really hard to bind the correct key-model pair in a LLM node, and you cannot do it dynamically like you can in n8n that is a little bit on the winning side here

---

Dify: has fewer integrations, by the use of plugins, many of the plugins are just duplicates that do the same thing and they are very often documented with Chinese images, the good thing is Dify has a "marketplace" that makes it easy to search them
n8n: has a lot more integrations for almost anything, many of them are natively supported by the team, the third party plugins are hosted on npmjs and you may need quite some time to find the right tool for the job
Notes: I think n8n wins because of the sheer amount of team-maintained high quality integrations

---

Dify: team-maintained plugins for communication are centered on Chinese apps (maybe because the team is predominantly Chinese),
n8n: team-maintained plugins for communication are more for worldwide audience (team is German?)
Notes: You can find WeChat plugins in Dify, but not Telegram and WhatsApp business (Telegram was there, but they removed it for some unknown reasons) it is something to keep in mind since nowadays, most people are used to chat to AI bots using chat applications that they already have in their smartphone. Until Dify is on-par with those plugins and start to really enforce worldwide documentation of plugins, the winner for me is n8n

---

Dify: White labeling is for Enterprise only
n8n: Extensive white labeling
Notes: n8n is a winner ONLY if you want to remove their logo from their frontend application. Without an Enterprise license, you cannot allow access to the white-labelled editor.

---

Dify: has dynamic fields binding based on just output variables
n8n: has a much powerful binding system based on schema output that can work really great especially after executing the previous node to get its output
Notes: Their main editor in the center of the page is hardly usable especially for prompts because they use very small text-areas, from a mere UI point of view Dify is better for the prompt engineering tasks, but better binding system gives the point to n8n

---

Dify: allows to convert workflows into tools to use them in other workflows, but this breaks inner "tool" observability (it also seems to me like it adds a little-bit of latency) when you update a workflow after publishing it you have to remember to publish the tool version too
n8n: allows the invocation of other workflows thanks to their "entrypoint" trigger mechanics and allows you to follow the execution flow in the other tool editor instead of disappearing
Notes: In Dify you can inspect the execution log in the original workflow or in external tracing tool (this means you have some means to make your life easier) You may consider the need to hit the "configure tool" after already "publishing" the workflow a plus, but to me it's something that should be improved (dify team easily do that)
I consider n8n the winner, because thanks to the "invoke other workflow" mechanism, you can easily export workflows, whereas in Dify you cannot export the tools and you have to regenerated them from scratch when importing the workflow somewhere else.

---

TLDR

In the end, I still use both, but Dify has a better licensing, has observability out of the box, has better logs for debugging and for the tinkering and developing part, this is something extraordinary valuable