Dify

This Dify review covers what Dify is used for, its key features, Dify pricing on the cloud and self-hosted, whether it is production ready, where it shines and falls short, and how it compares with other tools. Dify is an open source, low code platform from LangGenius to create LLM applications: chatbots, AI agents, agent workflows and RAG pipelines over your own knowledge bases. You design the logic on a visual canvas with a drag and drop interface, pick a model from different AI models (OpenAI, Anthropic, Gemini, open models and more), connect knowledge bases and tools, and publish the result as a web app, an embedded chat widget or an API. Deployment is flexible: use Dify Cloud, or run the platform self hosted with Docker on your own servers.

What is Dify used for?

Dify is used to build AI agents, build workflows and ship LLM applications without writing code for every step. Typical projects are customer support chatbots that answer from a knowledge base, internal assistants that search company documents, content generation pipelines, document extraction and classification, and AI workflows that call APIs and other tools. Teams use it to prototype an idea in a day, test prompts against different models, and move the same agent or workflow to production behind an API. For many teams it becomes the platform where every AI agent in the company lives.

It suits two groups. Non technical users and product teams can create workflows from templates and learn the basic concepts in a few hours. Development teams and technical teams use Dify as a backend for LLM apps: it handles prompts, retrieval, model management, logs and the API, while their own code handles the rest. According to VentureBeat, Dify runs on more than 1.4 million machines, and companies such as Maersk and Novartis build on its commercial versions.

Key features

• Workflow and chatflow canvas: a drag and drop editor where you connect LLM nodes, knowledge retrieval, conditions, loops, iteration, code, HTTP requests and tools to build complex workflows and agentic workflows. You can call one workflow inside another.
• Agents: build AI agents that use tool calling and reasoning to decide which tools to run, from a single agent to multi agent systems and autonomous agents inside a workflow.
• Knowledge bases and RAG pipelines: upload documents or sync data sources such as Notion and websites into knowledge bases, choose chunking and embedding settings, and retrieve with vector, full-text or hybrid search. Dify works with several vector databases.
• Model management: connect different AI models from many providers, including self-hosted models through Ollama or other OpenAI-compatible endpoints, and switch the model per node.
• Prompt IDE and debugging: test prompts, compare models side by side, and read step-by-step run logs and error messages for every node.
• Plugins and MCP: a plugin marketplace for models, tools and extensions, plus support for MCP servers so agents can use external tools.
• Publishing and API: publish as a web app, embed a chat widget, or call every app through a REST API. Annotations let you correct answers, and logs let you monitor usage.
• Code nodes: add custom code in Python or JavaScript when the built-in nodes are not enough.

Dify pricing

Dify Cloud has a free plan and two paid plans. The figures below come from the Dify pricing page as of September 2026.

• Sandbox plan (free): 200 message credits, 1 member, 5 apps, 50 knowledge documents, 50 MB of knowledge storage and 30 days of logs.
• Professional: $59 a month, or $590 a year. 5,000 message credits a month, 3 members, 50 apps, 500 documents, 5 GB of storage and unlimited logs.
• Team: $159 a month, or $1,590 a year. 10,000 message credits a month, 50 members, 200 apps, 1,000 documents and 20 GB of storage.
• Self-hosted Community: free, with all core features from the public repository in one workspace.
• Enterprise: custom pricing for self-hosted deployments with multiple workspaces, SSO and dedicated support.

The usage metric on Dify Cloud is message credits. They cover calls to the models Dify provides; you can also add your own provider API keys, and then you pay the model provider for those calls directly.

The cost surprise to check is that the subscription is only part of the bill. Model usage with your own keys is billed by OpenAI, Anthropic or another provider, and heavy RAG apps also need storage and knowledge-request capacity. Self-hosted Dify is free to run, but you pay for servers, updates and backups, and multiple workspaces or removing the Dify branding need a commercial license.

Self-hosted Dify: deployment and the open source license

Dify is self hosted with Docker Compose on Linux, macOS or Windows with WSL 2. The minimum is 2 CPU cores and 4 GB of RAM, and setup is a git clone, a copied environment file and docker compose up -d (Dify docs). For a production deployment, teams add a managed database, object storage and a vector database, and plan how they will update the platform to new releases. Community help runs through GitHub issues and the forum.

The code is published under a modified Apache 2.0 license. You may use it commercially, but two conditions apply: you may not run Dify as a multi-tenant service (one tenant is one workspace) without written permission, and you may not remove the Dify logo or copyright notice from the console and apps. That makes it source-available rather than a standard open source license, which matters if you want to resell Dify to many customers.

Is Dify production ready for AI agents and AI workflows?

For most internal assistants, support chatbots and AI workflows, yes. Dify has versioned apps, logs, annotations, API access with keys and enterprise options such as SSO. Its March 2026 funding round names Maersk, ETS, Anker Innovations and Novartis as users. The limits show up in large, custom systems: code nodes run in a sandbox that blocks some libraries, very large canvases get hard to read, and version control is built into Dify rather than in Git. Test your own workload, including peak traffic and document volume, before you commit.

Dify vs Microsoft Copilot Studio

Microsoft Copilot Studio is the low code agent platform most often compared with Dify. Both let you build AI agents on a canvas, ground them in knowledge bases and publish them to chat channels. Copilot Studio fits teams already on Microsoft 365: it connects natively to SharePoint, Dataverse, Teams and Power Automate, and it runs only in Microsoft's cloud. Dify is the more open AI platform: you choose any model, you can self-host it, and every agent or workflow is also an API for your own apps. Pick Copilot Studio when the data and the users live in Microsoft Office; pick Dify when you want model choice, custom code or your own deployment.

Where Dify shines — and where it falls short

Strengths:

• A visual canvas with drag and drop editing for AI agents, RAG and LLM workflows in one workspace.
• Works with almost any model provider, including self-hosted models.
• Open source code you can self-host, with a free Community edition.
• A built-in API, so developers can use Dify as the backend of their own app.

Trade-offs:

• A steeper learning curve for non technical users once workflows grow.
• The license limits multi-tenant and white-label use.
• Sandboxed code nodes limit custom code, and large canvases make complex workflows hard to follow.
• Model costs come on top of the subscription.

How Dify compares

• n8n: workflow automation with AI nodes, stronger for connecting business apps; Dify is stronger for RAG and agent apps.
• Voiceflow: conversational agents and chat and voice design for customer support teams.
• Make and Zapier: no-code automation across thousands of apps, with AI steps rather than a full LLM app builder.
• AI Builder: Microsoft's AI features for Power Platform. For teams on Microsoft 365, Copilot Studio is Microsoft's agent builder and the closest match; Dify wins on model choice and self-hosting, Copilot Studio on Microsoft Office and SharePoint data.
• Landbot, Chatfuel and ChatBot: chatbot builders for websites and messaging channels, simpler than Dify but less flexible.

Is Dify worth it?

Dify is worth it for teams that want to build AI agents, RAG chatbots and LLM apps quickly, test different models and still keep the option to self-host. The free Sandbox plan and the free Community edition make it cheap to try. As an AI platform for agent and workflow development it is one of the most complete open options. It is less of a fit if you need deep integrations with many business apps (n8n or Make are simpler), if you plan a multi-tenant SaaS on top of Dify, or if your team would rather write the whole stack in code. If you want help to build AI agents or workflows on Dify, or to run it self-hosted in production, the low-code agencies and developers on LowCodeDevs can take it on.

Sources

• Dify pricing: plans, message credits and limits.
• Dify docs: Docker Compose deployment: system requirements and setup.
• Dify license: the modified Apache 2.0 terms.
• Dify raises $30 million, VentureBeat: adoption figures and named customers.
• Dify Review 2026, DEV Community: a six-month user review (its "5,000 workflow executions" and 10 GB for Professional differ from the pricing page, which lists 5,000 message credits and 5 GB).
• 7 best Dify alternatives, Gumloop and Dify AI review, GPTBots: comparisons (written by competitors).