Press n or j to go to the next uncovered block, b, p or k for the previous block.
| 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 | 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x | import {
getStreamModeSplitPattern,
type SplitPatternType,
} from '@api/utils/object';
import { unwrapJsonResponseContent } from '@api/utils/structured-output';
import { error, warn } from '@shared/console-logging';
import type {
JSONToStreamGeneratorTransformFunction,
ResponseChunkStreamTransformFunction,
ResponseTransformFunction,
} from '@shared/types/ai-providers/config';
import type { SuperAgentsRequestData } from '@shared/types/api/request/body';
import type { SuperAgentsResponseBody } from '@shared/types/api/response/body';
import type { ChatCompletionResponseBody } from '@shared/types/api/routes/chat-completions-api';
import type { CompletionResponseBody } from '@shared/types/api/routes/completions-api';
import {
AIProvider,
ContentTypeName,
PRECONDITION_CHECK_FAILED_STATUS_CODE,
REQUEST_TIMEOUT_STATUS_CODE,
} from '@shared/types/constants';
// Stream processing constants
/**
* Delay after first chunk to allow client to establish connection and process headers
* This helps prevent race conditions where the client hasn't fully set up stream handling
*/
const FIRST_CHUNK_DELAY_MS = 25;
/**
* Small delay between chunks for Azure OpenAI to prevent rate limiting
* Azure OpenAI requires throttling between chunks to avoid connection issues
*/
const AZURE_CHUNK_DELAY_MS = 1;
/**
* Maximum size for accumulated stream chunks in bytes (10 MB)
* Protects against memory exhaustion from extremely long streams
*/
const MAX_ACCUMULATED_CHUNKS_SIZE = 10 * 1024 * 1024;
// Helper function to clean response headers by removing compression-related headers
function cleanResponseHeaders(
originalHeaders: Headers,
): Record<string, string> {
const cleanedHeaders: Record<string, string> = {};
const headersToExclude = new Set([
'content-encoding',
'content-length', // Will be wrong after decompression
'transfer-encoding',
'vary', // Often related to compression negotiation
]);
for (const [key, value] of originalHeaders.entries()) {
if (!headersToExclude.has(key.toLowerCase())) {
cleanedHeaders[key] = value;
}
}
return cleanedHeaders;
}
function readUInt32BE(buffer: Uint8Array, offset: number): number {
return (
((buffer[offset] << 24) |
(buffer[offset + 1] << 16) |
(buffer[offset + 2] << 8) |
buffer[offset + 3]) >>>
0
); // Ensure the result is an unsigned integer
}
function getPayloadFromAWSChunk(chunk: Uint8Array): string {
const decoder = new TextDecoder();
const chunkLength = readUInt32BE(chunk, 0);
const headersLength = readUInt32BE(chunk, 4);
// prelude 8 + Prelude crc 4 = 12
const headersEnd = 12 + headersLength;
const payloadLength = chunkLength - headersEnd - 4; // Subtracting 4 for the message crc
const payload = chunk.slice(headersEnd, headersEnd + payloadLength);
const decodedJson = JSON.parse(decoder.decode(payload));
return decodedJson.bytes
? Buffer.from(decodedJson.bytes, 'base64').toString()
: JSON.stringify(decodedJson);
}
function concatenateUint8Arrays(a: Uint8Array, b: Uint8Array): Uint8Array {
const result = new Uint8Array(a.length + b.length);
result.set(a, 0); // Copy contents of array 'a' into 'result' starting at index 0
result.set(b, a.length); // Copy contents of array 'b' into 'result' starting at index 'a.length'
return result;
}
export async function* readAWSStream(
reader: ReadableStreamDefaultReader,
transformFunction: ResponseChunkStreamTransformFunction | undefined,
fallbackChunkId: string,
strictOpenAiCompliance: boolean,
saRequestData: SuperAgentsRequestData,
onFirstChunk?: () => void,
): AsyncGenerator<string | Uint8Array, void, unknown> {
let buffer = new Uint8Array() as Uint8Array<ArrayBufferLike>;
let expectedLength = 0;
const streamState = {};
let isFirstChunk = true;
while (true) {
const { done, value } = await reader.read();
if (done) {
if (buffer.length) {
expectedLength = readUInt32BE(buffer, 0);
while (buffer.length >= expectedLength && buffer.length !== 0) {
const data = buffer.subarray(0, expectedLength);
buffer = buffer.subarray(expectedLength);
expectedLength = readUInt32BE(buffer, 0);
const payload = getPayloadFromAWSChunk(data);
if (transformFunction) {
const transformedChunk = transformFunction(
payload,
fallbackChunkId,
streamState,
strictOpenAiCompliance,
saRequestData,
);
if (Array.isArray(transformedChunk)) {
for (const item of transformedChunk) {
yield item;
}
} else {
yield transformedChunk;
}
} else {
yield data;
}
}
}
break;
}
if (expectedLength === 0) {
expectedLength = readUInt32BE(value, 0);
}
buffer = concatenateUint8Arrays(buffer, value);
while (buffer.length >= expectedLength && buffer.length !== 0) {
const data = buffer.subarray(0, expectedLength);
buffer = buffer.subarray(expectedLength);
expectedLength = readUInt32BE(buffer, 0);
const payload = getPayloadFromAWSChunk(data);
if (isFirstChunk && onFirstChunk) {
onFirstChunk();
isFirstChunk = false;
}
if (transformFunction) {
const transformedChunk = transformFunction(
payload,
fallbackChunkId,
streamState,
strictOpenAiCompliance,
saRequestData,
);
if (Array.isArray(transformedChunk)) {
for (const item of transformedChunk) {
yield item;
}
} else {
yield transformedChunk;
}
} else {
yield data;
}
}
}
}
export async function* readStream(
reader: ReadableStreamDefaultReader,
splitPattern: SplitPatternType,
transformFunction: ResponseChunkStreamTransformFunction | undefined,
isSleepTimeRequired: boolean,
fallbackChunkId: string,
strictOpenAiCompliance: boolean,
saRequestData: SuperAgentsRequestData,
onFirstChunk?: () => void,
): AsyncGenerator<string | Uint8Array, void, unknown> {
let buffer = '';
const decoder = new TextDecoder();
let isFirstChunk = true;
const streamState = {};
while (true) {
const { done, value } = await reader.read();
if (done) {
if (buffer.length > 0) {
if (transformFunction) {
const transformedChunk = transformFunction(
buffer,
fallbackChunkId,
streamState,
strictOpenAiCompliance,
saRequestData,
);
if (Array.isArray(transformedChunk)) {
for (const item of transformedChunk) {
yield item;
}
} else {
yield transformedChunk;
}
} else {
yield buffer;
}
}
break;
}
buffer += decoder.decode(value, { stream: true });
// keep buffering until we have a complete chunk
while (buffer.split(splitPattern).length > 1) {
const parts = buffer.split(splitPattern);
const lastPart = parts.pop() ?? ''; // remove the last part from the array and keep it in buffer
for (const part of parts) {
// Some providers send ping event which can be ignored during parsing
if (part.length > 0) {
if (isFirstChunk) {
if (onFirstChunk) {
onFirstChunk();
}
isFirstChunk = false;
await new Promise((resolve) =>
setTimeout(resolve, FIRST_CHUNK_DELAY_MS),
);
} else if (isSleepTimeRequired) {
await new Promise((resolve) =>
setTimeout(resolve, AZURE_CHUNK_DELAY_MS),
);
}
if (transformFunction) {
const transformedChunk = transformFunction(
part,
fallbackChunkId,
streamState,
strictOpenAiCompliance,
saRequestData,
);
if (Array.isArray(transformedChunk)) {
for (const item of transformedChunk) {
yield item;
}
} else {
yield transformedChunk;
}
} else {
yield part + splitPattern;
}
}
}
buffer = lastPart; // keep the last part (after the last '\n\n') in buffer
}
}
}
export async function handleTextResponse(
aiProviderResponse: Response,
responseTransformer: ResponseTransformFunction | undefined,
saRequestData: SuperAgentsRequestData,
): Promise<Response> {
const text = await aiProviderResponse.text();
if (responseTransformer) {
const transformedText = responseTransformer(
{ 'html-message': text },
aiProviderResponse.status,
aiProviderResponse.headers,
false,
saRequestData,
);
return new Response(JSON.stringify(transformedText), {
...aiProviderResponse,
status: aiProviderResponse.status,
headers: new Headers({
...cleanResponseHeaders(aiProviderResponse.headers),
'content-type': 'application/json',
}),
});
}
return new Response(text, aiProviderResponse);
}
export async function handleNonStreamingMode(
aiProviderResponse: Response,
responseTransformer: ResponseTransformFunction | undefined,
strictOpenAiCompliance: boolean,
saRequestData: SuperAgentsRequestData,
areSyncHooksAvailable: boolean,
): Promise<{
response: Response;
saResponseBody: SuperAgentsResponseBody | null;
originalBodyJson?: Record<string, unknown> | null;
}> {
// 408 is thrown whenever a request takes more than request_timeout to respond.
// In that case, response thrown by gateway is already in OpenAI format.
// So no need to transform it again.
if (
[
REQUEST_TIMEOUT_STATUS_CODE,
PRECONDITION_CHECK_FAILED_STATUS_CODE,
].includes(aiProviderResponse.status)
) {
return {
response: aiProviderResponse,
saResponseBody: await aiProviderResponse.clone().json(),
};
}
let originalResponseBodyJson: Record<string, unknown> | null = null;
const originalResponseBodyText: string = await aiProviderResponse
.clone()
.text();
try {
originalResponseBodyJson = JSON.parse(originalResponseBodyText);
} catch {
// Maybe the response is not meant to be JSON. Do nothing.
}
let transformedBodyJson: Record<string, unknown> | Blob | null =
originalResponseBodyJson;
if (responseTransformer && originalResponseBodyJson) {
transformedBodyJson = responseTransformer(
originalResponseBodyJson,
aiProviderResponse.status,
aiProviderResponse.headers,
strictOpenAiCompliance,
saRequestData,
);
}
// A provider that only saw `response_format` as a prompt instruction may have
// answered with the JSON inside a markdown fence. Unwrap it before validation
// so the caller gets content it can parse.
if (transformedBodyJson && !(transformedBodyJson instanceof Blob)) {
transformedBodyJson = unwrapJsonResponseContent(
saRequestData,
transformedBodyJson,
);
}
// Make sure that the response body is in the expected format.
let saResponseBody: SuperAgentsResponseBody | null = null;
if (transformedBodyJson) {
const saResponseBodyParseResult =
saRequestData.responseSchema.safeParse(transformedBodyJson);
if (!saResponseBodyParseResult.success) {
throw new Error(
`Invalid response body: ${saResponseBodyParseResult.error}`,
);
}
saResponseBody = saResponseBodyParseResult.data as SuperAgentsResponseBody;
}
if (!areSyncHooksAvailable) {
return {
response: new Response(
saResponseBody
? JSON.stringify(saResponseBody)
: originalResponseBodyText,
{
...aiProviderResponse,
headers: new Headers(
cleanResponseHeaders(aiProviderResponse.headers),
),
},
),
saResponseBody, // TODO: Review if this is necessary
originalBodyJson:
transformedBodyJson instanceof Blob ? null : transformedBodyJson,
};
}
return {
response: new Response(JSON.stringify(saResponseBody), {
...aiProviderResponse,
headers: new Headers(cleanResponseHeaders(aiProviderResponse.headers)),
}),
saResponseBody,
// Send original response if transformer exists
...(responseTransformer && {
originalBodyJson:
transformedBodyJson instanceof Blob ? null : transformedBodyJson,
}),
};
}
export function handleAudioResponse(response: Response): Response {
return new Response(response.body, response);
}
export function handleOctetStreamResponse(response: Response): Response {
return new Response(response.body, response);
}
export function handleImageResponse(response: Response): Response {
return new Response(response.body, response);
}
export function handleStreamingMode(
response: Response,
provider: AIProvider,
responseTransformer: ResponseChunkStreamTransformFunction | undefined,
aiProviderRequestURL: string,
saRequestData: SuperAgentsRequestData,
strictOpenAiCompliance: boolean,
onFirstChunk?: () => void,
onStreamEnd?: (accumulatedChunks: string) => void,
): Response {
const splitPattern = getStreamModeSplitPattern(
provider,
aiProviderRequestURL,
);
// If the provider doesn't supply completion id,
// we generate a fallback id using the provider name + timestamp.
const fallbackChunkId = `${provider}-${Date.now().toString()}`;
if (!response.body) {
throw new Error('Response format is invalid. Body not found');
}
const { readable, writable } = new TransformStream();
const writer = writable.getWriter();
const reader = response.body.getReader();
const isSleepTimeRequired = provider === AIProvider.AZURE_OPENAI;
const encoder = new TextEncoder();
const decoder = new TextDecoder();
// Accumulate chunks for logging
let accumulatedChunks = '';
if (provider === AIProvider.BEDROCK) {
(async () => {
try {
for await (const chunk of readAWSStream(
reader,
responseTransformer,
fallbackChunkId,
strictOpenAiCompliance,
saRequestData,
onFirstChunk,
)) {
const encodedChunk = encoder.encode(chunk as string);
const decodedChunk = decoder.decode(encodedChunk, { stream: true });
// Check size limit before accumulating
const newSize =
new TextEncoder().encode(accumulatedChunks).length +
new TextEncoder().encode(decodedChunk).length;
if (newSize > MAX_ACCUMULATED_CHUNKS_SIZE) {
warn(
'[Stream Handler] Accumulated chunks exceed size limit, stopping accumulation',
{ currentSize: newSize, maxSize: MAX_ACCUMULATED_CHUNKS_SIZE },
);
// Continue writing to client but stop accumulating
await writer.write(encodedChunk);
continue;
}
accumulatedChunks += decodedChunk;
await writer.write(encodedChunk);
}
} catch (streamError) {
error('[Stream Handler] Error processing Bedrock stream', streamError);
throw streamError;
} finally {
try {
writer.close();
} catch (closeError) {
error(
'[Stream Handler] Error closing Bedrock stream writer',
closeError,
);
}
if (onStreamEnd) {
onStreamEnd(accumulatedChunks);
}
}
})();
} else {
(async () => {
try {
for await (const chunk of readStream(
reader,
splitPattern,
responseTransformer,
isSleepTimeRequired,
fallbackChunkId,
strictOpenAiCompliance,
saRequestData,
onFirstChunk,
)) {
const encodedChunk = encoder.encode(chunk as string);
const decodedChunk = decoder.decode(encodedChunk, { stream: true });
// Check size limit before accumulating
const newSize =
new TextEncoder().encode(accumulatedChunks).length +
new TextEncoder().encode(decodedChunk).length;
if (newSize > MAX_ACCUMULATED_CHUNKS_SIZE) {
warn(
'[Stream Handler] Accumulated chunks exceed size limit, stopping accumulation',
{ currentSize: newSize, maxSize: MAX_ACCUMULATED_CHUNKS_SIZE },
);
// Continue writing to client but stop accumulating
await writer.write(encodedChunk);
continue;
}
accumulatedChunks += decodedChunk;
await writer.write(encodedChunk);
}
} catch (streamError) {
error('[Stream Handler] Error processing stream', streamError);
throw streamError;
} finally {
try {
writer.close();
} catch (closeError) {
error('[Stream Handler] Error closing stream writer', closeError);
}
if (onStreamEnd) {
onStreamEnd(accumulatedChunks);
}
}
})();
}
// Convert GEMINI/COHERE json stream to text/event-stream for non-proxy calls
const isGoogleCohereOrBedrock = [
AIProvider.GOOGLE,
AIProvider.COHERE,
AIProvider.BEDROCK,
].includes(provider);
// const isVertexLlama =
// proxyProvider === AIProviderName.enum['vertex-ai'] &&
// responseTransformer?.name ===
// VertexLlamaChatCompleteStreamChunkTransform.name;
// const isJsonStream = isGoogleCohereOrBedrock || isVertexLlama;
const isJsonStream = isGoogleCohereOrBedrock;
if (isJsonStream && responseTransformer) {
return new Response(readable, {
...response,
headers: new Headers({
...cleanResponseHeaders(response.headers),
'content-type': 'text/event-stream',
}),
});
}
return new Response(readable, response);
}
export async function handleJSONToStreamResponse(
response: Response,
provider: AIProvider,
responseTransformerFunction: JSONToStreamGeneratorTransformFunction,
): Promise<Response> {
const { readable, writable } = new TransformStream();
const writer = writable.getWriter();
const encoder = new TextEncoder();
const responseJSON: ChatCompletionResponseBody | CompletionResponseBody =
await response.clone().json();
if (
Object.prototype.toString.call(responseTransformerFunction) ===
'[object GeneratorFunction]'
) {
const generator = responseTransformerFunction(responseJSON, provider);
(async () => {
while (true) {
const chunk = generator.next();
if (chunk.done) {
break;
}
await writer.write(encoder.encode(chunk.value));
}
writer.close();
})();
} else {
const streamChunkArray = responseTransformerFunction(
responseJSON,
provider,
);
(async () => {
for (const chunk of streamChunkArray) {
await writer.write(encoder.encode(chunk));
}
writer.close();
})();
}
return new Response(readable, {
headers: new Headers({
...cleanResponseHeaders(response.headers),
'content-type': ContentTypeName.EVENT_STREAM,
}),
status: response.status,
statusText: response.statusText,
});
}
|