Chat Service API Types

AI Dev Kit supports multiple chat service APIs, each with different capabilities and use cases.

Available Chat Services

Default (Responses API / Chat Completions)

Automatically selects the best API for your provider:

• OpenAI: Uses Responses API (new unified API)

• Other providers: Falls back to Chat Completions API

When to use:

• Most general-purpose applications

• You want automatic API selection

• Standard request/response pattern

Example:

Chat Service API: Default
Model: gpt-4o

Chat Completions API

Traditional stateless chat API:

• Supported by most providers (OpenAI, Anthropic, Google, etc.)

• Client manages conversation history

• Full control over context

When to use:

• Need cross-provider compatibility

• Want full control over conversation state

• Building custom conversation management

Example:

Chat Service API: ChatCompletions
Model: claude-3-5-sonnet-20241022
Conversation Store: LocalFile

Assistants API (OpenAI)

Stateful assistant with server-side management:

• OpenAI manages conversation threads

• Built-in tools (file search, code interpreter)

• Persistent across sessions

When to use:

• Building with OpenAI specifically

• Need file search or code interpreter

• Want server-side conversation persistence

Example:

Chat Service API: AssistantsApi
Assistant Id: asst_xxxxx
Conversation Store: ThreadsApi
Enable File Search: true
Enable Code Interpreter: true

Realtime API (OpenAI)

Low-latency bidirectional voice conversations:

• WebSocket-based streaming

• Optimized for voice interactions

• Real-time audio processing

When to use:

• Building voice assistants

• Need ultra-low latency

• Real-time audio conversations

Example:

Chat Service API: RealtimeApi
Model: gpt-4o-realtime-preview
Enable Input Audio: true
Enable Output Audio: true

Feature Comparison

Feature	Chat Completions	Assistants API	Realtime API
Provider Support	All	OpenAI only	OpenAI only
Conversation Management	Client-side	Server-side	Server-side
Streaming	✅ Text	✅ Text	✅ Audio
Function Calling	✅	✅	✅
File Search	❌	✅	❌
Code Interpreter	❌	✅	❌
Voice I/O	Via separate APIs	Via separate APIs	✅ Native
Latency	Medium	Medium-High	Ultra-low

Configuration Examples

Multi-Provider Chat App

// Use Chat Completions for maximum compatibility
settings.ChatServiceApi = ChatService.ChatCompletions;

// Switch models at runtime
agent.Model = "gpt-4o"; // OpenAI
agent.Model = "claude-3-5-sonnet-20241022"; // Anthropic
agent.Model = "gemini-2.0-flash-exp"; // Google

OpenAI Assistant with Tools

settings.ChatServiceApi = ChatService.AssistantsApi;
settings.AssistantsApiOptions.AssistantId = "asst_xxxxx";
settings.EnableFileSearch = true;
settings.EnableCodeInterpreter = true;

behaviour.ConversationStoreType = ConversationStoreType.ThreadsApi;

Voice Assistant

settings.ChatServiceApi = ChatService.RealtimeApi;
settings.Model = "gpt-4o-realtime-preview";
settings.EnableInputAudio = true;
settings.EnableOutputAudio = true;
settings.Voice = "alloy";

behaviour.Stream = true;

Switching APIs at Runtime

// Not recommended - requires reinitializing agent
// Better to create separate agents for different APIs

// If you must switch:
agent.Dispose();
settings.ChatServiceApi = ChatService.AssistantsApi;
agent = new Agent(settings, behaviour);
await agent.InitializeAsync();

Next Steps

• Agent Settings - Configure models and parameters

• Conversation Store - Choose persistence strategy

• Tools - Add function calling capabilities