Dify
Dify is an LLM app development platform that combines visual AI workflows, document retrieval pipelines, autonomous agents with their own sandbox, model management and observability, with an API for everything it builds. It connects to hundreds of models including any OpenAI-compatible one, and to MCP servers. Its licence is Apache-2.0 with added conditions: no multi-tenant service and no removing the Dify logo without written permission, so this site labels it source available.
What does Dify do?
Dify is "an open-source LLM app development platform", in its README's words, that "combines AI workflow, RAG pipeline, agent capabilities, model management, observability features" for going from prototype to production. You build and test workflows "on a visual canvas", and its agents are "autonomous agents with a sandbox of their own: they run commands, install software, and handle files." Agents connect tools from the Dify Marketplace, MCP servers or your own APIs.
It supports "hundreds of proprietary / open-source LLMs from dozens of inference providers and self-hosted solutions", including any OpenAI-compatible model. Its retrieval pipeline extracts text from PDFs, presentations and other documents. A prompt IDE compares models, observability integrates with Opik, Langfuse and Arize Phoenix, and "all of Dify's offerings come with corresponding APIs."
Key facts
- Licence: the Dify Open Source License, a modified Apache-2.0 with added conditions.
- Condition one: "Unless explicitly authorized by Dify in writing, you may not use the Dify source code to operate a multi-tenant environment", where one tenant is one workspace.
- Condition two: you may not remove or modify the logo or copyright information in the Dify console or applications; this covers the web front end.
- Contributors agree Dify may change the licence terms and use contributed code commercially; the licence also refers to an appearance patent on the interaction design.
- Visual AI workflows, RAG pipelines from PDFs and slides, agents with their own sandbox, a prompt IDE, and an API for everything built.
- Models: hundreds from dozens of providers and self-hosted solutions, including any OpenAI-compatible model; MCP servers and a marketplace for tools.
- Install: Docker Compose v2.24 or later on a machine with at least 2 CPU cores and 4 GiB of RAM; Dify Cloud and Dify Enterprise are hosted and paid options.
- Latest tagged release when read: 1.17.1 on 2026-09-10; 156,206 stars on 2026-09-17.
What does it replace, and where does it fall short?
Dify replaces Zapier and Make where the automation is itself an AI application: a chatbot over your documents, an agent that runs tasks, a workflow of model calls, published with an API. It is the pick for AI-first workflows on the open-source alternatives to Make page.
Where it falls short of open source: you may not run it as a multi-tenant service or remove its logo without Dify's written permission, and Dify may change the terms, so this site labels it source available. It is built for AI applications rather than the long tail of app-to-app integrations Zapier covers. For MIT licensing and a visual AI builder, Langflow.
How does Dify run?
Clone, then in the docker folder copy .env.example to .env and run docker compose up; it needs Docker Compose 2.24 or later and at least 2 CPU cores and 4 GiB of RAM. Dify Cloud runs the same product hosted, with a sandbox plan the README says includes 200 free GPT-4 calls, and Dify Enterprise adds SSO, role-based security and support agreements.
Who is Dify for?
A team building AI applications, chat over documents, agents, model workflows, for its own organisation, on one workspace per deployment. A company that wants to host it for many customers needs Dify's written permission.
What limits does the README state?
From the README: a minimum of 2 CPU cores and 4 GiB of RAM. The licence conditions quoted above are the limits to read.
Questions people ask
Is Dify open source?
Not by the OSI definition. Its licence is Apache-2.0 with added conditions: no multi-tenant service without written permission, and no removing or changing the logo in the console or applications. Contributors also agree that Dify may change the terms. This site labels it source available.
Can I use Dify in my company?
Yes. Running it for your own organisation, with one workspace per deployment, is within the licence. Operating it as a multi-tenant service for other organisations needs Dify's written permission, and the logo in the front end must stay.
Does Dify support local models?
It supports hundreds of models from dozens of providers and self-hosted solutions, including any OpenAI-compatible model, by its README. Ollama is not named in the README, but any local server with an OpenAI-compatible API fits that description.
How does Dify compare with Make and Zapier?
Dify is built for AI applications: workflows of model calls, agents with a sandbox, retrieval over documents, each with an API. Make and Zapier are hosted services for connecting apps. The open-source alternatives to Make page compares Dify with n8n, Activepieces, Langflow, Windmill and Huginn.
What hardware does Dify need?
At least 2 CPU cores and 4 GiB of RAM, with Docker Compose 2.24 or later, by the README. Model inference is separate: hosted models need only keys, while local models need their own hardware.
Sources
- Dify README and LICENSE (Dify Open Source License): github.com/langgenius/dify, read 2026-09-17.
Compared with the others
On the open-source alternatives to Zapier page, Dify is the pick for ai-first workflows and agents. Visual AI workflows, retrieval pipelines, agents with their own sandbox, MCP tools and hundreds of models, with an API for everything built.
Also on that page: Activepieces for a no-code builder under mit, n8n for the largest integration library, Windmill for code-first automation for developers.
On the open-source alternatives to Make page, Dify is the pick for ai-first workflows and agents. Visual AI workflows, retrieval pipelines, agents with their own sandbox, MCP tools and hundreds of models, with an API for everything built.
Also on that page: n8n for the largest integration library, Activepieces for a no-code builder under mit, Langflow for visual ai flows as apis or mcp tools.
More workflow automation
A fair-code workflow automation platform with 1,500+ integrations, 9,000+ templates, AI agents over OpenAI, Anthropic, Google or open models, and code steps in JavaScript and Python, under the Sustainable Use License.
An MIT visual platform for building and deploying AI agents and workflows that turns every flow into an API or an MCP server, with Python-editable components and multi-agent orchestration.
An MIT system, running since 2013, for building agents that watch the web and act for you, which its README calls a hackable version of IFTTT or Zapier on your own server, with webhooks, scraping and JavaScript steps.
An MIT automation builder its README calls an open-source replacement for Zapier, with hundreds of TypeScript pieces that double as MCP servers, AI agents, human approval steps and a no-code builder.
Added September 17, 2026. Every claim above comes from the project's README, LICENSE or model card, read on September 17, 2026, or from the GitHub API on the date shown in the panel. Found an error? Write to reveneau@licheo.com and it is fixed in the next weekly pass. Repository: github.com/langgenius/dify.
Running one of these inside your own environment, with your own data and your own security rules, is the kind of work Reveneau does. Read how a forward deployed engagement works.