andy-stack_vaultkeeper-ai/Services/StreamingService.ts
Andrew Beal 0fb17e7b3a feat: add planning model selection and rate limit countdown UI
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.
2026-01-05 21:49:51 +00:00

204 lines
No EOL
6.2 KiB
TypeScript

import type { AIFunctionCall } from "AIClasses/AIFunctionCall";
import { Event } from "Enums/Event";
import { Exception } from "Helpers/Exception";
import { ApiError, ApiErrorType } from "Types/ApiError";
import { AbortService } from "./AbortService";
import { Resolve } from "./DependencyService";
import { EventService } from "./EventService";
import { Services } from "./Services";
import { sleep } from "Helpers/Helpers";
export interface IStreamChunk {
content: string;
isComplete: boolean;
error?: string;
errorType?: ApiErrorType;
functionCall?: AIFunctionCall;
shouldContinue?: boolean;
}
export class StreamingService {
private static readonly MAX_RETRIES = 3;
private static readonly RETRY_DELAYS = [1000, 2000, 4000]; // ms
private readonly abortService: AbortService;
private readonly eventService: EventService;
public constructor() {
this.abortService = Resolve<AbortService>(Services.AbortService);
this.eventService = Resolve<EventService>(Services.EventService);
}
public async* streamRequest(url: string, requestBody: unknown, parseStreamChunk: (chunk: string) => IStreamChunk,
additionalHeaders?: Record<string, string>, extractRetryDelay?: (error: ApiError) => number | undefined): AsyncGenerator<IStreamChunk, void, unknown> {
let lastError: Error | null = null;
for (let attempt = 0; attempt <= StreamingService.MAX_RETRIES; attempt++) {
try {
const response = await this.makeRequest(url, requestBody, additionalHeaders);
const reader = response.body?.getReader();
if (!reader) {
Exception.throw("Response body is not readable");
}
const streamCompleted = yield* this.processStream(reader, parseStreamChunk);
if (!streamCompleted) {
yield { content: "", isComplete: true };
}
return;
} catch (error) {
lastError = Exception.new(error);
if (AbortService.isAbortError(error)) {
throw error;
}
if (!this.shouldRetry(error, attempt)) {
Exception.log(error);
yield this.createErrorChunk(Exception.new(error));
return;
}
const delayMs = this.calculateRetryDelay(error, attempt, extractRetryDelay);
Exception.warn(`Rate limit exceeded, waiting for ${delayMs}ms...`);
await sleep(delayMs);
}
}
if (lastError) {
Exception.log(lastError);
yield this.createErrorChunk(lastError);
}
}
private async makeRequest(url: string, requestBody: unknown,
additionalHeaders?: Record<string, string>): Promise<Response> {
try {
const response = await fetch(url, {
method: "POST",
headers: {
"Content-Type": "application/json",
...additionalHeaders,
},
body: JSON.stringify(requestBody),
signal: this.abortService.signal(),
});
if (!response.ok) {
const responseBody = await response.text();
throw ApiError.fromResponse(response.status, response.statusText, responseBody, response.headers);
}
return response;
} catch (error) {
if (ApiError.isApiError(error) || AbortService.isAbortError(error)) {
throw error;
}
throw ApiError.fromNetworkError(Exception.new(error));
}
}
private async* processStream(reader: ReadableStreamDefaultReader<Uint8Array>,
parseStreamChunk: (chunk: string) => IStreamChunk): AsyncGenerator<IStreamChunk, boolean, unknown> {
let buffer = "";
let lastChunkWasComplete = false;
const decoder = new TextDecoder();
while (true) {
if (this.abortService.signal().aborted) {
this.abortService.throw();
}
const { done, value } = await reader.read();
buffer += decoder.decode(value, { stream: true });
const lines = buffer.split("\n");
buffer = lines.pop() || ""; // Keep potentially incomplete line in buffer
for (const line of lines) {
if (line.trim().startsWith("data:")) {
const jsonStr = line.trim().substring(5);
try {
const chunk = parseStreamChunk(jsonStr);
lastChunkWasComplete = chunk.isComplete;
yield chunk;
} catch (error) {
if (AbortService.isAbortError(error) || ApiError.isApiError(error)) {
throw error;
}
Exception.log(error);
yield {
content: "",
isComplete: true,
error: Exception.messageFrom(error),
errorType: ApiErrorType.UNKNOWN
};
}
}
}
if (done) {
break;
}
}
return lastChunkWasComplete;
}
private createErrorChunk(error: Error | ApiError): IStreamChunk {
if (error instanceof ApiError) {
return {
content: "",
isComplete: true,
error: error.info.userMessage,
errorType: error.info.type
};
}
return {
content: "",
isComplete: true,
error: Exception.messageFrom(error),
errorType: ApiErrorType.UNKNOWN
};
}
private shouldRetry(error: unknown, attempt: number): boolean {
if (AbortService.isAbortError(error)) {
return false; // Don't retry abort errors
}
if (error instanceof ApiError && !error.info.isRetryable) {
return false; // Don't retry non-retryable errors
}
return attempt < StreamingService.MAX_RETRIES;
}
private calculateRetryDelay(error: unknown, attempt: number,
extractRetryDelay?: (error: ApiError) => number | undefined
): number {
// Only use provider-specific delay for 429 rate limits
if (error instanceof ApiError && error.info.type === ApiErrorType.RATE_LIMIT && extractRetryDelay) {
const providerDelaySeconds = extractRetryDelay(error);
if (providerDelaySeconds !== undefined) {
const delayMs = providerDelaySeconds * 1000;
this.eventService.trigger(Event.RateLimitCountdown, delayMs);
return delayMs;
}
}
// Fall back to exponential backoff
return StreamingService.RETRY_DELAYS[attempt];
}
}