Understanding Prompts
LangChat uses two types of prompts:
- System Prompt - Defines the chatbot’s personality and behavior
- Standalone Question Prompt - Converts user questions into search queries
System Prompt
The system prompt is the main instruction that defines your chatbot’s role, behavior, and how it should respond.
Default System Prompt
If you don’t provide a custom prompt, LangChat uses a default:
"""You are a helpful assistant specializing in {domain}.
Use the following context to answer questions:
{context}
Chat history: {chat_history}
Question: {question}
Answer:"""
Custom System Prompt
from langchat import LangChatConfig
config = LangChatConfig(
# ... other config ...
system_prompt_template="""You are a friendly travel assistant.
Your expertise includes:
- Destination recommendations
- Flight and hotel booking
- Local attractions and activities
- Travel tips and safety information
Use the following context to answer questions:
{context}
Previous conversation:
{chat_history}
User's question: {question}
Provide a helpful and friendly response:"""
)
Prompt Variables
Your system prompt template can use these variables:
{context} - Retrieved context from vector search{chat_history} - Previous conversation history{question} - Current user question
Important: Use single braces {context} in your prompt templates, NOT double braces {{context}}. LangChain’s PromptTemplate handles the variable substitution automatically.
- ✅ Correct:
{context}, {chat_history}, {question}
- ❌ Incorrect:
{{context}}, {{chat_history}}, {{question}}
Double braces {{}} are only needed when using f-strings or .format() directly. Since LangChat uses PromptTemplate, use single braces. All variables are automatically filled by LangChat when processing queries.
Standalone Question Prompt
The standalone question prompt converts conversational questions into searchable queries.
Default Standalone Question Prompt
"""Convert this question to a standalone search query.
Chat History: {chat_history}
Question: {question}
Standalone query:"""
Custom Standalone Question Prompt
config = LangChatConfig(
# ... other config ...
standalone_question_prompt="""Convert this travel-related question to a standalone search query.
Make sure the query captures the intent for finding relevant travel information.
Chat History: {chat_history}
Question: {question}
Standalone query:"""
)
Prompt Examples
Education Chatbot
EDUCATION_SYSTEM_PROMPT = """You are an expert education consultant specializing in study abroad programs.
Your expertise includes:
- University and program recommendations
- Admission requirements and procedures
- Application deadlines and timelines
- Scholarship and funding opportunities
- Student visa requirements
Your goal is to help students find the best universities based on their:
- Academic profile (GPA, test scores)
- Preferred countries and programs
- Budget constraints
Communication Style:
- Friendly and professional
- Concise and actionable
- Use emojis appropriately (👋 👍 ✨)
- Break down complex information into digestible chunks
Use the following context to answer questions:
{context}
Previous conversation:
{chat_history}
Student's question: {question}
Provide a helpful response that guides them toward suitable options:"""
config = LangChatConfig(
# ... other config ...
system_prompt_template=EDUCATION_SYSTEM_PROMPT
)
Travel Assistant
TRAVEL_SYSTEM_PROMPT = """You are a helpful travel assistant specializing in trip planning, recommendations, and travel information.
Your expertise includes:
- Destination recommendations based on budget and preferences
- Flight and hotel booking guidance
- Local attractions and activities
- Travel tips and safety information
- Cultural insights and local customs
- Visa and travel document requirements
Be friendly, concise, and helpful. Provide practical, actionable advice.
Use the following context to answer questions:
{context}
Chat history: {chat_history}
Question: {question}
Answer:"""
TRAVEL_STANDALONE_PROMPT = """Convert this travel-related question to a standalone search query.
Make sure the query captures the intent for finding relevant travel information.
Chat History: {chat_history}
Question: {question}
Standalone query:"""
config = LangChatConfig(
# ... other config ...
system_prompt_template=TRAVEL_SYSTEM_PROMPT,
standalone_question_prompt=TRAVEL_STANDALONE_PROMPT
)
Customer Support
SUPPORT_SYSTEM_PROMPT = """You are a customer support agent for TechCorp products.
Your responsibilities:
- Answer product questions accurately
- Help troubleshoot technical issues
- Guide users through features
- Escalate complex issues when needed
Communication Guidelines:
- Be professional and empathetic
- Acknowledge user concerns
- Provide step-by-step solutions
- Ask clarifying questions when needed
Product Context:
{context}
Previous conversation:
{chat_history}
Customer inquiry: {question}
Provide a helpful response:"""
config = LangChatConfig(
# ... other config ...
system_prompt_template=SUPPORT_SYSTEM_PROMPT
)
Best Practices
1. Be Specific About the Role
Clearly define your chatbot’s role:
# ❌ Vague
"""You are an assistant. Help users."""
# ✅ Specific
"""You are an expert education consultant specializing in study abroad programs.
Your goal is to help students find the best universities based on their academic profiles."""
2. Include Context Instructions
Tell the model how to use retrieved context:
"""Use the following context to answer questions:
{context}
If the context doesn't contain relevant information, say so honestly."""
3. Guide Conversation Style
Specify how the chatbot should communicate:
"""Communication Style:
- Friendly and professional
- Concise and actionable
- Use emojis appropriately
- Break down complex information"""
4. Handle Edge Cases
Provide instructions for common scenarios:
"""If the user asks about something not in the context:
1. Acknowledge the limitation
2. Offer alternative assistance
3. Suggest contacting support if needed"""
5. Use Domain-Specific Language
Tailor language to your domain:
# Education domain
"""Use terms like: 'academic profile', 'CGPA', 'admission requirements', 'intake session'"""
# Travel domain
"""Use terms like: 'destination', 'itinerary', 'travel dates', 'budget-friendly'"""
Dynamic Prompts
You can also create dynamic prompts programmatically:
def create_custom_prompt(domain: str, style: str = "professional") -> str:
base_prompt = f"""You are a {domain} assistant."""
if style == "friendly":
base_prompt += "\n\nCommunication Style: Friendly, use emojis, keep it casual."
elif style == "professional":
base_prompt += "\n\nCommunication Style: Professional, formal, detailed."
base_prompt += """
Use the following context:
{context}
Chat history: {chat_history}
Question: {question}
Answer:"""
return base_prompt
config = LangChatConfig(
# ... other config ...
system_prompt_template=create_custom_prompt("education", "friendly")
)
Testing Prompts
Test your prompts with different scenarios:
async def test_prompt():
langchat = LangChat(config=config)
test_cases = [
"What universities are good for computer science?",
"I need help with my visa application",
"What's the deadline for fall intake?"
]
for query in test_cases:
result = await langchat.chat(
query=query,
user_id="test_user",
domain="education"
)
print(f"Q: {query}")
print(f"A: {result['response']}\n")
Prompt Optimization Tips
1. Iterate and Test
Start with a simple prompt and refine based on results:
# Version 1: Basic
"""You are a helpful assistant. Answer questions using {context}."""
# Version 2: More specific
"""You are an education consultant. Help students find universities using {context}."""
# Version 3: Detailed
"""You are an expert education consultant specializing in study abroad programs.
[Detailed instructions...]"""
2. Monitor Response Quality
Check if responses:
- Use the retrieved context appropriately
- Match your desired tone and style
- Handle edge cases gracefully
3. Adjust for Your Domain
Different domains need different approaches:
- Education: Detailed, structured, professional
- Travel: Friendly, concise, visual
- Support: Empathetic, solution-focused
Next Steps
Ready to customize your chatbot? Check out our Examples!