We used Dify to create and test an agentic workflow and an AI agent model with some of the tools and RAG models. We used it to test how it works and how to implement it for part of our core product in our company.

We created an AI marketing agent using Dify, and the idea was that it can look into your marketing platforms, for example, Facebook, Google Ads, and Google Analytics. Those are the marketing platforms that we targeted, and instead of manually moving through dashboards, the idea was an agent will have access to your marketing data and can go through this data and provide you with reports and insights and suggestions. We used Dify's visual workflow builder to build this.

I tested Dify about six months ago for some of the tasks that we had for building some kind of a product, and I used it for two months and then did not use it very much after that.

The visual workflow builder Dify offers is really helpful, and you do not have to code everything. You can use it to connect nodes and make a flow of how your agent should work. Dify has RAG functionality that is also in-built, and those are the features that we used, and those two features were very good.

The visual workflow builder made my work easier because it saved us time since we did not have to code everything. One of the other interesting things was that it was really easy to show or present to a non-technical person. Our CEO was non-technical, so for him, it was really easy to show it as a diagram and explain how it works, and he could even do some edits. The ability for non-technical people to look into it is a really great use case.

Using Dify has positively impacted our organization because we were able to cut down on some development time and do a lot of testing in a very small time period. Initially, we had about two weeks of time to implement the whole thing, but that was cut down to two days of time through using Dify.

My personal experience with Dify's UI is that it is not my favorite, as it can be improved a little bit, and sometimes the UI feels a little bit buggy. I am not sure if that is because it was a self-hosted version. The documentation can also be improved a little bit more. I think not a lot of people are using Dify currently, so that is why the documentation is not very great. If the documentation was improved, that would also be a really good thing.

Currently, Dify could improve by offering better observability like other platforms. We currently use OpenAI Agents SDK, which requires you to build everything by code, but the observability is really good. It has OpenAI Traces, and you can basically trace everything for a conversation. If Dify had that kind of tracing functionality, that would be great.

I actually tested Dify about six months ago for some of the tasks that we had, it was for building some kind of a product, And, I used it for like two months and then, did not use it very much after that.

Dify is stable.

We used Dify only for the POC, so we did not expose it to a lot of workload. One of the main concerns that we had is that it might not be very scalable because we are hosting it in a self-hosted environment, and we have to configure the architecture and everything. Rather than using a cloud-hosted platform, using a self-hosted platform means there can be scalability issues. We anticipated there would be scalability issues, but we did not go for that scale. While testing, sometimes because of the limitations of the server, it crashed, stopped working, or got delayed, so it has a little bit of scalability problems.

We used Dify for testing out a POC and different ways of how to implement the agent, so there is no direct return on investment, as the investment was zero, so there is no return.

My experience with pricing, setup cost, and licensing is that it was free to use. We were able to get the free license from the GitHub release and then deploy it in our organization.

We evaluated another option called Chatbot Kit, but I am not very sure about that because I do not remember everything. We also used another product, either Chatbot Kit or ManyChat, and then after Dify, we switched back to OpenAI Agents SDK.

The visual workflow builder made my work easier because it saved us time since we did not have to code everything. One of the other interesting things was that it was really easy to show or present to a non-technical person. Our CEO was non-technical, so for him, it was really easy to show it as a diagram and explain how it works, and he could even do some edits. The ability for non-technical people to look into it is a really great use case.

Dify is self-hostable, so we did not have to pay anything. We just had to host it, and we really own the whole thing, and we can see the code as well. Self-hostability is another great feature.

If you are a startup or someone who is trying to run a POC related to agents, you should use Dify. It is a really good alternative that you can use to test things out and build a POC. After that, make sure to move to a different platform because if you need to scale it up and if you need custom steps, that is the advice I would have. I would rate this review a six out of ten.