Module: core/conversation

Type Aliases

ConversationMessage

Ƭ ConversationMessage: ConversationMessageType<"user", typeof UserMessage> | ConversationMessageType<"assistant", typeof AssistantMessage> | ConversationMessageType<"system", typeof SystemMessage> | ConversationMessageType<"functionCall", typeof FunctionCall> | ConversationMessageType<"functionResponse", typeof FunctionResponse>

A type that represents a conversation message.

Defined in

packages/ai-jsx/src/core/conversation.tsx:163

Variables

ConversationHistoryContext

• Const ConversationHistoryContext: Context<Node>

Sets the node that the

component will resolve to.

Defined in

packages/ai-jsx/src/core/conversation.tsx:65

Functions

AssistantMessage

▸ AssistantMessage(«destructured»): Node

Provide an Assistant Message to the LLM, for use within a ChatCompletion.

The assistant message tells the model what it has previously said. See https://platform.openai.com/docs/guides/gpt/chat-completions-api for more detail.

Example

   <ChatCompletion>
     <UserMessage>I'd like to cancel my account.</UserMessage>
     <AssistantMessage>Sorry to hear that. Can you tell me why?</AssistantMessage>
     <UserMessage>It's too expensive.</UserMessage>
   </ChatCompletion>

==> "Ok, thanks for that feedback. I'll cancel your account."

Parameters

Name	Type
«destructured»	Object
› children	Node
› metadata?	Record<string, Jsonifiable>

Returns

Node

Defined in

packages/ai-jsx/src/core/conversation.tsx:58

---

ConversationHistory

▸ ConversationHistory(_, «destructured»): null | string | number | boolean | Element<any> | Node[] | IndirectNode

Renders to the conversation history provided through ConversationHistoryContext.

Parameters

Name	Type
_	Object
«destructured»	ComponentContext

Returns

null | string | number | boolean | Element<any> | Node[] | IndirectNode

Defined in

packages/ai-jsx/src/core/conversation.tsx:70

---

Converse

▸ Converse(«destructured», «destructured»): RenderableStream

A component that appends messages to a conversation according to a reply function.

The reply function is invoked with the new messages produced from its own replies until it fails to return a conversational message (e.g. UserMessage or AssistantMessage). The first invocation includes the messages from the children prop.

Example

   <Converse reply={function (messages, fullConversation) {
       const lastMessage = messages[messages.length - 1];
       if (lastMessage.type === "user") {
         return (
           <ChatCompletion functions={functions}>
             {fullConversation.map(msg => msg.element)}
           </ChatCompletion>
         );
       }

       if (lastMessage.type === "functionCall") {
         return <EvaluateFunction name={lastMessage.element.name} args={lastMessage.element.args} />
       }

       return null;
     }>
     <ConversationHistory />
   </ChatCompletion>

   ==> 'Hello there!'

Parameters

Name	Type
«destructured»	Object
› children	Node
› reply	(messages: ConversationMessage[], fullConversation: ConversationMessage[]) => Renderable
«destructured»	ComponentContext

Returns

RenderableStream

Defined in

packages/ai-jsx/src/core/conversation.tsx:284

---

FunctionCall

▸ FunctionCall(«destructured»): string

Provide a function call to the LLM, for use within a ChatCompletion.

The function call tells the model that a function was previously invoked by the model. See https://platform.openai.com/docs/guides/gpt/chat-completions-api for more detail. When the model returns a function call, @{link ChatCompletion} returns a @{link FunctionCall} component.

Example

   <ChatCompletion>
     <UserMessage>What is 258 * 322?</UserMessage>
     <FunctionCall name="evaluate_math" args={expression: "258 * 322"} />
     <FunctionResponse name="evaluate_math">83076</FunctionResponse>
   </ChatCompletion>

   ==> "That would be 83,076."

Parameters

Name	Type
«destructured»	Object
› args	Record<string, null | string | number | boolean>
› name	string
› metadata?	Record<string, Jsonifiable>
› partial?	boolean

Returns

string

Defined in

packages/ai-jsx/src/core/conversation.tsx:101

---

FunctionResponse

▸ FunctionResponse(«destructured»): Element

Renders to the output of a previous FunctionCall component, for use within a ChatCompletion.

See https://platform.openai.com/docs/guides/gpt/chat-completions-api for more detail.

Example

   <ChatCompletion>
     <UserMessage>What is 258 * 322?</UserMessage>
     <FunctionCall name="evaluate_math" args={expression: "258 * 322"} />
     <FunctionResponse name="evaluate_math">83076</FunctionResponse>
   </ChatCompletion>

   ==> "That would be 83,076."

Parameters

Name	Type
«destructured»	Object
› children	Node
› name	string
› failed?	boolean
› metadata?	Record<string, Jsonifiable>

Returns

Element

Defined in

packages/ai-jsx/src/core/conversation.tsx:130

---

ShowConversation

▸ ShowConversation(«destructured», «destructured»): RenderableStream

Allows the presentation of conversational components (UserMessage et al) to be altered.

Also accepts an onComplete prop, which will be invoked once per render with the entire conversation.

Example

    <ShowConversation present={(msg) => msg.type === "assistant" && <>Assistant: {msg.element}</>}>
        <UserMessage>This is not visible.</UserMessage>
        <AssistantMessage>This is visible!</AssistantMessage>
    </ShowConversation>

    ==> 'Assistant: This is visible!'

Parameters

Name	Type
«destructured»	Object
› children	Node
› onComplete?	(conversation: ConversationMessage[], render: <TIntermediate>(renderable: Renderable, opts?: RenderOpts<TIntermediate, string>) => RenderResult<TIntermediate, string><TIntermediate>(renderable: Renderable, opts: RenderOpts<TIntermediate, PartiallyRendered[]>) => RenderResult<TIntermediate, PartiallyRendered[]>) => void | Promise<void>
› present?	(message: ConversationMessage, index: number) => Node
«destructured»	ComponentContext

Returns

RenderableStream

Defined in

packages/ai-jsx/src/core/conversation.tsx:327

---

SystemMessage

▸ SystemMessage(«destructured»): Node

Provide a System Message to the LLM, for use within a ChatCompletion.

The system message can be used to put the model in character. See https://platform.openai.com/docs/guides/gpt/chat-completions-api for more detail.

Example

   <ChatCompletion>
     <SystemMessage>You are a helpful customer service agent.</SystemMessage>
   </ChatCompletion>

Parameters

Name	Type
«destructured»	Object
› children	Node
› metadata?	Record<string, Jsonifiable>

Returns

Node

Defined in

packages/ai-jsx/src/core/conversation.tsx:20

---

UserMessage

▸ UserMessage(«destructured»): Node

Provide a User Message to the LLM, for use within a ChatCompletion.

The user message tells the model what the user has said. See https://platform.openai.com/docs/guides/gpt/chat-completions-api for more detail.

Example

   <ChatCompletion>
     <UserMessage>I'd like to cancel my account.</UserMessage>
   </ChatCompletion>

   ==> 'Sorry to hear that. Can you tell me why?

Parameters

Name	Type
«destructured»	Object
› children	Node
› metadata?	Record<string, Jsonifiable>
› name?	string

Returns

Node

Defined in

packages/ai-jsx/src/core/conversation.tsx:38