Dify 1.18.0: Agents with Branch Routing and Portable App Packages
On October 23, 2026, release 1.18.0 of the Dify platform came out — a visual agent builder that assembles AI workflows, RAG pipelines and agent scenarios in a single workspace. The release focuses on letting the agent choose the next step itself and on making applications portable between environments with a single file.
What's new
- Output Routes for agents. The agent can now decide which workflow branch to take: each branch is given a selection condition, and a switch at the output returns the chosen route. This allows, for example, answering a simple request right away or handing it off to a narrow specialist.
- Portable application packages. An application, together with its resources, is packed into a single .ifpkg file. Workflow and Chatflow packages carry embedded agents, skills and their files; on import, resources are recreated in the target workspace and references are rebound.
- Text streaming in Agent v2. Regular Agent v2 nodes now stream text as it is generated. Structured results, routed results and nodes with error handling via default values remain buffered.
- OpenAPI and difyctl. The new difyctl CLI client and an extended OpenAPI let you build and manage applications from the command line, not just through the web interface.
- Agent Knowledge Base as opt-in. A separate experience for working with the agent's knowledge base has appeared, enabled on request.
What the platform does
Dify is an open-source LLM application development platform with a visual canvas for workflows, a RAG pipeline from document upload to retrieval, agents with their own sandbox and observability tools. Agents can run commands, install software and work with files, connect tools from the Dify Marketplace, MCP servers or their own APIs. The platform also offers LLMOps capabilities: log monitoring, prompt comparison and model improvement based on production data.
Your own model instead of the cloud
Your own model is connected via an OpenAI-compatible API — the documentation states support for any OpenAI API-compatible models, including self-hosted solutions. The platform is installed via Docker Compose: Docker and Docker Compose v2.24.0 or newer are enough, with minimum requirements of 2 CPU cores and 4 GB of RAM. The license is Apache 2.0 with additional conditions: commercial use is permitted, including backend services, but multi-tenant mode requires written permission or a commercial license; removing the logo and copyright in the frontend is prohibited.
How it differs from its neighbors
Among visual agent builders, Dify stands out because its documentation states sandboxed code execution and voice capabilities (text-to-speech) — Langflow, AutoGPT Platform and n8n do not mention these features. On connecting your own model, Dify overlaps with n8n via an OpenAI-compatible API, whereas Langflow and AutoGPT Platform state Ollama. Stateful graphs and Docker deployment are present in all four platforms.
| Dify | Langflow | AutoGPT Platform | n8n | |
|---|---|---|---|---|
| Type | builder | builder | builder | builder |
| Language | Python | Python | Python | TypeScript |
| License | Apache 2.0 with add. cond… | MIT | PolyForm Shield Licen… | Sustainable Use Licen… |
| GitHub stars | 159,392 | 155,815 | 187,450 | 206,788 |
| Releases in 90 days | 4 | 15 | 12 | 58 |
| Multiple agents in a chain | — | yes | — | — |
| Custom tools and functions | yes | yes | — | — |
| MCP support | yes | yes | — | — |
| Knowledge base (RAG) | yes | yes | — | — |
| Sandboxed code execution | yes | — | — | — |
| Human approval | — | — | — | yes |
| Graphs and state persistence | yes | yes | yes | yes |
| Visual editor | yes | yes | — | yes |
| Tracing and debugging | yes | yes | — | yes |
| Voice and speech | yes | — | — | — |
| Own server and Docker | yes | yes | yes | yes |
| Own model: Ollama | — | yes | yes | — |
| Own model: LM Studio | — | — | — | yes |
| Own model: OpenAI-compat. API | yes | — | — | yes |
Cross-checked against the projects' READMEs and documentation as of 10/24/2026: "yes" means the feature is stated there, a dash means it is not mentioned (which does not mean the feature is absent). Stars and releases are per GitHub data.
Updating is worthwhile for those already running agents on Dify who want branch routing without manual switching, and worth a closer look for those building agents with a sandbox and moving applications between environments. More platform breakdowns and news about running neural networks on your own hardware are in the CPU3D news feed.
