mirror of
https://github.com/andy-stack/vaultkeeper-ai.git
synced 2026-07-22 06:42:03 +00:00
Introduce separate planning model setting to allow using different models for planning vs execution. Add visual countdown display when rate limits are hit, with improved retry delay parsing across providers (Claude, OpenAI, Gemini). Refactor settings tab into Views directory and enhance mobile layout for input controls.
414 lines
No EOL
16 KiB
TypeScript
414 lines
No EOL
16 KiB
TypeScript
import { BaseAIClass } from "AIClasses/BaseAIClass";
|
|
import type { IStreamChunk } from "Services/StreamingService";
|
|
import type { Conversation } from "Conversations/Conversation";
|
|
import type { ConversationContent } from "Conversations/ConversationContent";
|
|
import type { Attachment } from "Conversations/Attachment";
|
|
import { AIProvider, AIProviderURL } from "Enums/ApiProvider";
|
|
import { AIFunctionCall } from "AIClasses/AIFunctionCall";
|
|
import { fromString as aiFunctionFromString } from "Enums/AIFunction";
|
|
import type { IAIFunctionDefinition } from "AIClasses/FunctionDefinitions/IAIFunctionDefinition";
|
|
import type { ResponseEvent, ResponseOutputTextDelta, ResponseOutputItemDone, ResponseErrorEvent, ResponseFailedEvent, OpenAIFunctionTool, ResponsesAPIInput } from "./OpenAITypes";
|
|
import { Exception } from "Helpers/Exception";
|
|
import { ApiError, ApiErrorType } from "Types/ApiError";
|
|
import { MimeType, toMimeType } from "Enums/MimeType";
|
|
import { isTextFile } from "Enums/FileType";
|
|
import { MimeTypeToFileTypes } from "Enums/FileTypeMimeTypeMapping";
|
|
import { parseFunctionCall, parseFunctionResponse } from "Helpers/ResponseHelper";
|
|
|
|
export class OpenAI extends BaseAIClass {
|
|
|
|
private readonly SUPPORTED_MIMETYPES = [
|
|
MimeType.TEXT_PLAIN,
|
|
MimeType.APPLICATION_PDF,
|
|
MimeType.IMAGE_JPEG,
|
|
MimeType.IMAGE_PNG,
|
|
MimeType.IMAGE_WEBP
|
|
];
|
|
|
|
public constructor() {
|
|
super(AIProvider.OpenAI);
|
|
}
|
|
|
|
public async* streamRequest(conversation: Conversation, isPlanningAgent: boolean): AsyncGenerator<IStreamChunk, void, unknown> {
|
|
|
|
// Refresh file cache only if conversation has attachments
|
|
if (conversation.hasAttachments()) {
|
|
await this.aiFileService.refreshCache();
|
|
}
|
|
|
|
const systemPrompt = `${this.systemPrompt}\n\n${this.userInstruction}`;
|
|
|
|
const input = await this.extractContents(conversation.contents);
|
|
|
|
const tools = [{
|
|
type: "web_search"
|
|
}, ...this.mapFunctionDefinitions(this.toolDefinitions)];
|
|
|
|
const requestBody = {
|
|
model: this.model(isPlanningAgent),
|
|
instructions: systemPrompt,
|
|
input: input,
|
|
tools: tools,
|
|
stream: true
|
|
};
|
|
|
|
const headers = {
|
|
"Authorization": `Bearer ${this.apiKey}`,
|
|
"Content-Type": "application/json"
|
|
};
|
|
|
|
yield* this.streamingService.streamRequest(
|
|
AIProviderURL.OpenAI,
|
|
requestBody,
|
|
(chunk: string) => this.parseStreamChunk(chunk),
|
|
headers,
|
|
(error) => this.extractRetryDelay(error)
|
|
);
|
|
}
|
|
|
|
protected parseStreamChunk(chunk: string): IStreamChunk {
|
|
try {
|
|
// OpenAI Responses API sends "[DONE]" as the final message, which is not valid JSON
|
|
if (chunk.trim() === "[DONE]") {
|
|
return { content: "", isComplete: true };
|
|
}
|
|
|
|
const event = JSON.parse(chunk) as ResponseEvent;
|
|
|
|
let text = "";
|
|
let functionCall: AIFunctionCall | undefined = undefined;
|
|
let isComplete = false;
|
|
let shouldContinue = false;
|
|
|
|
// Handle different event types
|
|
switch (event.type) {
|
|
case "response.output_text.delta": {
|
|
// Text content streaming
|
|
const deltaEvent = event as ResponseOutputTextDelta;
|
|
text = deltaEvent.delta;
|
|
break;
|
|
}
|
|
|
|
case "response.refusal.delta": {
|
|
// Model refused to respond - treat as text for now
|
|
const refusalEvent = event as ResponseOutputTextDelta;
|
|
text = refusalEvent.delta;
|
|
break;
|
|
}
|
|
|
|
case "error":
|
|
case "response.error": {
|
|
const errorEvent = event as ResponseErrorEvent;
|
|
this.throwRetryableError(
|
|
errorEvent.message,
|
|
errorEvent.code || undefined,
|
|
ApiErrorType.SERVER_ERROR
|
|
);
|
|
break;
|
|
}
|
|
|
|
case "response.failed": {
|
|
const errorEvent = event as ResponseFailedEvent;
|
|
this.throwRetryableError(
|
|
errorEvent.response?.error?.message || "Response failed",
|
|
errorEvent.response?.error?.code || undefined,
|
|
ApiErrorType.SERVER_ERROR
|
|
);
|
|
break;
|
|
}
|
|
|
|
case "response.function_call_arguments.delta": {
|
|
// Function call arguments streaming - we can ignore these
|
|
// as we'll get the complete call in the "done" event
|
|
break;
|
|
}
|
|
|
|
case "response.function_call_arguments.done": {
|
|
// Function call arguments streaming - we can ignore these
|
|
// The complete function call info comes in response.output_item.done
|
|
break;
|
|
}
|
|
|
|
case "response.completed":
|
|
case "response.done": {
|
|
// Response completed
|
|
isComplete = true;
|
|
break;
|
|
}
|
|
|
|
case "response.output_item.done": {
|
|
// Complete output item received - this includes function calls with name
|
|
const itemDoneEvent = event as ResponseOutputItemDone;
|
|
|
|
// Check if this is a function call
|
|
if (itemDoneEvent.item.type === "function_call" &&
|
|
itemDoneEvent.item.name &&
|
|
itemDoneEvent.item.arguments) {
|
|
try {
|
|
const args = JSON.parse(itemDoneEvent.item.arguments) as Record<string, unknown>;
|
|
functionCall = new AIFunctionCall(
|
|
aiFunctionFromString(itemDoneEvent.item.name),
|
|
args as Record<string, object>,
|
|
itemDoneEvent.item.call_id || itemDoneEvent.item_id,
|
|
undefined // thoughtSignature not used by OpenAI
|
|
);
|
|
// When we receive a function call, we should continue the conversation
|
|
shouldContinue = true;
|
|
} catch (error) {
|
|
Exception.log(error);
|
|
}
|
|
}
|
|
break;
|
|
}
|
|
|
|
case "response.created":
|
|
case "response.in_progress":
|
|
case "response.content_part.added":
|
|
case "response.content_part.done":
|
|
case "response.output_item.added":
|
|
case "response.output_text.done":
|
|
case "response.web_search_call.in_progress":
|
|
case "response.web_search_call.searching":
|
|
case "response.web_search_call.completed":
|
|
// These events can be used for more granular tracking if needed
|
|
// For now, we handle content through the delta events
|
|
break;
|
|
|
|
default:
|
|
// log in dev but just ignore unhandled cases in prod
|
|
Exception.log(`Unknown event type: ${event.type}`);
|
|
break;
|
|
}
|
|
|
|
return {
|
|
content: text,
|
|
isComplete: isComplete,
|
|
functionCall: functionCall,
|
|
shouldContinue: shouldContinue,
|
|
};
|
|
} catch (error) {
|
|
return this.createErrorChunk(error);
|
|
}
|
|
}
|
|
|
|
protected async extractContents(conversationContent: ConversationContent[]): Promise<ResponsesAPIInput[]> {
|
|
const results: ResponsesAPIInput[] = [];
|
|
|
|
for (const content of this.filterConversationContents(conversationContent)) {
|
|
const contentToExtract = content.content ?? "";
|
|
|
|
// Case 1: Assistant message with function call
|
|
if (content.functionCall) {
|
|
const parsedContent = parseFunctionCall(content.functionCall);
|
|
|
|
if (parsedContent) {
|
|
// Check if function call has required id field (for OpenAI Responses API)
|
|
if (parsedContent.functionCall.id && parsedContent.functionCall.id.trim() !== "") {
|
|
// Add assistant text message if present
|
|
if (contentToExtract.trim() !== "") {
|
|
results.push({
|
|
role: content.role as "user" | "assistant",
|
|
content: contentToExtract
|
|
});
|
|
}
|
|
|
|
// Add function call as separate input item
|
|
results.push({
|
|
type: "function_call",
|
|
call_id: parsedContent.functionCall.id,
|
|
name: parsedContent.functionCall.name,
|
|
arguments: JSON.stringify(parsedContent.functionCall.args)
|
|
});
|
|
} else {
|
|
// No id (from other provider or legacy) - convert to text message
|
|
const legacyText = this.convertFunctionCallToText(parsedContent);
|
|
const messageContent = contentToExtract.trim();
|
|
const combinedContent = messageContent !== ""
|
|
? `${messageContent}\n\n${legacyText}`
|
|
: legacyText;
|
|
|
|
results.push({
|
|
role: content.role as "user" | "assistant",
|
|
content: combinedContent
|
|
});
|
|
}
|
|
} else {
|
|
// Fall back to regular message if parsing fails
|
|
results.push({
|
|
role: content.role as "user" | "assistant",
|
|
content: contentToExtract.trim() !== "" ? contentToExtract : "Error parsing function call"
|
|
});
|
|
}
|
|
continue;
|
|
}
|
|
|
|
// Case 2: Binary file attachments
|
|
if (content.attachments && content.attachments.length > 0) {
|
|
const { formattedParts, uploadErrors } = await this.processAttachments<ResponsesAPIInput>(
|
|
content.attachments,
|
|
(attachments) => this.formatBinaryFiles(attachments)
|
|
);
|
|
|
|
results.push(...formattedParts);
|
|
|
|
for (const uploadError of uploadErrors) {
|
|
// OpenAI formatBinaryFiles returns array with role wrapper, so add as separate message
|
|
results.push({
|
|
role: "user",
|
|
content: Exception.messageFrom(uploadError)
|
|
});
|
|
}
|
|
continue;
|
|
}
|
|
|
|
// Case 3: Function call response
|
|
if (content.functionResponse) {
|
|
const parsedContent = parseFunctionResponse(content.functionResponse);
|
|
|
|
if (parsedContent) {
|
|
// Check if response has required id field (for OpenAI Responses API)
|
|
if (parsedContent.id && parsedContent.id.trim() !== "") {
|
|
results.push({
|
|
type: "function_call_output",
|
|
call_id: parsedContent.id,
|
|
output: JSON.stringify(parsedContent.functionResponse.response)
|
|
});
|
|
} else {
|
|
// No id (from Gemini or legacy) - convert to text message
|
|
const legacyText = this.convertFunctionResponseToText(parsedContent);
|
|
results.push({
|
|
role: content.role as "user" | "assistant",
|
|
content: legacyText
|
|
});
|
|
}
|
|
} else {
|
|
// Fall back to regular user message if parsing fails
|
|
results.push({
|
|
role: content.role as "user" | "assistant",
|
|
content: content.functionResponse
|
|
});
|
|
}
|
|
continue;
|
|
}
|
|
|
|
// Case 4: Regular text message (user or assistant)
|
|
if (contentToExtract.trim() !== "") {
|
|
results.push({
|
|
role: content.role as "user" | "assistant",
|
|
content: contentToExtract
|
|
});
|
|
}
|
|
}
|
|
|
|
return results;
|
|
}
|
|
|
|
protected mapFunctionDefinitions(aiFunctionDefinitions: IAIFunctionDefinition[]): OpenAIFunctionTool[] {
|
|
return aiFunctionDefinitions.map((functionDefinition) => ({
|
|
type: "function",
|
|
name: functionDefinition.name,
|
|
description: functionDefinition.description,
|
|
parameters: functionDefinition.parameters
|
|
}));
|
|
}
|
|
|
|
public formatBinaryFiles(attachments: Attachment[]): string {
|
|
const contentBlocks: unknown[] = [];
|
|
|
|
for (const attachment of attachments) {
|
|
const fileID = attachment.getFileID(this.provider);
|
|
if (!fileID) {
|
|
continue; // Skip - upload failed, error message added in extractContents()
|
|
}
|
|
|
|
const mimeType = toMimeType(attachment.mimeType);
|
|
|
|
let isPlainText = false;
|
|
|
|
//This content can be sent up with the 'MimeType.TEXT_PLAIN' mime type
|
|
if (MimeTypeToFileTypes[mimeType].some(fileType => isTextFile(fileType))) {
|
|
isPlainText = true;
|
|
}
|
|
|
|
if (!isPlainText && !this.isSupportedMimeType(mimeType)) {
|
|
contentBlocks.push({ type: "input_text", text: `Unsupported mime type '${mimeType}': ${attachment.fileName}` });
|
|
continue;
|
|
}
|
|
|
|
contentBlocks.push(
|
|
{ type: "input_text", text: `Binary data for ${attachment.fileName} follows in next message` },
|
|
{
|
|
type: isPlainText || mimeType === MimeType.APPLICATION_PDF ? "input_file" : "input_image",
|
|
file_id: fileID
|
|
}
|
|
);
|
|
}
|
|
|
|
return JSON.stringify([{
|
|
role: "user",
|
|
content: contentBlocks
|
|
}]);
|
|
}
|
|
|
|
private extractRetryDelay(error: ApiError): number | undefined {
|
|
if (error.info.type !== ApiErrorType.RATE_LIMIT || !error.info.responseHeaders) {
|
|
return undefined;
|
|
}
|
|
|
|
const headers = error.info.responseHeaders;
|
|
|
|
// 1. Prefer standard Retry-After header (seconds or HTTP-date)
|
|
const retryAfter = headers.get('retry-after');
|
|
if (retryAfter) {
|
|
const seconds = Number(retryAfter);
|
|
if (!Number.isNaN(seconds)) {
|
|
return Math.max(0, seconds);
|
|
}
|
|
}
|
|
|
|
// 2. Fallback to provider-specific headers (e.g., OpenAI)
|
|
const resetHeader =
|
|
headers.get('x-ratelimit-reset-requests') ??
|
|
headers.get('x-ratelimit-reset-tokens');
|
|
|
|
if (resetHeader) {
|
|
return this.parseDurationToSeconds(resetHeader);
|
|
}
|
|
|
|
return undefined;
|
|
}
|
|
|
|
/**
|
|
* Parses duration strings (e.g., "15s", "600ms", "2m", "1h") into seconds.
|
|
* Returns undefined if parsing fails.
|
|
*/
|
|
private parseDurationToSeconds(value: string): number | undefined {
|
|
const trimmed = value.trim();
|
|
const numericValue = parseFloat(trimmed);
|
|
|
|
if (Number.isNaN(numericValue)) {
|
|
return undefined;
|
|
}
|
|
|
|
// Parse based on suffix
|
|
if (trimmed.endsWith('ms')) {
|
|
return Math.max(0, Math.ceil(numericValue / 1000));
|
|
}
|
|
if (trimmed.endsWith('s')) {
|
|
return Math.max(0, numericValue);
|
|
}
|
|
if (trimmed.endsWith('m')) {
|
|
return Math.max(0, numericValue * 60);
|
|
}
|
|
if (trimmed.endsWith('h')) {
|
|
return Math.max(0, numericValue * 3600);
|
|
}
|
|
|
|
// Fallback: treat as raw seconds
|
|
return Math.max(0, numericValue);
|
|
}
|
|
|
|
private isSupportedMimeType(mimeType: MimeType): boolean {
|
|
return this.SUPPORTED_MIMETYPES.includes(mimeType);
|
|
}
|
|
} |