How langgenius/dify works
Dify is an LLM application development platform: users configure workflows, prompts, models, and knowledge sources, then run deployed applications or RAG pipelines. Its backend includes application-definition and invocation components, retrieval, and persistence; the README also identifies agent capabilities and observability. The map combines documented product capabilities with relationships shown in sampled source or the source index; unsampled subsystems are represented only where the README or tree supports them.
Main entry points: generate_task_pipeline.py Workflow Runner · rag_pipeline_run_task.py RAG Pipeline Task · app Console API · workflow.py Workflow Definition
Components 18
App Authoring
- Console API
api/controllers/console/app - Workflow Definition
api/models/workflow.py - Prompt IDE
api/core/app/app_config/easy_ui_based_app/prompt_template/manager.py
Execution Runtime
- App Definition Query
api/services/app_definition_query_service.py - Invocation Context
api/core/app/entities/app_invoke_entities.py - Workflow Runner
api/core/app/apps/workflow/generate_task_pipeline.py - Run Event Queue
api/core/app/apps/pipeline/pipeline_queue_manager.py - Agent Runtime
api/clients/agent_backend/client.py
Knowledge and RAG
- RAG Pipeline
api/controllers/console/datasets/rag_pipeline/rag_pipeline_workflow.py - RAG Pipeline Task
api/tasks/rag_pipeline/rag_pipeline_run_task.py - Retrieval Service
api/core/rag/datasource/retrieval_service.py
Platform Services
- Model Providers
api/controllers/console/workspace/model_providers.py - File Storage
api/extensions/ext_storage.py - Observability
api/configs/observability/otel/otel_config.py
Persistence
- Definition Repository
api/repositories/app_definition_query_repository.py - Dataset Models
api/models/dataset.py - Database Extension
api/extensions/ext_database.py
Other components
- App Builder
Connections 14
- App Builder → Console API: configures apps
README.md - Console API → Workflow Definition: edits workflows
README.md - App Builder → Prompt IDE: crafts prompts
README.md - App Builder → Model Providers: selects models
README.md - Workflow Definition → Workflow Runner: defines runs
README.md - App Definition Query → Definition Repository: queries definitions
api/services/app_definition_query_service.py - App Builder → RAG Pipeline: builds RAG
README.md - RAG Pipeline Task → Retrieval Service: uses retrieval
- RAG Pipeline Task → Dataset Models: reads dataset
api/tasks/rag_pipeline/rag_pipeline_run_task.py - RAG Pipeline Task → Workflow Runner: builds pipeline
api/tasks/rag_pipeline/rag_pipeline_run_task.py - Workflow Runner → Run Event Queue: publishes events
- Workflow Runner → Invocation Context: uses context
- App Builder → Observability: inspects runs
- File Storage → Database Extension: persists metadata