Do you need to know how to code to use Dify?

Not for the essentials. The visual canvas lets you wire up models, knowledge bases, and tools without a line of code. The learning curve starts when you want to call an external API that isn't in the catalog: then you need to write a bit of Python in a code node.

Does my client data leave my infrastructure?

In the self-hosted Community version, no: everything stays on your servers, and you choose which model to call. In the Cloud version, data passes through Dify's servers (hosted in the United States and other countries) and through the provider of the model you use. For full confidentiality, prefer self-hosting with a local model.

Is Dify really open source?

The Community version is published under the Dify Open Source License, a license based on Apache 2.0 with additional conditions. You can use it, modify it, and install it on your own servers freely. The Cloud and Enterprise versions are commercial services built on the same code, with extra features. Redistribution or resale is governed by the license's additional terms.

How much does Dify cost for a five-person team?

The Professional tier is $590 per workspace per year, including 3 members. For five people, you need to move up to the Team tier at $1,590 a year (up to 50 members). Add the cost of the model provider once the message credits run out: for example, an OpenAI API key billed by usage.

What's the difference with n8n?

n8n automates flows between existing applications. Dify builds AI applications around a model: a chatbot fed by your documents, an agent able to call tools, a RAG pipeline. Both are open source and self-hostable, but they solve different problems.

Dify: our review

Dify is an AI application development platform, usable in the cloud or installed on your own servers. It lets you build agents, workflows, and knowledge bases without coding, while keeping control of the model and the infrastructure that process your data.

Cloud or self-hostedFree version, then paid·from $590/year

Go to Dify's website

What we take away from it

What counts in its favor

A free Community version installable on your own servers: business data never passes through a third-party cloud, and you choose which model processes it.
The publisher doesn't reuse data to train models, and its contracts with its subprocessors (OpenAI, Anthropic, etc.) also forbid them from doing so.
The visual interface for building workflows and agents lets you prototype an application in a few hours, without a developer.

What counts against it

Dify's cloud hosts data across several countries with no European-residency guarantee: the United States, Australia, Canada, China. The self-hosted version is the only way to control location.
Message credits included in the cloud tiers cover little use: 5,000 a month on Professional, after which you must bring your own API key from a model provider, adding a separate bill.
The publisher is a US company (Delaware), despite French roots: the governing law is California's, not French GDPR.

The best open-source platform for building AI agents without depending on a third-party cloud. Use it self-hosted if European data location is a requirement.

The facts

First paid tier: $590/year — Professional, per workspace
Data hosting: Cloud or self-hosted · source
Categories: AI agents,RAG and knowledge bases
Data verified on: 3 August 2026

What happens to your data

Is your data used to train their models?
Subprocessors: The cloud platform delegates inference to the model providers the user chooses. Dify states that its contracts with these providers forbid them from using the data to train their models. The list of subprocessors is published on the site.
Retention: Data is kept as long as the account is active. The publisher reserves the right to keep it beyond that if the law requires, or to anonymize it. Self-hosted, the retention period depends on your own configuration.
Data processing agreement: Document published by the publisher

Dify in detail

Dify is a platform that lets you build artificial-intelligence applications without writing a line of code, then deploy them on your own servers. Where a tool like ChatGPT remains a service you consume, Dify is a workshop: you assemble models, documents, tools, and business logic there to produce a custom agent or application.

What it actually does

Dify is built around three blocks. The first is the workflow studio: a visual canvas where you wire a language model to data sources, API calls, and conditions. The result can be a chatbot that answers from a document base, an agent able to send an email after analyzing a document, or a multi-step content-generation pipeline.

The second block is the knowledge pipeline (RAG): you import your documents — PDF, Word, web pages — and Dify indexes them so the model can search them for answers. This is the mechanism that turns a general-purpose model into an expert on your own document base, with no fine-tuning needed.

The third block is the marketplace: ready-to-use integrations with about fifty tools and models, from Slack to Notion, as well as OpenAI, Anthropic, and DeepSeek APIs. You bring your own API key and Dify orchestrates the calls.

Dify's strength is its two-speed deployment model: a turnkey cloud to prototype in minutes, and a Community version you install on your own servers with Docker to go into production without the data leaving your infrastructure.

Why it interests us (or not)

For an SMB or a firm handling confidential documents, Dify checks two boxes few platforms combine. First, self-hosting: the Community version installs on an internal server and the data doesn't leave it, except for calls to the model you choose. Second, the no-training commitment: the vendor explicitly states it doesn't reuse users' data, and its contracts with its subprocessors (OpenAI, Anthropic, etc.) also forbid them from doing so.

The flip side is that this freedom comes at a technical-skill cost. The Docker install is documented and takes half an hour, but maintaining an instance in production — updates, backups, monitoring — requires someone comfortable with a command line. It's not a tool you hand to a business use case with no support.

Another point: the publisher LangGenius is a Delaware company, whose governing law is California's. Its roots are French — the founding team is based in Paris — but the contract is American. For an organization that must document sovereign hosting, the self-hosted Community version is the only path.

Where it falls short

Dify's cloud hosts data in several countries with no guarantee of European residency: servers are in the United States, and data can pass through Australia, Canada, China, and the European Economic Area depending on processing needs. The self-hosted version solves this problem, but requires handling maintenance yourself.

Message credits included in the paid tiers are modest: 5,000 a month on Professional, 10,000 on Team. Once used up, you must bring your own API key from a model provider, adding a separate, variable bill. The actual cost of regular use is therefore higher than the tier's posted price.

Lastly, the documentation is in English, and so is the interface. A team working exclusively in French will have to work around this barrier, even if the tool naturally accepts prompts and documents in French.

Our recommendation

Dify is the best starting point for an organization that wants to build its own AI agents without locking itself into one provider. The self-hosted Community version with an open model or an API whose usage terms you control is the setup we recommend for professional use. If you don't have the technical resources to install it, our support service covers exactly this need.

Who it's for, and who it isn't

Suited for

Small business or firm that wants to build AI agents on its own data without handing it to a US cloud
Team with at least minimal technical skill for a Docker install, even without a dedicated developer
Organization evaluating several AI models that wants to be able to switch without rebuilding the whole application

Not a fit

Very small business with no technical resources: the self-hosted install requires knowing how to run a Docker container
Organization that requires exclusively European cloud hosting: Dify's cloud spans multiple countries, and self-hosting requires maintenance

Frequently asked questions

Do you need to know how to code to use Dify?: Not for the essentials. The visual canvas lets you wire up models, knowledge bases, and tools without a line of code. The learning curve starts when you want to call an external API that isn't in the catalog: then you need to write a bit of Python in a code node.
Does my client data leave my infrastructure?: In the self-hosted Community version, no: everything stays on your servers, and you choose which model to call. In the Cloud version, data passes through Dify's servers (hosted in the United States and other countries) and through the provider of the model you use. For full confidentiality, prefer self-hosting with a local model.
Is Dify really open source?: The Community version is published under the Dify Open Source License, a license based on Apache 2.0 with additional conditions. You can use it, modify it, and install it on your own servers freely. The Cloud and Enterprise versions are commercial services built on the same code, with extra features. Redistribution or resale is governed by the license's additional terms.
How much does Dify cost for a five-person team?: The Professional tier is $590 per workspace per year, including 3 members. For five people, you need to move up to the Team tier at $1,590 a year (up to 50 members). Add the cost of the model provider once the message credits run out: for example, an OpenAI API key billed by usage.
What's the difference with n8n?: n8n automates flows between existing applications. Dify builds AI applications around a model: a chatbot fed by your documents, an agent able to call tools, a RAG pipeline. Both are open source and self-hostable, but they solve different problems.