Dify
Build and operate LLM applications with visual workflows, RAG pipelines, and tool-enabled agents.
- Source repo
- langgenius/dify
- Stars
- ★ 160k
- Last updated
- today
- License
- NOASSERTION
- Primary language
- Python
- FA score
- 55/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformOpenAI API
- Cost
- Free tier plus a paid hosted plan
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A product team wants to design and test a multi-step AI Workflow visually before shipping it.
- Not a fit if
- Teams that need a fully offline setup with no cloud service
- Users wanting zero-config use without Docker or Compose
- Teams needing a plain OSI license without extra conditions
- Source review
- 55/100 · Major gaps
What does this agent do, and when should you use it?
Dify is an open-source LLM application development platform that combines visual AI workflows, RAG pipelines, agent capabilities, model management, and observability. Teams build and test flows on a visual canvas, while the Prompt IDE supports prompt authoring, model comparison, and chat-app features such as text-to-speech. Its RAG Pipeline spans document ingestion through retrieval and includes text extraction for PDFs, PPTs, and other common document formats. Agents can be defined with LLM Function Calling or ReAct and extended with built-in or custom tools. Dify is accessed through a browser dashboard and corresponding APIs, with Dify Cloud, Docker Compose self-hosting, and enterprise deployment options described.
Dify lets users build and test Workflows on a visual canvas, then author prompts and compare model performance in the Prompt IDE. It integrates hundreds of proprietary and open-source LLMs from dozens of inference providers, plus self-hosted solutions and OpenAI API-compatible models. Its RAG Pipeline ingests documents and retrieves from them; the README specifically documents text extraction from PDFs, PPTs, and other common document formats. Users define Agents using LLM Function Calling or ReAct and attach built-in or custom tools; the README names Google Search, DALL·E, Stable Diffusion, and WolframAlpha among 50+ built-in tools. It monitors application logs and performance, exposes corresponding APIs, and starts a self-hosted instance with docker compose up -d, followed by browser initialization at http://localhost/install.
- A product team wants to design and test a multi-step AI Workflow visually before shipping it.
- A knowledge-application team needs to ingest PDFs, PPTs, and similar documents and retrieve relevant content for an LLM app.
- A developer needs to choose between LLM Function Calling and ReAct for an agent, then add built-in or custom tools.
- An application team needs to work across GPT, Mistral, Llama3, or OpenAI API-compatible model paths.
- An operations team wants to improve prompts, datasets, and models from application logs, performance data, and annotations.
- An organization wants a Docker Compose deployment in its own environment while integrating the platform through APIs.
How do you install or deploy this agent?
Minimum requirements are a 2-core CPU and 4 GiB RAM. Install Docker and Docker Compose v2.24.0 or later, then run from the repository directory:
cd dify
cd docker
cp .env.example .env
docker compose up -dOpen http://localhost/install to initialize the dashboard. The supplied material does not document provider credential configuration; for configuration changes, edit docker/.env and rerun docker compose up -d from docker.
How do you use this agent?
Open http://localhost/install and complete initialization, then build and test an AI flow in the Workflow canvas. Use the Prompt IDE to author prompts and compare models, configure a RAG Pipeline for document ingestion and retrieval, or define a Function Calling or ReAct Agent and add tools. Corresponding APIs are available, but the supplied material does not specify endpoints, authentication, or request examples.
What are this agent's strengths and limitations?
- Brings Workflow, Prompt IDE, RAG Pipeline, Agents, model management, and LLMOps into one platform.
- The documented RAG flow includes ingestion and retrieval, with out-of-the-box text extraction for PDFs, PPTs, and other common formats.
- The stated model coverage spans dozens of inference providers, proprietary and open-source LLMs, self-hosted solutions, and OpenAI API-compatible models.
- Offers hosted cloud, Docker Compose self-hosting, and enterprise paths, with corresponding APIs for its capabilities.
- Self-hosting requires at least a 2-core CPU, 4 GiB RAM, Docker, and Docker Compose v2.24.0 or later.
- The supplied quick start covers Docker Compose only; the linked Kubernetes, Helm, Terraform, and CDK deployments are community contributions.
- The supplied material does not provide API endpoints, authentication details, provider-key setup, or production deployment procedures.
- Licensing needs verification: repository metadata says NOASSERTION, while the README states a Dify Open Source License based on Apache 2.0 with additional conditions.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Dify This agent | 55 · Major gaps | Self-hosted serviceFreemium | ★ 160k | today | Python | OpenAI API |
| Agents Universe | 68 · Some gaps | Self-hosted serviceFree + model costs | ★ 830 | today | Python | OpenAI API · Claude API |
| 12-Factor Agents | 24 · Major gaps | Library / SDKFree | ★ 27k | 1y ago | TypeScript | — |
| Agents Towards Production | Insufficient evidence | Library / SDKFree + model costs | ★ 22k | 4d ago | Jupyter Notebook | — |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows: README mentions security disclosure process (SECURITY.md), but does not detail permission model, user confirmation mechanisms, data flow transparency, sensitive data handling, dependency security audits, external effects, rollback, or source attribution. Therefore, each criterion scores low, only 1 point for existence of security policy. Deduction: lack of specific implementation details.
Evidence shows: unit and integration tests exist (e.g., api-tests.yml), but failure messages are not detailed. Self-consistency, dependency availability, and failure messages are partially supported but not adequately handled. Deduction: limited test coverage, failure messages not explicit.
Evidence shows: README describes multiple usage scenarios (cloud, self-hosted, enterprise) and provides deployment options (Docker, Kubernetes, Terraform). Capability boundaries and trigger precision are not explicit, but environment fit is good. Deduction: capability boundaries and trigger precision lack clear documentation.
Evidence shows: README provides installation instructions, examples, and community links, with clear information architecture. License is a custom open-source license, but versioning changelog is not explicit. Known limitations are not detailed. Deduction: versioning changelog and known limitations insufficient.
Evidence shows: README describes output usability (API, visual interface) and marginal value (rapid prototype to production). Cost-benefit is not explicit, but overall effective. Deduction: cost-benefit lacks specific data.
Evidence shows: claims in README are supported by documentation links, but no independent verification. Cross-source corroboration is limited, and fact/inference separation is not explicit. Deduction: lack of independent verification and explicit fact/inference separation.
- The repository uses a custom open-source license with additional conditions that may affect commercial use.
- Security policy exists, but detailed permission model and data flow transparency are not provided.
- Versioning changelog is not explicit; users need to track updates themselves.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
1.19.0 v1.19.0 - Trace Your Workflows, Layer by Layer
Release notes excerpt
Dify 1.19.0 brings nested workflow tracing to your observability tools. Follow a workflow into the Workflow Tools it calls, inspect the steps inside each invocation, and keep the LLM context alongside the execution tree. This release also makes Workflow Tools portable with App packages, adds Skill Hub automation, and makes Agent files easier to upload and preview. The bundled plugin daemon is upgraded to 0.6.14-local. [Release highlights](#user-content-release-highlights) · [Upgrade guide](#user-content-upgrade-guide) · [Detailed changes](#user-content-detailed-changes) …View on GitHub ↗ -
1.18.0 v1.18.0 — Smarter Agent Routing, Portable Apps, Deeper Workflow Visibility
Release notes excerpt
Dify 1.18.0 brings Agents that route workflows, portable app packages, deeper visibility into nested workflow runs, and the opt-in Agent Knowledge Base experience. OpenAPI and the new difyctl expand how you build and manage apps from the command line. > [!IMPORTANT] > Read the upgrade guide below before deploying. Upgrade the API, workers, Agent Backend, and Agent runtime images together. Local Agent sandboxes require a non-empty authentication token, and Agent Backend uses a new run protocol. Source installations require Python 3.13. Install the new difyctl client and sig …View on GitHub ↗ -
1.17.1 v1.17.1 - Bug Fixes and Improvements
Release notes excerpt
> [!WARNING] > Self-hosted deployments using the bundled Weaviate must complete a manual, staged upgrade before starting 1.17.1. The bundled Weaviate server moves from 1.27.0 to 1.39.2 — 12 minor versions — and skipping minors is unsupported. Pulling and restarting can silently and permanently break vector search. See Upgrade Guide below. Fresh deployments, external Weaviate, and other vector stores are unaffected. New Features Dataset-Scoped Knowledge Base API Keys …View on GitHub ↗
FAQ
Can it be fully self-hosted?
docker/.env as the place for configuration changes.