Assistants

​

Overview

• Custom Assistant: Build using your choice of LLMs and tools
• OpenAI Assistant: Leverage OpenAI’s Assistant API with built-in features
• Azure Assistant: Coming soon for Azure OpenAI deployments

​

Assistant Types

​

Custom Assistant

• Choose from multiple LLM providers (OpenAI, Anthropic, Cohere, etc.)
• Full control over prompt engineering
• Custom tool integration
• Flexible conversation memory
• Cost-effective with open-source models

• Domain-specific assistants
• Multi-provider deployments
• Custom business logic
• Budget-conscious applications

​

OpenAI Assistant

• Built-in function calling
• Code interpreter for data analysis
• File search capabilities
• Multi-turn conversations with context
• Thread management
• Vector store integration

• gpt-4o and gpt-4o-mini
• gpt-4-turbo and gpt-4-turbo-preview
• gpt-4
• gpt-3.5-turbo

• Customer support with knowledge base
• Code generation and debugging
• Data analysis and visualization
• Document processing and Q&A

​

Creating a Custom Assistant

Available LLMs:
├─ OpenAI (GPT-4, GPT-3.5)
├─ Anthropic (Claude)
├─ Google (PaLM, Gemini)
├─ Cohere
├─ HuggingFace
└─ Local Models (Ollama, LM Studio)

You are a helpful customer support assistant for TechCorp.

Your responsibilities:
- Answer questions about products and services
- Help troubleshoot common issues
- Escalate complex problems to human agents
- Always be professional and empathetic

Knowledge base:
- Product catalog
- Return policy (30 days)
- Shipping information
- Technical documentation

​

Creating an OpenAI Assistant

{
  "name": "Support Bot",
  "description": "Helps customers with product questions",
  "model": "gpt-4.1",
  "instructions": "You are a friendly and knowledgeable support agent..."
}

{
  "name": "get_order_status",
  "description": "Retrieves the current status of a customer order",
  "parameters": {
    "type": "object",
    "properties": {
      "order_id": {
        "type": "string",
        "description": "The unique order identifier"
      }
    },
    "required": ["order_id"]
  }
}

{
  "temperature": 0.7,
  "top_p": 1.0,
  "max_prompt_tokens": 4096,
  "max_completion_tokens": 2048,
  "metadata": {
    "version": "1.0",
    "department": "customer_support"
  }
}

​

Using Assistants in Chatflows

​

Custom Assistant

1. Add Custom Assistant node to canvas
2. Select your saved assistant
3. Connect to chat interface
4. Configure additional settings

​

OpenAI Assistant

1. Add OpenAI Assistant node to canvas
2. Select assistant from your OpenAI account
3. Configure thread management
4. Connect to chat interface

[User Input] → [OpenAI Assistant] → [Response]
                       ↓
              [Vector Store / Tools]

​

Thread Management (OpenAI Assistants)

​

Create Thread

import requests

API_URL = "http://localhost:3000/api/v1/openai-assistants/thread"

payload = {
    "assistantId": "asst_abc123"
}

response = requests.post(API_URL, json=payload)
thread = response.json()
print(f"Thread ID: {thread['id']}")

​

Send Message

message_payload = {
    "threadId": thread['id'],
    "message": "What is your return policy?"
}

response = requests.post(
    f"{API_URL}/message",
    json=message_payload
)
print(response.json())

​

Retrieve Thread History

history = requests.get(f"{API_URL}/thread/{thread['id']}/messages")
for message in history.json():
    print(f"{message['role']}: {message['content']}")

​

Assistant API

​

Create Custom Assistant

curl -X POST http://localhost:3000/api/v1/assistants/custom \
  -H "Authorization: Bearer <your_api_key>" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Support Assistant",
    "instructions": "You are a helpful support agent",
    "llmConfig": {
      "model": "gpt-4",
      "temperature": 0.7
    },
    "tools": ["web_search", "calculator"]
  }'

​

Create OpenAI Assistant

curl -X POST http://localhost:3000/api/v1/assistants/openai \
  -H "Authorization: Bearer <your_api_key>" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Code Helper",
    "model": "gpt-4.1",
    "instructions": "You help with coding questions",
    "tools": [{"type": "code_interpreter"}]
  }'

​

List Assistants

curl -X GET http://localhost:3000/api/v1/assistants \
  -H "Authorization: Bearer <your_api_key>"

​

Delete Assistant

curl -X DELETE http://localhost:3000/api/v1/assistants/{assistantId} \
  -H "Authorization: Bearer <your_api_key>"

​

Best Practices

​

Instruction Writing

✅ Good:
"You are a technical support assistant for SaaS products.
Always ask clarifying questions before troubleshooting.
Provide step-by-step solutions.
Escalate to human agents when users are frustrated."

❌ Bad:
"You help with technical stuff."

​

Tool Selection

• Customer Support: Knowledge base search, ticket creation
• Sales: Product catalog, pricing calculator, CRM integration
• Technical: Code execution, API documentation, debugging tools

​

File Organization

• Use descriptive file names
• Organize by topic or category
• Keep files updated
• Remove outdated information

​

Performance Optimization

• Use appropriate model for task complexity
• Limit file sizes for faster processing
• Cache frequent queries
• Monitor token usage and costs

​

Security

• Never include sensitive data in instructions
• Validate function call parameters
• Use credential management for API keys
• Implement rate limiting

​

Troubleshooting

​

Assistant Not Responding

• Check API key validity
• Verify model availability
• Review error logs
• Test with simple query

​

Incorrect Function Calls

• Review function definitions
• Improve instruction clarity
• Add examples to instructions
• Validate parameter schemas

​

File Search Not Working

• Verify files are uploaded
• Check vector store attachment
• Ensure file format is supported
• Review chunking configuration

​

High Costs

• Monitor token usage per request
• Optimize instructions length
• Use appropriate model tier
• Implement caching strategies

​

Related Resources

• Chatflows
• Tools
• OpenAI Assistant API
• Function Calling