Dify: Build LLM Apps Without the Plumbing (GitHub, Scanned)

Dify is a self-hosted platform for building applications on top of language models without writing the plumbing. A visual workflow editor chains prompts, model calls, code steps, conditions and tool calls; a knowledge base ingests documents for retrieval; an agent builder adds tool use and memory; and every app you build gets a web interface, an API and a log of every run. It connects to hundreds of models, local and hosted.

It is on this list because it is the most complete of the open builders and because it is genuinely used in production: a large Python API, a Next.js front end and a plugin system with its own marketplace, all under active development. The licence is Apache 2.0 with two conditions, no multi-tenant hosting without a commercial licence and no removing the Dify logo from the front end, which matters if you plan to resell it and not at all if you plan to use it.

• Repository: github.com/langgenius/dify
• Licence: custom (Other)
• Language: TypeScript. Stars: 155.9K. Forks: 24.6K. Last push: Sep 16, 2026.
• Scan: clean, read the notes, Sep 16, 2026, commit 38f9d85

Who it is for

Product teams prototyping an AI feature before committing engineering time, internal-tools builders who need a chat over company documents by Friday, and developers who want an agent runtime with tracing that they can host themselves.

Getting started

1. Clone and start with Docker Compose (needs Docker and about 4 GB of free memory)

git clone https://github.com/langgenius/dify.git && cd dify/docker && cp .env.example .env && docker compose up -d

2. Open the setup page and create the admin account

open http://localhost/install

3. Add a model provider under Settings, then create your first app from a template

# Settings > Model Provider > add Ollama, OpenAI or another provider

Safety scan

We cloned langgenius/dify at commit 38f9d85 on Sep 16, 2026 and ran the checks described on the GitHub Tools page: credential patterns, decode-and-execute code, install-time scripts, committed binaries, risky CI workflows, every host the code talks to, known vulnerabilities in pinned dependencies, and project hygiene. A person read every hit. This is what we found.

• No secrets in the source: the one candidate is a fake GitHub token in a test. No self-decoding code. The bare-IP hits are test cases for the SSRF proxy, which exist to prove private addresses are blocked.
• The pattern hits are tests quoting install commands and a commented usage line in the CLI installer script. The copy of the Monaco code editor under web/public/vs is a shipped asset and is skipped, as is all JavaScript under a public folder.
• Four install-time scripts: web/package.json refuses installs that are not pnpm, and the dev-proxy package builds itself on prepare. Nothing fetches remote code during install. The scripts under cli/scripts download release binaries from GitHub for the optional Dify CLI, not the server.
• Known advisories are few for a code base this size. The one critical is chromadb 0.5.20 in api/uv.lock, one of a dozen optional vector stores, loaded only if you choose it; the Docker setup defaults to a different store.
• 37 workflows, 59 of 66 third-party actions pinned to a commit, Dependabot on, security policy and code of conduct present. Among the best hygiene on this list.
• Licence: Apache 2.0 with two conditions, no multi-tenant hosting without a commercial licence and no removing the logo from the console.

What the scanner counted

The raw findings

Every hit the scanner wrote out, with a link to the exact line at the scanned commit. Secrets candidates are redacted.

By the numbers

Lines by language: TypeScript 1.3M, Python 979.5K, JSON 206.3K, Markdown 48.6K, YAML 23.3K, Go 13.3K.

Questions

Is Dify free to self-host?

Yes, for your own use, including commercial use inside your company or as the back end of your own product. The licence is Apache 2.0 with two extra conditions: you may not run it as a multi-tenant service for other organisations without a commercial licence, and you may not remove the Dify logo from the console. Dify Cloud is the paid hosted version.

What is the difference between Dify and LangChain?

LangChain is a code library; Dify is an application with a visual editor, a database and an API that you host. You can build the same things in either, but Dify gives non-developers a way in and gives developers logging, versioning and a deployment target without assembling them. Many teams prototype in Dify and rebuild the winners in code.

Does Dify work with local models?

Yes. Ollama, LM Studio, LocalAI and any OpenAI-compatible server can be added as a model provider, and a workflow can mix local and hosted models step by step. Embedding models for the knowledge base can be local too.

This post is part of GitHub Tools, where every repository is cloned and scanned before it is written up. The scan is a snapshot of one commit on one day; the repository has moved on since, so check it before you install.