What Dify is
Dify is an open-source platform for building LLM applications. It offers a visual workflow canvas, a RAG pipeline from ingestion to retrieval, agents with tools from its marketplace, MCP servers or custom APIs, a prompt IDE, backend-as-a-service APIs, and integrations with observability tools such as Langfuse, Opik and Arize Phoenix. You can use Dify Cloud, self-host the Community Edition with Docker Compose, or ask about Dify Enterprise, which adds SSO, RBAC and support SLAs. Its Code node runs Python or JavaScript in a sandbox, and its Human Input node pauses a workflow for a person to respond.
App builder versus governed agent platform
Dify and EDDI overlap on RAG, tools and agents, but they optimise for different people. Dify optimises for builders who want to assemble applications visually and ship them quickly, with a code node as the escape hatch. EDDI optimises for operators who need agent behavior to be declarative, reviewable configuration with governance around it. Many teams reasonably start in Dify, and some move once tenancy, licensing or audit requirements arrive.
Builder convenience versus operating surface
| Dimension | Dify | EDDI |
|---|---|---|
| Primary abstraction | Visual workflow and chatflow canvas, with a prompt IDE | Config-as-code: JSON resources authored in the Manager or API, versioned like source |
| License | Dify Open Source License: Apache 2.0 plus conditions. Multi-tenant operation of the source needs written authorization, and the frontend logo and copyright must stay | Apache 2.0, OSI approved |
| Custom logic | Code node runs Python or JavaScript in a sandbox that blocks file system, outbound network and system commands, with a fixed package set | No runtime code evaluation; logic is declared, or sits behind an MCP tool or HTTP call |
| Human in the loop | Human Input node pauses a workflow for a form answered in the web app or by email; default wait of 3 days, then a timeout branch | Gates on turns, tool calls and group phases; timeout policies include WAIT_INDEFINITELY, AUTO_APPROVE, AUTO_REJECT and ABORT; Manager and Slack approvals |
| Tools and integrations | Marketplace, MCP servers and custom APIs | 84 MCP tools, an MCP client, OpenAPI import, httpCall, A2A |
| RAG | Built-in knowledge pipeline with document extraction | 8 embedding providers, 6 vector stores, plus zero-infrastructure httpCall RAG |
| Multi-agent | Agents and workflow nodes composed on the canvas | Group discussions with 7 styles, recorded decisions and dissent, optional human approval per phase |
| Identity and access | Workspaces; SSO and RBAC through Dify Enterprise | Optional Keycloak/OIDC with eddi-admin and eddi-approver roles in the open-source build |
| Observability and audit | Logs and monitoring, integrations with Langfuse, Opik and Arize Phoenix | Prometheus metrics plus an HMAC-signed, append-only, sequence-checked audit ledger |
| Deployment | Dify Cloud, Docker Compose, Enterprise on request | Docker, Helm, Kustomize, OpenShift (Red Hat certified image, Operator via OperatorHub) |
Details about the other tool were checked against its official documentation and repository in October 2026. These projects change quickly, so confirm current behavior with the vendor before you decide.
Choosing between them
Where Dify is the better choice
- You want a visual canvas that non-engineers can read and extend, with quick feedback while building.
- A plugin marketplace and ready workflow templates save you real integration time.
- You want a hosted option (Dify Cloud) and do not want to operate anything yourself.
- A Code node is a feature you want: small data transforms in Python or JavaScript inside the workflow.
- You run one internal deployment for your own team, so the multi-tenant license condition does not apply to you.
Where EDDI is the better choice
- Your legal team needs a plain Apache 2.0 license, including for serving many tenants or customers from one deployment.
- The security review forbids code execution inside the automation platform, sandboxed or not.
- Specific tool calls or group phases need a human decision, recorded in a signed ledger.
- Agent behavior must be versioned, diffed and promoted between environments as configuration.
- You need Kubernetes, OpenShift or air-gapped deployment with local models.
Moving an app from Dify
Dify apps are visual graphs and EDDI agents are configuration, so this is a translation rather than an import. The parts that need the most thought are Code nodes, which become configuration or a declared tool.
| Dify | EDDI |
|---|---|
| Chatflow or workflow | EDDI agent: pipeline plus behavior rules |
| LLM node and prompt | LLM task with a system message and versioned prompt snippets |
| Knowledge base | EDDI RAG knowledge base with an embedding provider and vector store |
| Tool, plugin or MCP server | MCP client connection, httpCall, or OpenAPI-imported API |
| Code node | Behavior rules and prompt logic, or an MCP tool if it needs real code |
| Human Input node | PAUSE_CONVERSATION rule or toolApprovals, with a timeout policy |
| Provider API keys | Secrets Vault entries, referenced as ${vault:key} |
Frequently asked
Is Dify open source?
Dify publishes its source under the Dify Open Source License, which is Apache 2.0 with additional conditions: you may not use the source to run a multi-tenant environment without written authorization, and you may not remove the logo or copyright information from the frontend. That is not the plain Apache 2.0 that EDDI uses, so read it with your legal team if either condition could apply.
Does EDDI have a visual builder like Dify?
EDDI Manager is a production React UI for building, deploying and monitoring agents, available in 11 languages. It is a management interface rather than a node canvas: you configure agents, prompts, tools and pipelines, and every change is a versioned resource rather than a diagram.
Why does EDDI avoid a code node?
A sandboxed code node is a reasonable feature, and Dify restricts what it can reach. But any code execution inside the platform is an attack surface that a security review has to examine. EDDI keeps agent behavior declarative, so what a tool can do is what you declare and approve.
Can we use Dify and EDDI together?
Yes. Dify can use MCP servers and custom APIs, and EDDI exposes an MCP server and REST API, so a Dify workflow can call an EDDI agent as a governed step. EDDI can call the Dify API through httpCall.
Other comparisons
- EDDI vs. LangChain and LangGraph
- EDDI vs. Microsoft Agent Framework (AutoGen and Semantic Kernel)
- EDDI vs. CrewAI
- EDDI vs. n8n
- EDDI vs. Flowise
- EDDI vs. Rasa
- All alternatives: builders, libraries, and cloud platforms
Ready for production?
Get managed deployment, SLA-backed support, and hands-on help from the team that builds EDDI.
Talk to us about enterprise deployment