tokendiet 1.4.0 → 1.6.0

package/dist/cli.js

@@ -26,6 +26,7 @@ program
     .option('--cloud-url <url>', 'Cloud API URL', 'https://tokendiet-production.up.railway.app')
     .option('--port <port>', 'Port to listen on (default: 3100)', '3100')
     .option('--verbose', 'Show full NestJS logs')
+    .option('--dry-run', 'Disable all stripping (baseline measurement)')
     .action(async (opts) => {
     let orgToken = await resolveToken(opts.orgToken);
     if (!orgToken) {
@@ -37,6 +38,7 @@ program
         cloudUrl: opts.cloudUrl,
         port: parseInt(opts.port, 10),
         verbose: opts.verbose ?? false,
+        dryRun: opts.dryRun ?? false,
     });
     while (result.unauthorized) {
         (0, utils_1.clearSavedToken)();
@@ -51,6 +53,7 @@ program
             cloudUrl: opts.cloudUrl,
             port: parseInt(opts.port, 10),
             verbose: opts.verbose ?? false,
+            dryRun: opts.dryRun ?? false,
         });
     }
     if (result.success) {

package/dist/commands/start.d.ts

@@ -3,6 +3,7 @@ export interface StartOptions {
     cloudUrl: string;
     port: number;
     verbose?: boolean;
+    dryRun?: boolean;
 }
 export interface StartResult {
     success: boolean;

package/dist/commands/start.js

@@ -125,6 +125,8 @@ async function startGateway(options) {
     };
     if (options.verbose)
         env.TOKENDIET_VERBOSE = '1';
+    if (options.dryRun)
+        env.TOKENDIET_DRY_RUN = '1';
     if (gatewayRunId)
         env.TOKENDIET_GATEWAY_RUN_ID = gatewayRunId;
     const claudeSettings = (0, utils_1.readJsonFile)(utils_1.claudeSettingsPath);

package/dist/main.js

@@ -10,7 +10,7 @@ async function bootstrap() {
     const port = process.env.PORT ?? 3100;
     await app.listen(port);
     if (!process.env.TOKENDIET_VERBOSE) {
-        (0, cli_output_1.printBanner)(Number(port));
+        (0, cli_output_1.printBanner)(Number(port), process.env.TOKENDIET_DRY_RUN === '1');
     }
 }
 bootstrap();

package/dist/proxy/analytics.service.d.ts

@@ -25,6 +25,7 @@ export declare class AnalyticsService {
     private readonly gatewayRunId;
     private readonly upstreamUrl;
     private readonly analyticsPath;
+    private readonly isDryRun;
     private sessionTotals;
     constructor(config: ConfigService, cloudApiService: CloudApiService);
     getSessionTotals(): SessionTotals;

package/dist/proxy/analytics.service.js

@@ -71,6 +71,7 @@ let AnalyticsService = AnalyticsService_1 = class AnalyticsService {
     gatewayRunId;
     upstreamUrl;
     analyticsPath;
+    isDryRun;
     sessionTotals = { tokensSaved: 0, dollarSavings: 0, requestCount: 0 };
     constructor(config, cloudApiService) {
         this.config = config;
@@ -78,6 +79,7 @@ let AnalyticsService = AnalyticsService_1 = class AnalyticsService {
         this.sessionId = this.config.get('TOKENDIET_SESSION_ID') ?? null;
         this.gatewayRunId = this.config.get('TOKENDIET_GATEWAY_RUN_ID') ?? null;
         this.upstreamUrl = this.config.get('PROXY_TARGET_URL') ?? DEFAULT_UPSTREAM;
+        this.isDryRun = this.config.get('TOKENDIET_DRY_RUN') === '1';
         this.analyticsPath = ANALYTICS_PATH;
     }
     getSessionTotals() {
@@ -99,6 +101,7 @@ let AnalyticsService = AnalyticsService_1 = class AnalyticsService {
                 strippingApplied: this.buildStrippingEntries(preStripResult, usage),
                 originalRequestTokenCount: preStripResult.input_tokens,
                 countMethod: preStripResult.countMethod,
+                isDryRun: this.isDryRun,
                 upstreamResponseUsage: {
                     input_tokens: usage.input_tokens,
                     output_tokens: usage.output_tokens ?? 0,

package/dist/proxy/cli-output.d.ts

@@ -1,9 +1,9 @@
-export declare function printBanner(port: number): void;
+export declare function printBanner(port: number, dryRun?: boolean): void;
 interface Totals {
     tokensSaved: number;
     dollarSavings: number;
     requestCount: number;
 }
-export declare function printRequestSavings(model: string, tokensSaved: number, costSaved: number, totals: Totals): void;
+export declare function printRequestSavings(model: string, tokensSaved: number, costSaved: number, originalTokenCount: number, totals: Totals, dryRun?: boolean): void;
 export declare function _reset(): void;
 export {};

package/dist/proxy/cli-output.js

@@ -11,12 +11,14 @@ const MAX_VISIBLE = 5;
 const COL_MODEL = 30;
 const COL_TOKENS = 12;
 const COL_COST = 10;
-const TABLE_WIDTH = 2 + COL_MODEL + 2 + COL_TOKENS + 2 + COL_COST + 2;
+const COL_PCT = 6;
+const TABLE_WIDTH = 2 + COL_MODEL + 2 + COL_TOKENS + 2 + COL_COST + 2 + COL_PCT + 2;
 const recentRequests = [];
 let linesWritten = 0;
-function printBanner(port) {
+function printBanner(port, dryRun = false) {
     console.log('');
-    console.log(chalk_1.default.bold.green(' TokenDiet') + ' running on port ' + chalk_1.default.cyan(String(port)));
+    const tag = dryRun ? chalk_1.default.bold.yellow(' [DRY RUN]') : '';
+    console.log(chalk_1.default.bold.green(' TokenDiet') + tag + ' running on port ' + chalk_1.default.cyan(String(port)));
     console.log('');
 }
 function formatNumber(n) {
@@ -34,7 +36,8 @@ function formatHeader() {
     return chalk_1.default.dim('  ' +
         'Model'.padEnd(COL_MODEL) + '  ' +
         'Tokens saved'.padStart(COL_TOKENS) + '  ' +
-        'Cost saved'.padStart(COL_COST));
+        'Cost saved'.padStart(COL_COST) + '  ' +
+        '%'.padStart(COL_PCT));
 }
 function formatSeparator() {
     return chalk_1.default.dim('  ' + '\u2500'.repeat(TABLE_WIDTH - 2));
@@ -43,12 +46,14 @@ function formatRequest(entry, dim) {
     const model = truncateModel(entry.model);
     const tokens = formatNumber(entry.tokensSaved);
     const cost = formatCost(entry.costSaved);
+    const pct = entry.pctSaved.toFixed(2) + '%';
     const wrap = dim ? chalk_1.default.dim : (s) => s;
     const arrow = dim ? chalk_1.default.dim('\u2193 ') : chalk_1.default.green('\u2193 ');
     return (arrow +
         wrap(model.padEnd(COL_MODEL)) + '  ' +
         (dim ? chalk_1.default.dim : chalk_1.default.yellow)(tokens.padStart(COL_TOKENS)) + '  ' +
-        (dim ? chalk_1.default.dim : chalk_1.default.green)(cost.padStart(COL_COST)));
+        (dim ? chalk_1.default.dim : chalk_1.default.green)(cost.padStart(COL_COST)) + '  ' +
+        (dim ? chalk_1.default.dim : chalk_1.default.cyan)(pct.padStart(COL_PCT)));
 }
 function formatTotals(totals) {
     const label = `Total (${totals.requestCount} reqs)`;
@@ -58,13 +63,17 @@ function formatTotals(totals) {
         chalk_1.default.yellow(tokens.padStart(COL_TOKENS)) + '  ' +
         chalk_1.default.green(cost.padStart(COL_COST)));
 }
-function printRequestSavings(model, tokensSaved, costSaved, totals) {
-    recentRequests.unshift({ model, tokensSaved, costSaved });
+function printRequestSavings(model, tokensSaved, costSaved, originalTokenCount, totals, dryRun = false) {
+    const pctSaved = originalTokenCount > 0 ? (tokensSaved / originalTokenCount) * 100 : 0;
+    recentRequests.unshift({ model, tokensSaved, costSaved, pctSaved });
     if (recentRequests.length > MAX_VISIBLE)
         recentRequests.length = MAX_VISIBLE;
     if (linesWritten > 0) {
         process.stdout.write(`\x1b[${linesWritten}A\x1b[0J`);
     }
+    if (dryRun) {
+        console.log(chalk_1.default.yellow.bold('  [DRY RUN] ') + chalk_1.default.dim('Stripping disabled — baseline measurement'));
+    }
     console.log(formatHeader());
     console.log(formatSeparator());
     for (let i = 0; i < recentRequests.length; i++) {
@@ -72,7 +81,7 @@ function printRequestSavings(model, tokensSaved, costSaved, totals) {
     }
     console.log(formatSeparator());
     console.log(formatTotals(totals));
-    linesWritten = recentRequests.length + 4;
+    linesWritten = recentRequests.length + 4 + (dryRun ? 1 : 0);
 }
 function _reset() {
     recentRequests.length = 0;

package/dist/proxy/cloud-api.service.d.ts

@@ -24,6 +24,7 @@ export interface CloudApiRequestPayload {
     }>;
     originalRequestTokenCount: number;
     countMethod: string;
+    isDryRun: boolean;
     upstreamResponseUsage: {
         input_tokens: number;
         output_tokens: number;

package/dist/proxy/cloud-api.service.js

@@ -40,7 +40,8 @@ let CloudApiService = CloudApiService_1 = class CloudApiService {
                 this.logger.warn(`Cloud API returned ${response.status}`);
                 return null;
             }
-            return (await response.json());
+            const body = (await response.json());
+            return body;
         }
         catch (err) {
             this.logger.warn(`Cloud API request failed: ${err.message}`);

package/dist/proxy/proxy.controller.d.ts

@@ -19,6 +19,7 @@ export declare class ProxyController {
     private readonly analyticsService;
     private readonly logger;
     private readonly proxy;
+    private readonly isDryRun;
     constructor(config: ConfigService, tokenCountService: TokenCountService, analyticsService: AnalyticsService);
     messages(req: Request, res: Response, next: NextFunction): Promise<void>;
     countTokens(req: Request, res: Response, next: NextFunction): Promise<void>;

package/dist/proxy/proxy.controller.js

@@ -81,10 +81,12 @@ let ProxyController = ProxyController_1 = class ProxyController {
     analyticsService;
     logger = new common_1.Logger(ProxyController_1.name);
     proxy;
+    isDryRun;
     constructor(config, tokenCountService, analyticsService) {
         this.config = config;
         this.tokenCountService = tokenCountService;
         this.analyticsService = analyticsService;
+        this.isDryRun = this.config.get('TOKENDIET_DRY_RUN') === '1';
         this.proxy = (0, http_proxy_middleware_1.createProxyMiddleware)({
             target: this.config.get('PROXY_TARGET_URL') ?? DEFAULT_UPSTREAM,
             selfHandleResponse: true,
@@ -95,8 +97,11 @@ let ProxyController = ProxyController_1 = class ProxyController {
                     req.__requestedAt = new Date().toISOString();
                     req.__requestPath = req.originalUrl ?? req.path;
                     req.__preStripCountPromise = this.tokenCountService.count(req.__originalBody);
-                    req.body = (0, message_stripper_1.stripMessageBody)(req.body);
-                    req.body = (0, system_tools_stripper_1.stripSystemAndTools)(req.body);
+                    if (!this.isDryRun) {
+                        req.body = (0, message_stripper_1.stripMessageBody)(req.body);
+                        req.body = (0, system_tools_stripper_1.stripSystemAndTools)(req.body);
+                    }
+                    req.__localDelta = this.tokenCountService.countLocalDelta(req.__originalBody, req.body);
                     (0, http_proxy_middleware_1.fixRequestBody)(_proxyReq, req);
                 },
                 proxyRes: (0, http_proxy_middleware_1.responseInterceptor)(async (responseBuffer, proxyRes, req) => {
@@ -120,6 +125,17 @@ let ProxyController = ProxyController_1 = class ProxyController {
                     }
                     if (!preStripResult)
                         return responseBuffer;
+                    this.logger.debug(`countMethod: ${preStripResult.countMethod}`);
+                    const localDelta = req.__localDelta;
+                    if (preStripResult.countMethod === 'local' && localDelta) {
+                        const usageTotal = usage.input_tokens
+                            + (usage.cache_read_input_tokens ?? 0)
+                            + (usage.cache_creation_input_tokens ?? 0);
+                        preStripResult = {
+                            input_tokens: usageTotal + localDelta.delta,
+                            countMethod: 'delta',
+                        };
+                    }
                     try {
                         const cloudResponse = await this.analyticsService.record({
                             model,
@@ -132,7 +148,7 @@ let ProxyController = ProxyController_1 = class ProxyController {
                             requestedAt,
                         });
                         if (cloudResponse) {
-                            (0, cli_output_1.printRequestSavings)(model, cloudResponse.savings?.tokensSaved ?? 0, cloudResponse.savings?.costSaved ?? 0, this.analyticsService.getSessionTotals());
+                            (0, cli_output_1.printRequestSavings)(model, cloudResponse.savings?.tokensSaved ?? 0, cloudResponse.savings?.costSaved ?? 0, cloudResponse.originalRequestTokenCount ?? 0, this.analyticsService.getSessionTotals(), this.isDryRun);
                         }
                     }
                     catch (err) {

package/dist/proxy/token-count.service.d.ts

@@ -1,5 +1,5 @@
 import { ConfigService } from '@nestjs/config';
-export type CountMethod = 'api' | 'local';
+export type CountMethod = 'api' | 'local' | 'delta';
 export interface TokenCountResult {
     input_tokens: number;
     countMethod: CountMethod;
@@ -28,6 +28,24 @@ export declare class TokenCountService {
         system?: any;
         tools?: any;
     }): Promise<TokenCountResult>;
+    countLocalDelta(originalBody: {
+        messages?: Array<{
+            role: string;
+            content: string | any[];
+        }>;
+        system?: any;
+        tools?: any;
+    }, strippedBody: {
+        messages?: Array<{
+            role: string;
+            content: string | any[];
+        }>;
+        system?: any;
+        tools?: any;
+    }): {
+        delta: number;
+        countMethod: 'delta';
+    };
     countLocally(body: {
         messages?: Array<{
             role: string;

package/dist/proxy/token-count.service.js

@@ -65,9 +65,14 @@ let TokenCountService = TokenCountService_1 = class TokenCountService {
             params.tools = sanitized.tools;
         }
         const result = await this.client.messages.countTokens(params);
-        this.logger.log(`count_tokens response: ${JSON.stringify(result)}`);
+        this.logger.debug(`count_tokens response: ${JSON.stringify(result)}`);
         return { input_tokens: result.input_tokens, countMethod: 'api' };
     }
+    countLocalDelta(originalBody, strippedBody) {
+        const originalCount = this.countLocally(originalBody).input_tokens;
+        const strippedCount = this.countLocally(strippedBody).input_tokens;
+        return { delta: Math.max(0, originalCount - strippedCount), countMethod: 'delta' };
+    }
     countLocally(body) {
         const { encode } = require('gpt-tokenizer');
         let total = 0;

package/dist/token-stripper/provider-detector.util.d.ts

@@ -9,8 +9,3 @@ export interface LLMProvider {
     modelInUrlPath?: boolean;
     modelUrlPattern?: RegExp;
 }
-export declare class ProviderDetector {
-    detect(hostname: string, path: string): LLMProvider | null;
-    isLLMHost(hostname: string): boolean;
-    getAllHosts(): string[];
-}

package/dist/token-stripper/provider-detector.util.js

@@ -1,13 +1,5 @@
 "use strict";
-var __decorate = (this && this.__decorate) || function (decorators, target, key, desc) {
-    var c = arguments.length, r = c < 3 ? target : desc === null ? desc = Object.getOwnPropertyDescriptor(target, key) : desc, d;
-    if (typeof Reflect === "object" && typeof Reflect.decorate === "function") r = Reflect.decorate(decorators, target, key, desc);
-    else for (var i = decorators.length - 1; i >= 0; i--) if (d = decorators[i]) r = (c < 3 ? d(r) : c > 3 ? d(target, key, r) : d(target, key)) || r;
-    return c > 3 && r && Object.defineProperty(target, key, r), r;
-};
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.ProviderDetector = void 0;
-const common_1 = require("@nestjs/common");
 const KNOWN_PROVIDERS = [
     {
         name: 'anthropic',
@@ -93,28 +85,4 @@ const KNOWN_PROVIDERS = [
         modelUrlPattern: /\/v1\/engines\/([^/]+)\/completions/,
     },
 ];
-let ProviderDetector = class ProviderDetector {
-    detect(hostname, path) {
-        for (const provider of KNOWN_PROVIDERS) {
-            if (!hostname.includes(provider.host)) {
-                continue;
-            }
-            const pathMatches = provider.requestPaths.some((validPath) => path.startsWith(validPath));
-            if (pathMatches) {
-                return provider;
-            }
-        }
-        return null;
-    }
-    isLLMHost(hostname) {
-        return KNOWN_PROVIDERS.some((p) => hostname.includes(p.host));
-    }
-    getAllHosts() {
-        return KNOWN_PROVIDERS.map((p) => p.host);
-    }
-};
-exports.ProviderDetector = ProviderDetector;
-exports.ProviderDetector = ProviderDetector = __decorate([
-    (0, common_1.Injectable)()
-], ProviderDetector);
 //# sourceMappingURL=provider-detector.util.js.map

package/dist/utils/message-stripper.d.ts

@@ -1,3 +1,5 @@
+export declare function isCatNFormatted(text: string): boolean;
+export declare function stripReadPrefixes(text: string): string;
 export declare function normalizeWhitespace(text: string): string;
 export declare function normalizeToolResultWhitespace(text: string): string;
 export declare function stripMessageBody(body: unknown): unknown;

package/dist/utils/message-stripper.js

@@ -1,8 +1,37 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
+exports.isCatNFormatted = isCatNFormatted;
+exports.stripReadPrefixes = stripReadPrefixes;
 exports.normalizeWhitespace = normalizeWhitespace;
 exports.normalizeToolResultWhitespace = normalizeToolResultWhitespace;
 exports.stripMessageBody = stripMessageBody;
+const CAT_N_PREFIX = /^[ \t]*\d+[\t→]/;
+function isCatNFormatted(text) {
+    const lines = text.split('\n');
+    let checked = 0;
+    for (const line of lines) {
+        if (line.trim() === '')
+            continue;
+        if (!CAT_N_PREFIX.test(line))
+            return false;
+        checked++;
+        if (checked >= 3)
+            break;
+    }
+    return checked > 0;
+}
+function stripReadPrefixes(text) {
+    if (!isCatNFormatted(text))
+        return normalizeToolResultWhitespace(text);
+    return text
+        .split('\n')
+        .map((line) => line.replace(CAT_N_PREFIX, ''))
+        .join('\n')
+        .replace(/[ \t]+$/gm, '')
+        .replace(/\n{3,}/g, '\n\n')
+        .replace(/^\n+/, '')
+        .replace(/\n+$/, '');
+}
 function normalizeWhitespace(text) {
     return text
         .replace(/[ \t]+/g, ' ')
@@ -22,7 +51,9 @@ function stripMessageBody(body) {
     const obj = body;
     if (!Array.isArray(obj.messages))
         return body;
-    return {
+    let catNStripped = 0;
+    let toolResultsTotal = 0;
+    const result = {
         ...obj,
         messages: obj.messages.map((msg) => {
             if (!msg || typeof msg !== 'object')
@@ -32,40 +63,55 @@ function stripMessageBody(body) {
                 return { ...m, content: normalizeWhitespace(m.content) };
             }
             if (Array.isArray(m.content)) {
-                return {
-                    ...m,
-                    content: m.content.map((block) => {
-                        if (!block || typeof block !== 'object')
-                            return block;
-                        const b = block;
-                        if (b.type === 'text') {
-                            return { ...b, text: normalizeWhitespace(b.text) };
+                const normalized = m.content.map((block) => {
+                    if (!block || typeof block !== 'object')
+                        return block;
+                    const b = block;
+                    if (b.type === 'text') {
+                        const text = normalizeWhitespace(b.text);
+                        if (text === '')
+                            return null;
+                        return { ...b, text };
+                    }
+                    if (b.type === 'tool_result') {
+                        toolResultsTotal++;
+                        if (typeof b.content === 'string') {
+                            const wasCatN = isCatNFormatted(b.content);
+                            if (wasCatN)
+                                catNStripped++;
+                            return { ...b, content: stripReadPrefixes(b.content) };
                         }
-                        if (b.type === 'tool_result') {
-                            if (typeof b.content === 'string') {
-                                return { ...b, content: normalizeToolResultWhitespace(b.content) };
-                            }
-                            if (Array.isArray(b.content)) {
-                                return {
-                                    ...b,
-                                    content: b.content.map((sub) => {
-                                        if (sub &&
-                                            typeof sub === 'object' &&
-                                            sub.type === 'text' &&
-                                            typeof sub.text === 'string') {
-                                            return { ...sub, text: normalizeToolResultWhitespace(sub.text) };
-                                        }
-                                        return sub;
-                                    }),
-                                };
-                            }
+                        if (Array.isArray(b.content)) {
+                            return {
+                                ...b,
+                                content: b.content.map((sub) => {
+                                    if (sub &&
+                                        typeof sub === 'object' &&
+                                        sub.type === 'text' &&
+                                        typeof sub.text === 'string') {
+                                        const wasCatN = isCatNFormatted(sub.text);
+                                        if (wasCatN)
+                                            catNStripped++;
+                                        return { ...sub, text: stripReadPrefixes(sub.text) };
+                                    }
+                                    return sub;
+                                }),
+                            };
                         }
-                        return block;
-                    }),
+                    }
+                    return block;
+                }).filter((block) => block !== null);
+                return {
+                    ...m,
+                    content: normalized,
                 };
             }
             return msg;
         }),
     };
+    if (toolResultsTotal > 0) {
+        console.log(`[stripMessageBody] tool_results: ${toolResultsTotal}, cat-n stripped: ${catNStripped}`);
+    }
+    return result;
 }
 //# sourceMappingURL=message-stripper.js.map

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "tokendiet",
-  "version": "1.4.0",
+  "version": "1.6.0",
   "description": "Token optimization gateway CLI for Claude Code",
   "author": "vippelina",
   "private": false,