OpenAI - AI Assistant Fields
OpenAI's AI Assistant API (used in tools like ChatGPT and custom AI assistants) includes several key fields that define the assistant's behaviour, memory, and response style. Here’s a detailed breakdown of the primary fields:
Core Fields in OpenAI AI Assistant API
1. instructions (System Message / System Prompt)
-
-
Defines the assistant's role, behaviour, and response style.
-
Similar to a system prompt in ChatGPT.
-
Can specify tone, format, personality, and constraints.
-
Example: "instructions": "You are APE.I, a helpful AI chatbot for ConverseIQ, assisting users with CRM, live chat, and AI chatbot setup. Respond with concise and professional answers."
2. name (Assistant Name)
- The custom name assigned to the AI assistant.
- Helps personalise responses and branding.
Example: "name": "APE.I"
3. model (AI Model Selection)
Specifies which OpenAI model the assistant uses. Options include:"gpt-4-turbo" (faster, cost-efficient version of GPT-4) "gpt-4" (higher performance, more expensive) "gpt-3.5-turbo" (cost-effective, slightly less capable)
Example: "model": "gpt-4-turbo"
4. temperature (Creativity & Variability)
Controls how creative or predictable the assistant’s responses are.
Range: 0 (deterministic, always the same response) → 2 (highly creative, varied responses).
Recommended values:
- 0.2–0.5 for factual responses (support, troubleshooting).
- 0.7–1.2 for creative responses (marketing, storytelling).
Example: "temperature": 0.7
5. tools (External Function Calling)
- Defines which tools the assistant can use, such as:
- Code execution (Python calculations, data processing).
- Web browsing (fetching real-time info).
- File storage (handling and retrieving user files).
6. memory (Context & Long-Term Storage)
Determines if the assistant remembers previous interactions.
- Can store user preferences, past conversations, or ongoing tasks.
- Can be toggled on/off (API-based memory is still evolving).
Example (Storing User Preferences): "memory": { "enabled": true, "custom_data": { "preferred_language": "Australian English", "past_purchases": ["Dell XPS 15", "Logitech MX Master 3"] } }
file_ids (Attached Files for Processing) Allows the assistant to process uploaded files, such as:Spreadsheets (financial reports, CSVs). Documents (PDFs, contracts). Code files (Python scripts, logs).
Example: "file_ids": ["file-abc123xyz"]
response_format (Output Type) Controls whether responses are plain text, JSON, or structured formats. Useful for API-based interactions, chatbot responses, or data extractions.
Example (For JSON Response Format): "response_format": "json"
functions (Custom Function Calling) Allows the assistant to call specific predefined functions. Functions can be used to fetch CRM data, update tickets, trigger actions, etc.
Example (Defining a Function to Fetch User Details): "functions": [ { "name": "get_user_info", "description": "Fetches user profile details from the CRM.", "parameters": { "user_id": {"type": "string", "description": "The user's unique ID."} } } ]
Summary: Key Fields in OpenAI AI Assistant API
- instructions - Defines role, behaviour, response style
- name - Assigns a custom assistant name
- model - Chooses which AI model to use.
- temperature - Controls response creativity vs. predictability.
- tools - Enables function calling (code execution, web browsing, APIs).
- memory - Determines if the AI remembers past interactions.
- file_ids - Allows processing of uploaded files.
- response_format - Specifies output format (text, JSON, etc.).
- functions - Defines callable functions for external integrations.
ConverseIQ integrates with both OpenAI and OpenRouter to provide AI-powered conversational capabilities. However, it’s important to understand the current limitations and supported features around function calling within these integrations.
Function Calling with OpenAI
Our integration with OpenAI currently supports basic function calling. This allows you to define functions and pass them to the model, enabling it to decide when to call those functions based on the context of the conversation.
