agents / Exports / BaseAIProvider

Class: BaseAIProvider<TConfig, TMessage, TResponse>

Base AI provider implementation with proper type constraints All AI providers should extend this class

======================================== CRITICAL IMPLEMENTATION GUIDELINES

ALL AI PROVIDER IMPLEMENTATIONS (OpenAI, Anthropic, Google, etc.) MUST:

1. EXTEND THIS CLASS:

   export class OpenAIProvider extends BaseAIProvider {
       override readonly name = 'openai';
       override readonly version = '1.0.0';

2. USE IMPORTS FROM @robota-sdk/agents:

   import { BaseAIProvider } from '@robota-sdk/agents';
   import type {
       UniversalMessage,
       ChatOptions,
       ToolCall,
       ToolSchema,
       AssistantMessage
   } from '@robota-sdk/agents';

3. USE OVERRIDE KEYWORD FOR ALL INHERITED METHODS:

   • override async chat(...)
   • override async *chatStream(...)
   • override supportsTools()
   • override validateConfig()
   • override async dispose()

4. DO NOT REDEFINE TYPES THAT EXIST IN @robota-sdk/agents:

   • UniversalMessage
   • ChatOptions
   • ToolCall
   • ToolSchema
   • AssistantMessage
   • SystemMessage
   • UserMessage
   • ToolMessage

5. HANDLE MESSAGE CONTENT PROPERLY:

   • For tool calls: content should be null (not empty string)
   • For regular messages: content can be string or null
   • Always preserve null values from API responses

6. CALL SUPER() IN CONSTRUCTOR:

   constructor(options: ProviderOptions) {
       super();
       // provider-specific initialization
   }

This ensures ExecutionService can properly identify providers and prevents type conflicts across the codebase.

========================================

Type parameters

Name	Type	Description
TConfig	ProviderConfig	Provider configuration type (defaults to ProviderConfig for type safety)
TMessage	UniversalMessage	Message type (defaults to UniversalMessage for backward compatibility)
TResponse	UniversalMessage	Response type (defaults to UniversalMessage for backward compatibility)

Implements

• TypeSafeAIProvider<TConfig, TMessage, TResponse>

Table of contents

Constructors

• constructor

Properties

• name
• version

Methods

• configure
• chat
• chatStream
• supportsTools
• validateConfig
• dispose

Constructors

constructor

• new BaseAIProvider<TConfig, TMessage, TResponse>(): BaseAIProvider<TConfig, TMessage, TResponse>

Type parameters

Name	Type
TConfig	ProviderConfig
TMessage	UniversalMessage
TResponse	UniversalMessage

Returns

BaseAIProvider<TConfig, TMessage, TResponse>

Properties

name

• Readonly Abstract name: string

Implementation of

TypeSafeAIProvider.name

Defined in

packages/agents/src/abstracts/base-ai-provider.ts:110

---

version

• Readonly Abstract version: string

Implementation of

TypeSafeAIProvider.version

Defined in

packages/agents/src/abstracts/base-ai-provider.ts:111

Methods

configure

▸ configure(config): Promise<void>

Configure the provider with type-safe configuration

Parameters

Name	Type
config	TConfig

Returns

Promise<void>

Implementation of

TypeSafeAIProvider.configure

Defined in

packages/agents/src/abstracts/base-ai-provider.ts:117

---

chat

▸ chat(messages, options?): Promise<TResponse>

Each provider must implement chat using their own native SDK types internally

Parameters

Name	Type	Description
messages	TMessage[]	Array of messages from conversation history
options?	ChatOptions	Chat options including tools, model settings, etc.

Returns

Promise<TResponse>

Promise resolving to a response

Implementation of

TypeSafeAIProvider.chat

Defined in

packages/agents/src/abstracts/base-ai-provider.ts:128

---

chatStream

▸ chatStream(messages, options?): AsyncIterable<TResponse, any, any>

Each provider must implement streaming chat using their own native SDK types internally

Parameters

Name	Type	Description
messages	TMessage[]	Array of messages from conversation history
options?	ChatOptions	Chat options including tools, model settings, etc.

Returns

AsyncIterable<TResponse, any, any>

AsyncIterable of response chunks

Implementation of

TypeSafeAIProvider.chatStream

Defined in

packages/agents/src/abstracts/base-ai-provider.ts:136

---

supportsTools

▸ supportsTools(): boolean

Default implementation - most modern providers support tools

Returns

boolean

true if tool calling is supported

Implementation of

TypeSafeAIProvider.supportsTools

Defined in

packages/agents/src/abstracts/base-ai-provider.ts:142

---

validateConfig

▸ validateConfig(): boolean

Default implementation - providers can override for specific validation

Returns

boolean

true if configuration is valid

Implementation of

TypeSafeAIProvider.validateConfig

Defined in

packages/agents/src/abstracts/base-ai-provider.ts:150

---

dispose

▸ dispose(): Promise<void>

Default implementation - providers can override for cleanup

Returns

Promise<void>

Implementation of

TypeSafeAIProvider.dispose

Defined in

packages/agents/src/abstracts/base-ai-provider.ts:157