Stacks and Flows

Dify

Self-hostable platform for agentic AI apps — RAG pipelines, agent workflows, and model management in one stack.

platform freemium Updated 2026-09-13
Visit Dify →

Use when

  • Teams wanting a self-hosted RAG + agent platform without stitching 4 tools together
  • Teams whose residency requirements call for an operator-controlled deployment
  • Builders who want a visual layer over LangChain-style primitives without writing code

Operational upside

  • All-in-one platform — agent workflows, RAG, prompt management, and model routing in one product
  • Source-available under the Dify Open Source License and self-hostable with documented deployment options
  • Visual flow builder accessible to non-developers, with code escape hatches for engineers
  • Multi-language product surface and support for multiple model providers
  • Managed Cloud and self-hosted Community deployment options

Avoid or plan for

  • Dify Open Source License adds multi-tenant service and branding conditions — read before embedding or reselling
  • The integrated approach trades some component-level flexibility for one operating surface
  • Upgrade and migration work depends on the self-hosted version and deployment method
  • Model-provider coverage and controls should be checked against the providers a team actually uses

What it is

Dify is a source-available LLM application development platform that bundles an agent workflow builder, a RAG pipeline builder, prompt and model management, and an API serving layer. It runs as a managed cloud service or as a self-hosted Community deployment. Its Dify Open Source License adds conditions beyond standard Apache 2.0, including restrictions relevant to multi-tenant services and frontend branding.

The project provides a multi-language interface and supports a broad set of model providers. Adoption metrics such as GitHub stars change frequently and are not used here as a proxy for production suitability.

Who it’s for

Dify is a candidate for teams that want agent workflows, retrieval, model configuration, and an application interface in one operating surface. A self-hosted deployment may support residency or control requirements, but the team still has to verify storage, model endpoints, telemetry, backups, and access paths against its policy.

It’s a poor fit for teams who already have a strong opinion on each component (you’ll fight the platform’s defaults) and for teams whose use case is narrow enough that a single SDK suffices.

Strengths

  • Integrated application surface. Agent flows, RAG, prompt management, model providers, and API publication live in one product surface.
  • Visual + code. The flow builder is approachable for non-developers and has Python/JS code blocks for escape hatches when needed.
  • Self-hostable. Dify documents a Docker Compose Community deployment; teams should validate other deployment methods and operational requirements against current documentation.
  • Multi-provider. The platform supports configuring multiple model providers, subject to the capabilities of each integration.

Weaknesses / Watch out

  • License nuance. The Dify Open Source License is based on Apache 2.0 but adds conditions, including commercial-license requirements for operating a multi-tenant service and restrictions on removing frontend logo or copyright information. Read the current license before embedding or reselling Dify.
  • Commercial terms. Referral or partner mechanics are not part of this product-fit assessment; confirm any current program directly with the vendor.
  • Integrated-platform trade-off. A single operating surface reduces integration work but couples the application to Dify’s workflow, data, and provider abstractions.
  • Upgrade ownership. Self-hosted teams should pin versions, read migration notes, test upgrades, and retain a rollback path.

Best paired with

  • Anthropic Claude or OpenAI as model backends — Dify’s model routing handles multiple providers cleanly.
  • Self-hosted Postgres + Qdrant or pgvector as the persistence and vector layer in self-hosted deployments.
  • n8n for triggers and side-effect workflows that should live outside the agent loop — let Dify handle agent reasoning, n8n handle the boring SaaS plumbing.

Frontier verification — self-hosting transfers ownership

Dify’s Community deployment gives teams a self-hosted application surface, but the deployment method does not transfer patching, backups, secrets, queues, databases, or upgrade testing back to the vendor. Those become operator responsibilities.

Before choosing it for residency or control, review the current Dify license and document which team owns recovery and upgrades. “Self-hosted” is an operating model, not merely a location checkbox.

Current sources: self-hosted installation documentation and the repository license. Verified 2026-09-13.

Verdict

Conditionally recommended when an integrated, self-hostable application surface is the requirement. A narrower SDK may be easier to own when only one capability is needed. Review the current license before embedding, reselling, or removing required branding.

Sources

FAQ

Is Dify free?
Dify is freemium. Check the official pricing page for current tiers and limits.
What is Dify best for?
Teams wanting a self-hosted RAG + agent platform without stitching 4 tools together Teams whose residency requirements call for an operator-controlled deployment Builders who want a visual layer over LangChain-style primitives without writing code
What are the main downsides of Dify?
Dify Open Source License adds multi-tenant service and branding conditions — read before embedding or reselling The integrated approach trades some component-level flexibility for one operating surface Upgrade and migration work depends on the self-hosted version and deployment method
Who should use Dify?
Self-hostable platform for agentic AI apps — RAG pipelines, agent workflows, and model management in one stack. See our review for the full pros and cons.