Spaces:
Paused
Paused
File size: 18,876 Bytes
35743bd | 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 | // @ts-nocheck
/**
* Token Usage Tracking - Extract, normalize, estimate and log token usage
*/
import { appendRequestLog } from "@/lib/usageDb";
import {
getLoggedInputTokens,
getLoggedOutputTokens,
getPromptCacheCreationTokens,
getPromptCacheReadTokens,
} from "@/lib/usage/tokenAccounting";
import { FORMATS } from "../translator/formats.ts";
// ANSI color codes
export const COLORS = {
reset: "\x1b[0m",
red: "\x1b[31m",
green: "\x1b[32m",
yellow: "\x1b[33m",
blue: "\x1b[34m",
cyan: "\x1b[36m",
};
/**
* Safety buffer added to reported token usage to prevent clients from hitting
* context window limits. Accounts for overhead from system prompts,
* tool definitions, and format translation that may not be reflected in raw usage.
*
* Configurable via:
* - Settings API / Dashboard: `usageTokenBuffer` (persisted in DB)
* - Environment variable: `USAGE_TOKEN_BUFFER`
* - Defaults to 2000 if neither is set
*
* Set to 0 to disable the buffer entirely (raw provider token counts).
*/
const DEFAULT_BUFFER_TOKENS = 2000;
let _cachedBuffer: number | null = null;
let _cacheTimestamp = 0;
const CACHE_TTL_MS = 30_000; // Re-read from DB/env every 30s
function getBufferTokens(): number {
const now = Date.now();
const isExpired = _cachedBuffer !== null && now - _cacheTimestamp >= CACHE_TTL_MS;
if (_cachedBuffer !== null && !isExpired) {
return _cachedBuffer;
}
// Priority: env var > cached DB value > default
const envVal = process.env.USAGE_TOKEN_BUFFER;
if (envVal !== undefined) {
const parsed = parseInt(envVal, 10);
if (!isNaN(parsed) && parsed >= 0) {
_cachedBuffer = parsed;
_cacheTimestamp = now;
return parsed;
}
}
// Return cached value or default; kick off async DB read to update cache.
// On first call (_cachedBuffer is null), use the default.
// On TTL expiry (_cachedBuffer is stale), continue returning the stale value
// while refreshing asynchronously — prevents blocking the hot path.
if (_cachedBuffer === null || isExpired) {
if (_cachedBuffer === null) {
_cachedBuffer = DEFAULT_BUFFER_TOKENS;
}
_cacheTimestamp = now;
_loadBufferFromDb();
}
return _cachedBuffer;
}
async function _loadBufferFromDb(): Promise<void> {
try {
const { getSettings } = await import("@/lib/db/settings");
const settings = await getSettings();
const val = settings.usageTokenBuffer;
if (typeof val === "number" && val >= 0) {
_cachedBuffer = val;
_cacheTimestamp = Date.now();
}
} catch {
// DB not ready yet or settings unavailable — keep current value
}
}
/** Force-refresh the buffer from settings (e.g. after a settings update). */
export function invalidateBufferTokensCache(): void {
_cachedBuffer = null;
_cacheTimestamp = 0;
}
// Get HH:MM:SS timestamp
function getTimeString() {
return new Date().toLocaleTimeString("en-US", {
hour12: false,
hour: "2-digit",
minute: "2-digit",
second: "2-digit",
});
}
/**
* Add buffer tokens to usage to prevent context errors
* @param {object} usage - Usage object (supported format)
* @returns {object} Usage with buffer added
*/
export function addBufferToUsage(usage) {
if (!usage || typeof usage !== "object") return usage;
const buffer = getBufferTokens();
if (buffer === 0) return usage;
const result = { ...usage };
// Claude format
if (result.input_tokens !== undefined) {
result.input_tokens += buffer;
}
// OpenAI format
if (result.prompt_tokens !== undefined) {
result.prompt_tokens += buffer;
}
// Calculate or update total_tokens
if (result.total_tokens !== undefined) {
result.total_tokens += buffer;
} else if (result.prompt_tokens !== undefined && result.completion_tokens !== undefined) {
// Calculate total_tokens if not exists
result.total_tokens = result.prompt_tokens + result.completion_tokens;
}
return result;
}
export function filterUsageForFormat(usage, targetFormat) {
if (!usage || typeof usage !== "object") return usage;
// Cross-map between Claude-style and OpenAI-style field names before filtering.
// Some providers return input_tokens/output_tokens even when using OpenAI format.
const convertedUsage = { ...usage };
if (targetFormat === FORMATS.CLAUDE || targetFormat === FORMATS.OPENAI_RESPONSES) {
// OpenAI → Claude: prompt_tokens → input_tokens
if (convertedUsage.prompt_tokens !== undefined && convertedUsage.input_tokens === undefined) {
convertedUsage.input_tokens = convertedUsage.prompt_tokens;
}
if (
convertedUsage.completion_tokens !== undefined &&
convertedUsage.output_tokens === undefined
) {
convertedUsage.output_tokens = convertedUsage.completion_tokens;
}
} else {
// Claude → OpenAI: input_tokens → prompt_tokens
if (convertedUsage.input_tokens !== undefined && convertedUsage.prompt_tokens === undefined) {
convertedUsage.prompt_tokens = convertedUsage.input_tokens;
}
if (
convertedUsage.output_tokens !== undefined &&
convertedUsage.completion_tokens === undefined
) {
convertedUsage.completion_tokens = convertedUsage.output_tokens;
}
// Ensure total_tokens is set
if (
convertedUsage.total_tokens === undefined &&
convertedUsage.prompt_tokens !== undefined &&
convertedUsage.completion_tokens !== undefined
) {
convertedUsage.total_tokens = convertedUsage.prompt_tokens + convertedUsage.completion_tokens;
}
}
// Helper to pick only defined fields from usage
const pickFields = (fields) => {
const filtered = {};
for (const field of fields) {
if (convertedUsage[field] !== undefined) {
filtered[field] = convertedUsage[field];
}
}
return filtered;
};
// Define allowed fields for each format
const formatFields = {
[FORMATS.CLAUDE]: [
"input_tokens",
"output_tokens",
"cache_read_input_tokens",
"cache_creation_input_tokens",
"estimated",
],
[FORMATS.GEMINI]: [
"promptTokenCount",
"candidatesTokenCount",
"totalTokenCount",
"cachedContentTokenCount",
"thoughtsTokenCount",
"estimated",
],
[FORMATS.OPENAI_RESPONSES]: [
"input_tokens",
"output_tokens",
"input_tokens_details",
"output_tokens_details",
"estimated",
],
// OpenAI format (default for OPENAI, CODEX, KIRO, etc.)
default: [
"prompt_tokens",
"completion_tokens",
"total_tokens",
"cached_tokens",
"reasoning_tokens",
"prompt_tokens_details",
"completion_tokens_details",
"estimated",
],
};
// Get fields for target format
let fields = formatFields[targetFormat];
// Use same fields for similar formats
if (targetFormat === FORMATS.GEMINI_CLI || targetFormat === FORMATS.ANTIGRAVITY) {
fields = formatFields[FORMATS.GEMINI];
} else if (targetFormat === FORMATS.OPENAI_RESPONSE) {
fields = formatFields[FORMATS.OPENAI_RESPONSES];
} else if (!fields) {
fields = formatFields.default;
}
return pickFields(fields);
}
/**
* Normalize usage object - ensure all values are valid numbers
*/
export function normalizeUsage(usage) {
if (!usage || typeof usage !== "object" || Array.isArray(usage)) return null;
const normalized = {};
const assignNumber = (key, value) => {
if (value === undefined || value === null) return;
const numeric = Number(value);
if (Number.isFinite(numeric)) normalized[key] = numeric;
};
assignNumber("prompt_tokens", usage?.prompt_tokens);
assignNumber("completion_tokens", usage?.completion_tokens);
assignNumber("total_tokens", usage?.total_tokens);
assignNumber("cache_read_input_tokens", usage?.cache_read_input_tokens);
assignNumber("cache_creation_input_tokens", usage?.cache_creation_input_tokens);
assignNumber("cached_tokens", usage?.cached_tokens);
assignNumber("reasoning_tokens", usage?.reasoning_tokens);
if (Object.keys(normalized).length === 0) return null;
return normalized;
}
/**
* Check if usage has valid token data
* Valid = has at least one token field with value > 0
* Invalid = empty object {}, null, undefined, no token fields, or all zeros
*/
export function hasValidUsage(usage) {
if (!usage || typeof usage !== "object") return false;
// Check for known token fields with value > 0
const tokenFields = [
"prompt_tokens",
"completion_tokens",
"total_tokens", // OpenAI
"input_tokens",
"output_tokens", // Claude
"promptTokenCount",
"candidatesTokenCount", // Gemini
];
for (const field of tokenFields) {
if (typeof usage[field] === "number" && usage[field] > 0) {
return true;
}
}
return false;
}
/**
* Extract usage from supported formats (Claude, OpenAI, Gemini, Responses API)
*/
export function extractUsage(chunk) {
if (!chunk || typeof chunk !== "object") return null;
// Claude/Antigravity streaming: message_start event carries INPUT tokens
// FIX #74: This event was not handled — input_tokens were being dropped
// Structure: { type: "message_start", message: { usage: { input_tokens: N, output_tokens: 0 } } }
//
// Note: Claude's input_tokens is only the non-cached portion.
// Sum cache tokens into prompt_tokens for a correct total (consistent with
// extractUsageFromResponse in usageExtractor.ts for non-streaming).
if (chunk.type === "message_start" && chunk.message?.usage) {
const u = chunk.message.usage;
const inputTokens = u.input_tokens || u.prompt_tokens || 0;
const cacheRead = u.cache_read_input_tokens || 0;
const cacheCreation = u.cache_creation_input_tokens || 0;
if (inputTokens > 0 || cacheRead > 0 || cacheCreation > 0) {
return normalizeUsage({
prompt_tokens: inputTokens + cacheRead + cacheCreation,
completion_tokens: u.output_tokens || u.completion_tokens || 0,
cache_read_input_tokens: u.cache_read_input_tokens,
cache_creation_input_tokens: u.cache_creation_input_tokens,
});
}
}
// Claude format (message_delta event) — typically carries OUTPUT tokens
if (chunk.type === "message_delta" && chunk.usage && typeof chunk.usage === "object") {
const deltaInput = chunk.usage.input_tokens || 0;
const deltaCacheRead = chunk.usage.cache_read_input_tokens || 0;
const deltaCacheCreation = chunk.usage.cache_creation_input_tokens || 0;
return normalizeUsage({
prompt_tokens: deltaInput + deltaCacheRead + deltaCacheCreation,
completion_tokens: chunk.usage.output_tokens || 0,
cache_read_input_tokens: chunk.usage.cache_read_input_tokens,
cache_creation_input_tokens: chunk.usage.cache_creation_input_tokens,
});
}
// OpenAI Responses API format (response.completed or response.done)
if (
(chunk.type === "response.completed" || chunk.type === "response.done") &&
chunk.response?.usage &&
typeof chunk.response.usage === "object"
) {
const usage = chunk.response.usage;
return normalizeUsage({
prompt_tokens: usage.input_tokens || usage.prompt_tokens || 0,
completion_tokens: usage.output_tokens || usage.completion_tokens || 0,
cached_tokens:
usage.input_tokens_details?.cached_tokens ??
usage.prompt_tokens_details?.cached_tokens ??
usage.cache_read_input_tokens,
cache_creation_input_tokens: usage.cache_creation_input_tokens,
reasoning_tokens:
usage.output_tokens_details?.reasoning_tokens ??
usage.completion_tokens_details?.reasoning_tokens ??
usage.reasoning_tokens,
});
}
// OpenAI format
if (
chunk.usage &&
typeof chunk.usage === "object" &&
(chunk.usage.prompt_tokens !== undefined || chunk.usage.input_tokens !== undefined)
) {
return normalizeUsage({
prompt_tokens: chunk.usage.prompt_tokens ?? chunk.usage.input_tokens ?? 0,
completion_tokens: chunk.usage.completion_tokens ?? chunk.usage.output_tokens ?? 0,
cached_tokens:
chunk.usage.prompt_tokens_details?.cached_tokens ??
chunk.usage.input_tokens_details?.cached_tokens,
reasoning_tokens:
chunk.usage.completion_tokens_details?.reasoning_tokens ??
chunk.usage.output_tokens_details?.reasoning_tokens,
});
}
// Gemini format (Antigravity)
if (chunk.usageMetadata && typeof chunk.usageMetadata === "object") {
return normalizeUsage({
prompt_tokens: chunk.usageMetadata?.promptTokenCount || 0,
completion_tokens: chunk.usageMetadata?.candidatesTokenCount || 0,
total_tokens: chunk.usageMetadata?.totalTokenCount,
cached_tokens: chunk.usageMetadata?.cachedContentTokenCount,
reasoning_tokens: chunk.usageMetadata?.thoughtsTokenCount,
});
}
return null;
}
// Heuristic token estimation constants
const CHARS_PER_TOKEN_SCHEMA = 6; // ~6 chars/token for JSON schemas (more verbose per token)
/**
* Improved token estimation heuristic (no dependency).
* Splits text on common token boundaries (whitespace, punctuation, camelCase)
* and applies a sub-word correction factor. Better accuracy for:
* - English text (~4 chars/token)
* - CJK text (~1 char/token for ideographs)
* - Code (~3.5 chars/token, more punctuation-heavy)
*
* @param {string} text - Text to estimate tokens for
* @returns {number} Estimated token count
*/
function estimateTokenCount(text) {
if (!text || typeof text !== "string") return 0;
// Count CJK ideographs separately — each is roughly 1 token
const cjkMatches = text.match(/[\u3000-\u9fff\uf900-\ufaff\u{20000}-\u{2fa1f}]/gu);
const cjkCount = cjkMatches ? cjkMatches.length : 0;
// Remove CJK chars for the remaining estimation
const nonCJK = text.replace(/[\u3000-\u9fff\uf900-\ufaff]/g, " ");
// Split on token boundaries: whitespace, punctuation, camelCase transitions
const tokens = nonCJK
.split(/(\s+|[^\w\s]|(?<=[a-z])(?=[A-Z]))/)
.filter((t) => t && t.trim().length > 0);
// Apply sub-word correction: BPE tokenizers often split long words
// into sub-word pieces, so raw token count underestimates slightly
const estimatedNonCJK = Math.ceil(tokens.length * 1.3);
return cjkCount + estimatedNonCJK;
}
/**
* Estimate input tokens from request body.
* Separates tool definitions (JSON schemas) from message content
* for more accurate estimation since JSON schemas are more verbose but
* compress into fewer tokens than plain text.
*/
export function estimateInputTokens(body) {
if (!body || typeof body !== "object") return 0;
try {
let toolTokens = 0;
let messageTokens = 0;
// Separate tool definitions from the rest of the body
if (body.tools && Array.isArray(body.tools)) {
const toolStr = JSON.stringify(body.tools);
toolTokens = Math.ceil(toolStr.length / CHARS_PER_TOKEN_SCHEMA);
// Estimate messages without tools
const { tools, ...bodyWithoutTools } = body;
messageTokens = estimateTokenCount(JSON.stringify(bodyWithoutTools));
} else {
messageTokens = estimateTokenCount(JSON.stringify(body));
}
return messageTokens + toolTokens;
} catch (err) {
// Fallback if stringify fails
return 0;
}
}
/**
* Estimate output tokens from content length.
* Uses improved heuristic when possible, falls back to length-based estimation.
*/
export function estimateOutputTokens(contentLength) {
if (!contentLength || contentLength <= 0) return 0;
// When we only have a character count, use 4 chars/token with sub-word correction
return Math.max(1, Math.ceil(contentLength / 3.5));
}
/**
* Format usage object based on target format
* @param {number} inputTokens - Input/prompt tokens
* @param {number} outputTokens - Output/completion tokens
* @param {string} targetFormat - Target format from FORMATS
*/
export function formatUsage(inputTokens, outputTokens, targetFormat) {
// Claude format uses input_tokens/output_tokens
if (targetFormat === FORMATS.CLAUDE) {
return addBufferToUsage({
input_tokens: inputTokens,
output_tokens: outputTokens,
estimated: true,
});
}
// Default: OpenAI format (works for openai, gemini, responses, etc.)
return addBufferToUsage({
prompt_tokens: inputTokens,
completion_tokens: outputTokens,
total_tokens: inputTokens + outputTokens,
estimated: true,
});
}
/**
* Estimate full usage when provider doesn't return it
* @param {object} body - Request body for input token estimation
* @param {number} contentLength - Content length for output token estimation
* @param {string} targetFormat - Target format from FORMATS constant
*/
export function estimateUsage(body, contentLength, targetFormat = FORMATS.OPENAI) {
return formatUsage(estimateInputTokens(body), estimateOutputTokens(contentLength), targetFormat);
}
/**
* Log usage with cache info (green color)
*/
export function logUsage(
provider,
usage,
model: string | null = null,
connectionId: string | null = null,
apiKeyInfo = null
) {
if (!usage || typeof usage !== "object") return;
const p = provider?.toUpperCase() || "UNKNOWN";
// Support both formats:
// - OpenAI: prompt_tokens, completion_tokens
// - Claude: input_tokens, output_tokens
const inTokens = getLoggedInputTokens(usage);
const outTokens = getLoggedOutputTokens(usage);
void apiKeyInfo;
const normalizedConnectionId = typeof connectionId === "string" ? connectionId : undefined;
const accountPrefix = normalizedConnectionId
? normalizedConnectionId.slice(0, 8) + "..."
: "unknown";
let msg = `[${getTimeString()}] 📊 ${COLORS.green}[USAGE] ${p} | in=${inTokens} | out=${outTokens} | account=${accountPrefix}${COLORS.reset}`;
// Add estimated flag if present
if (usage.estimated) {
msg += ` ${COLORS.yellow}(estimated)${COLORS.reset}`;
}
// Add cache info if present (unified from different formats)
const cacheRead = getPromptCacheReadTokens(usage);
if (cacheRead) msg += ` | cache_read=${cacheRead}`;
const cacheCreation = getPromptCacheCreationTokens(usage);
if (cacheCreation) msg += ` | cache_create=${cacheCreation}`;
const reasoning = usage.reasoning_tokens;
if (reasoning) msg += ` | reasoning=${reasoning}`;
console.log(msg);
// Streaming requests persist usage once in chatCore's completion callback.
// Keep this helper side-effect free apart from console visibility.
const tokens = {
input: inTokens,
output: outTokens,
cacheRead: cacheRead || 0,
cacheCreation: cacheCreation || 0,
reasoning: reasoning || 0,
};
appendRequestLog({
model: typeof model === "string" ? model : undefined,
provider: typeof provider === "string" ? provider : undefined,
connectionId: normalizedConnectionId,
tokens,
status: "200 OK",
}).catch(() => {});
}
|