FEAT: NEW WORKFLOW ENGINE

[takatost]

Introducing Dify Workflow! 🎉

The much-anticipated workflow feature is here: In a nutshell, workflow provides a visual canvas for defining complex tasks as smaller, manageable steps (nodes). This reduces reliance on prompt engineering and LLM agent capabilities, taking the stability and reproducibility of your LLM applications to the next level by letting you be in control.
There are two Workflow application types with this update:

• Workflow App
  Targeting Automation and Batch Processing: This is ideal for translation, data analysis, content generation, email automation, and more.
  

• Chatflow App (A Sub-Type of Chatbot)
  For Conversational Applications: Suitable for customer service, semantic search, and more conversational apps requiring multi-step logic in crafting the response.
  Compared to the regular Workflow app type, Chatflow adds chat-specific features such as conversation history support (Memory), tagged replies, an Answer node type for streaming responses, and support for rich text and images.

  

For more information, please visit: https://docs.dify.ai/features/workflow/introduce

Other Enhancements:

• Optimized UI flow for app creation.

• Conversion support from various basic application types to Workflow-based applications.

  • Basic / Expert mode Chatbot apps → Chatflow

  • Text Generator → Workflow

    

• Dify's official app templates are now available in self-hosted mode.

• Support for adding descriptions to applications.

• Support for porting applications in and out of Dify with DSL.

• Under the hood, we also refactored the underlying execution logic of all app types for cleaner architecture and a tidier repo.

Update Guide

If you need to upgrade from 0.6.0-preview-workflow.1, you will need to connect to PostgreSQL and execute the following SQL (migration inserted in the main branch) to ensure data integrity.

ALTER TABLE dataset_keyword_tables ADD COLUMN data_source_type VARCHAR(255) NOT NULL DEFAULT 'database';
ALTER TABLE embeddings ADD COLUMN provider_name VARCHAR(40) NOT NULL DEFAULT '';
ALTER TABLE embeddings DROP CONSTRAINT embedding_hash_idx;
ALTER TABLE embeddings ADD CONSTRAINT embedding_hash_idx UNIQUE (model_name, hash, provider_name);

Docker compose deployments:

1. Get the latest code from the main branch:

   git checkout main
   git pull origin main

2. Go to the next step and update to the latest image:

   cd docker
   docker compose up -d

3. We also moved the agent data within the database, Execute the below script to complete the migrate: (NEW)

   docker compose exec api flask convert-to-agent-apps

Source Code deployments:

1. Stop API server, Worker and Web frontend Server.

2. Get the latest code from the main branch:

   git checkout main
   git pull origin main

3. Update Python dependencies:

   cd api
   pip install -r requirements.txt

4. Then, let's run the migration script:

   flask db upgrade

5. We also moved the agent data within the database, Execute the below script to complete the migrate: (NEW)

   flask convert-to-agent-apps

6. Finally, run API server, Worker and Web frontend Server again.

[sentry-io]

Suspect Issues

This pull request was deployed and Sentry observed the following issues:

• ‼️ InvokeError: [zhipuai] Error: Error code: 400, with error text {"error":{"code":"1214","message":"top_p参数非法。请检... core.model_runtime.model_providers.__base.large... View Issue
• ‼️ ProviderTokenNotInitError: Model gpt-3.5-turbo credentials is not initialized. core.model_manager in _fetch_credentials_from_b... View Issue
• ‼️ AttributeError: 'NoneType' object has no attribute 'strategy' core.app.apps.agent_chat.app_runner in run View Issue
• ‼️ TypeError: can only concatenate str (not "NoneType") to str core.app.apps.base_app_runner in _handle_invoke... View Issue
• ‼️ AttributeError: 'DatasetRetrieval' object has no attribute 'to_dataset_retriever_tool' core.tools.tool.dataset_retriever_tool in get_d... View Issue

_{Did you find this useful? React with a 👍 or 👎}