Dify x HK OSS
Dify x HK OSS
From Zero to Production: Building GenAI Applications
Workflow Orchestration, Knowledge Base Management, and Production Deployment with Dify
An open-source LLM application development platform connecting large model capabilities to production-grade applications.
Driven by open source, GitHub Top 100 Project
| Core Logic | Deterministic If-Then-Else control |
| Data Processing | Structured (SQL/JSON) Schema-first |
| Testing | Unit Tests (Pass/Fail) Logic coverage |
| Core Logic | Probabilistic Prompt + Context guidance |
| Data Processing | Unstructured (Embedding) Semantic retrieval |
| Testing | Evaluation (Eval) Accuracy, hallucination, relevance |
Visually build complex AI processing pipelines, from simple conversations to multi-step automation.
Conversational AI with context memory, ideal for customer service and Q&A scenarios
Autonomous task execution with tool calling, ideal for research and analysis
Multi-turn conversation flows with conditional branching, ideal for forms and wizards
Batch processing, automated tasks, stateless execution, ideal for data processing
| Category | Node | Description |
|---|---|---|
| I/O | Start | Define input parameter schema, trigger workflow execution |
| End | Define output format, terminate workflow | |
| Processing | Code | Python/Node.js execution, data transformation |
| Template | String template rendering, prompt construction | |
| AI | LLM | Large model invocation, multi-provider support, parameter config |
| Knowledge | Knowledge base retrieval, RAG context injection | |
| Control Flow | If/Else | Conditional branching, expression-based routing |
| Iteration | List iteration, batch data processing |
Complete RAG solution: full-cycle management from document ingestion to intelligent retrieval.
| Strategy | Use Case | Chunk Size | Overlap |
|---|---|---|---|
| Recursive | General document processing | 500-1000 chars | 50-100 |
| Semantic | Technical content, code | Sentence-level | N/A |
| Fixed | Structured data | Custom | 0 |
| Markdown | Documentation, Wiki | Header hierarchy | Context |
Deploy Dify quickly with Docker, configure for production environments.
# 1. Clone repository
git clone https://github.com/langgenius/dify.git
cd dify/docker
# 2. Configure environment
cp .env.example .env
# Edit .env to set OPENAI_API_KEY, SECRET_KEY, etc.
# 3. Start services
docker compose up -d
# 4. Access console
open http://localhost/install
| Variable | Description | Example |
|---|---|---|
| CONSOLE_API_URL | Console API URL | http://localhost:5001 |
| APP_API_URL | App API URL | http://localhost:5001 |
| DB_DATABASE | PostgreSQL database name | dify |
| REDIS_HOST | Redis host | redis |
| VECTOR_STORE | Vector database type | weaviate/qdrant/milvus |
| S3_BUCKET_NAME | File storage bucket name | dify-files |
Advanced Practice · Hands-on Projects
Extend Workflow capabilities with Python/Node.js
def main(payload: dict) -> dict:
country = payload.get("country", "")
score = payload.get("risk_score", 0)
# Custom routing logic
if country in ["US", "UK"] and score < 0.3:
return {"route": "fast-lane", "reason": "low_risk"}
elif score > 0.8:
return {"route": "block", "reason": "high_risk"}
return {"route": "standard", "score": score}
Finding the sweet spot between visual orchestration and code
| Dimension | Low-Code Orchestration (DSL) | Pro-Code |
|---|---|---|
| Core Use | Flow orchestration, conditionals, RAG pipelines | Complex algorithms, custom business logic |
| Delivery Speed | Fast iteration, business users can participate | Requires dev cycle, but highly flexible |
| Maintainability | Visual and intuitive, easy to understand | Requires documentation, code review |
| Best Practice | 80% DSL + 20% Code — Glue logic visually, core computation in code | |
Extend Dify's capabilities, integrate third-party services and custom tools.
Connect OpenAI, Claude, Llama and other LLMs
Google Search, code execution, database queries
Function Calling, ReAct, Plan & Execute
Systematically evaluate retrieval quality and generation effectiveness
| Recall | Proportion of relevant documents successfully retrieved |
| Precision | Proportion of retrieved results that are relevant |
| MRR | Mean Reciprocal Rank of first relevant document |
| NDCG | Ranking-aware weighted relevance score |
| Faithfulness | Is the answer based on retrieved context? |
| Relevance | How well does the answer match the question? |
| Completeness | Does it cover all aspects of the question? |
| Hallucination Rate | Proportion of fabricated information |
Evolution from simple conversations to autonomous task execution
Complete journey from raw documents to usable memories
Protect your applications from abuse and attacks
From logs to metrics, full-stack tracking of application performance
Execution traces, variable values, error stacks
Latency, token consumption, success rate
Cross-node call chains, dependencies
Systematically identify and resolve issues
Grow from user to contributor, build the open source ecosystem together.
Submit PRs, fix bugs, implement new features
Create Model/Tool/Agent plugins for the community
Help localize documentation, lower barriers to entry
Automatically transcribe Apple Watch voice memos, generate structured notes and sync to Notion.
chatbot:
system_prompt: |
You are a GDPR compliance assistant. You must:
1. Cite specific articles when answering
2. Label confidence (High/Medium/Low)
3. Recommend legal consultation for complex cases
4. Provide information only, not legal advice
guardrails:
confidence_threshold: 0.6
fallback: "Need more context, connecting you to a human agent."
Questions? Feel free to reach out anytime.