Dify
AIInvoke your Dify LLM apps from FlowRunner. Agents send chat and completion messages, run and monitor Dify workflows, manage conversations and annotation replies, upload files, and convert between audio and text on Dify Cloud or self-hosted deployments.
What This Integration Enables
Dify is where your team builds LLM apps. FlowRunner is where those apps meet real business processes. This connector turns any Dify app, chat, agent, completion, or workflow, on Dify Cloud or your own deployment, into a callable step: agents send chat and completion messages with conversation continuity, run and monitor Dify workflows, upload files for multimodal apps, convert between audio and text, and manage the conversation history behind it all. It also governs what those apps are allowed to promise. Dify's annotation replies return a curated answer verbatim whenever a matching question arrives, bypassing the model entirely. That is the most powerful quality control in the product, and it is exactly the kind of write that should carry a [human-in-the-loop](/concepts/human-in-the-loop) gate: an agent can find the answers that need curating, but a person decides the words your app will repeat to every future customer.
Without FlowRunner
With FlowRunner
Use Case Scenarios
Drafted replies with a graded escalation path
Queued tickets pulled from [Airtable](/integrations/airtable) run through the support app with Send Chat Message, each keeping its conversation ID so follow-ups carry context. The agent sorts the drafts: routine answers write back to the queue for sending, while anything touching refunds, exceptions, or legal language posts to the support channel in [Slack](/integrations/slack) for a senior agent to edit and approve. The same pattern powers a channel assistant, answering questions in Slack with per-user conversation continuity and posting replies back to the thread.
Dify workflows as steps in a bigger machine
A document-heavy process needs the multi-step logic the team already built as a Dify workflow app. FlowRunner uploads the source files with Upload File, executes the app with Run Workflow, and checks results with Get Workflow Run, stopping a runaway execution with Stop Workflow Task when needed. Around that step, FlowRunner does what Dify does not: fetches the inputs from the systems that hold them, routes the outputs to the people who act on them, and logs each run in [Google Sheets](/integrations/google-sheets) for review. Audio to Text and Text to Audio extend the same flows to voice.
The feedback-to-annotation loop
Every week, the agent sweeps conversation history with List Messages and collects the replies users disliked through message feedback. For recurring questions with bad answers, it drafts a corrected question-and-answer pair and presents the batch to the app owner. Approved pairs become annotations through Create Annotation; outdated ones get revised with Update Annotation or retired with Delete Annotation. The app's worst answers get systematically replaced by its best ones, and a person signs every word that gets locked in.
Human-in-Loop Highlight
Create Annotation is a quiet operation with a loud consequence: the answer it stores is returned verbatim to every future user whose question matches, bypassing the model and every prompt improvement you make afterward. So the curation agent never publishes an annotation it drafted. It posts the batch to the app owner: "Six recurring questions drew negative feedback this week. Here is each question, the reply users disliked, and the proposed replacement answer. Approve, edit, or reject each pair; approved pairs become the app's fixed reply for that question." The owner edits two, approves four, rejects none. The words the app will now repeat indefinitely were chosen by a person, on purpose.
Agent Capabilities
23 actionsChat and Completion
4- Send Chat Message Sends a message to a chat, agent, or chatflow app and returns the answer, with conversation IDs carrying context across turns. The core step for assistants and drafted-reply pipelines.
- Stop Chat Message Stops an in-progress chat generation by task ID. Used to cut off runaway generations.
- Send Completion Message Runs a text-generation app and returns the output. Used for one-shot summaries, rewrites, and translations.
- Stop Completion Message Stops an in-progress completion generation by task ID.
Workflows
3- Run Workflow Executes a Dify workflow app's latest published version and returns the final outputs. The step that embeds your team's multi-step Dify logic inside a FlowRunner flow.
- Get Workflow Run Retrieves a workflow run's status and outputs by run ID. Used to check long executions.
- Stop Workflow Task Stops an in-progress workflow task by task ID.
Conversations and Messages
6- List Conversations Lists a user's conversations, most recent first. Used to resume the right thread for a returning user.
- Rename Conversation Renames a conversation, or auto-generates a title from its content. Used to keep histories navigable.
- Delete Conversation Permanently deletes a conversation and its message history for a user. Used in retention workflows, with review, since the history cannot be recovered.
- List Messages Retrieves a conversation's message history. The evidence-gathering step behind quality sweeps.
- Submit Message Feedback Submits or revokes like or dislike feedback on a message, with optional written content. Used to capture quality signals that feed the annotation loop.
- Get Suggested Questions Retrieves the AI-generated follow-up questions for a message. Used to guide users toward answerable next steps.
Files and Audio
3- Upload File Uploads a file to Dify for attachment to chat, completion, or workflow messages. The entry point for multimodal apps.
- Audio to Text Transcribes an audio file using the app's configured speech-to-text model. Used to bring voice input into text workflows.
- Text to Audio Converts text or an existing answer to speech, saves the audio to FlowRunner file storage, and returns its URL.
App Configuration
3- Get App Parameters Retrieves the app's configuration: opening statement, input form schema, and file-upload limits. Used so workflows send inputs the app actually accepts.
- Get App Info Retrieves the app's name, mode, tags, and author. Used to verify a connection points at the intended app.
- Get App Meta Retrieves app meta information, including tool icons.
Annotations
4- List Annotations Lists the curated question-and-answer pairs configured for an app. The inventory step in curation sweeps.
- Create Annotation Creates a curated pair that becomes the app's verbatim reply for matching questions. The operation this page's human gate exists for.
- Update Annotation Updates an existing annotation's question and answer. Used when policy changes make an approved answer stale.
- Delete Annotation Permanently deletes an annotation. Used to retire fixed replies that no longer apply.
Frequently Asked Questions
What can FlowRunner do with Dify?
FlowRunner agents can run Send Chat Message, Stop Chat Message, and Send Completion Message in Dify, plus 20 more actions.
Does connecting Dify to FlowRunner require OAuth?
No. Dify connects to FlowRunner with an API key, no OAuth flow required.
Can Dify trigger a FlowRunner workflow automatically?
Dify doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.
Start building with Dify
$100 in credits. No card required. Connect in minutes.