Dify 1.15.0 adds CLI, CoT reasoning, and smarter workflows
New difyctl lets you run workflows from terminal; chain-of-thought now streams live.
Dify 1.15.0 is a significant update to the open-source LLM application platform, bringing a new command-line interface, deeper reasoning visibility, and more flexible workflow interactions.
The headline feature is difyctl, a CLI client that lets developers run Dify workflows directly from the terminal, enabling integration with scripts, CI pipelines, and personal agents. It supports scoped environment variables and consistent error handling, and binaries are available for macOS, Linux, and Windows with checksum verification.
Another major addition is live chain-of-thought (CoT) streaming in workflows and chat flows. The model's reasoning now appears in a dedicated "thinking" panel while keeping the final answer clean. This reasoning persists after page refresh and is also visible in CLI output and workflow run previews. Additionally, human-in-the-loop forms now support dropdown selects and file/multi-file uploads, not just free text, allowing structured input during paused workflows.
The release also adds support for slow, long-running generation models (e.g., image/video) via a polling mechanism, so workflows no longer time out waiting. Knowledge import now extracts embedded images from Excel files, preserving diagrams and screenshots. UX improvements include a redesigned onboarding, faster navigation palette, collapsible workflow editor panels, and safer delete confirmations. Security fixes include a path traversal patch (CVE-2026-41948) and hardened outbound HTTP timeouts.
- difyctl CLI allows running Dify workflows from any terminal, with cross-platform binaries and checksum verification.
- Live chain-of-thought reasoning streams in workflow UI and CLI, persisting after refresh.
- Human-in-the-loop forms now support dropdowns and file uploads; polling enables slow model support.
Why It Matters
Dify 1.15.0 empowers developers to automate LLM workflows from the command line and gain deeper reasoning transparency.