Feat/custom token rates for endpoints

api/models/tx.js

@@ -1,6 +1,49 @@
 const { matchModelName } = require('../utils');
 const defaultRate = 6;
 
+const customTokenOverrides = {};
+const customCacheOverrides = {};
+
+/**
+ * Allows overriding the default token multipliers.
+ *
+ * @param {Object} overrides - An object mapping model keys to their custom token multipliers.
+ * @param {Object} overrides.<model> - An object containing custom multipliers for the model.
+ * @param {number} overrides.<model>.prompt - The custom prompt multiplier for the model.
+ * @param {number} overrides.<model>.completion - The custom completion multiplier for the model.
+ *
+ * @example
+ * // Override the multipliers for "gpt-4o-mini" and "gpt-3.5":
+ * setCustomTokenOverrides({
+ *   "gpt-4o-mini": { prompt: 0.2, completion: 0.5 },
+ *   "gpt-3.5": { prompt: 1.0, completion: 2.0 }
+ * });
+ */
+const setCustomTokenOverrides = (overrides) => {
+  Object.assign(customTokenOverrides, overrides);
+};
+
+/**
+ * Allows overriding the default cache multipliers.
+ * The override values should be nested under a key named "Cache".
+ *
+ * @param {Object} overrides - An object mapping model keys to their custom cache multipliers.
+ * @param {Object} overrides.<model> - An object that must include a "Cache" property.
+ * @param {Object} overrides.<model>.Cache - An object containing custom cache multipliers for the model.
+ * @param {number} overrides.<model>.Cache.write - The custom cache write multiplier for the model.
+ * @param {number} overrides.<model>.Cache.read - The custom cache read multiplier for the model.
+ *
+ * @example
+ * // Override the cache multipliers for "gpt-4o-mini" and "gpt-3.5":
+ * setCustomCacheOverrides({
+ *   "gpt-4o-mini": { cache: { write: 0.2, read: 0.5 } },
+ *   "gpt-3.5": { cache: { write: 1.0, read: 1.5 } }
+ * });
+ */
+const setCustomCacheOverrides = (overrides) => {
+  Object.assign(customCacheOverrides, overrides);
+};
+
 /**
  * AWS Bedrock pricing
  * source: https://aws.amazon.com/bedrock/pricing/
@@ -243,20 +286,23 @@ const getCacheMultiplier = ({ valueKey, cacheType, model, endpoint, endpointToke
     return endpointTokenConfig?.[model]?.[cacheType] ?? null;
   }
 
-  if (valueKey && cacheType) {
-    return cacheTokenValues[valueKey]?.[cacheType] ?? null;
+  if (!valueKey && model) {
+    valueKey = getValueKey(model, endpoint);
   }
-
-  if (!cacheType || !model) {
+  if (!valueKey) {
     return null;
   }
 
-  valueKey = getValueKey(model, endpoint);
-  if (!valueKey) {
-    return null;
+  // Check for custom cache overrides under the "cache" property.
+  if (
+    customCacheOverrides[valueKey] &&
+    customCacheOverrides[valueKey].cache &&
+    customCacheOverrides[valueKey].cache[cacheType] != null
+  ) {
+    return customCacheOverrides[valueKey].cache[cacheType];
   }
 
-  // If we got this far, and values[cacheType] is undefined somehow, return a rough average of default multipliers
+  // Fallback to the default cacheTokenValues.
   return cacheTokenValues[valueKey]?.[cacheType] ?? null;
 };
 
@@ -267,4 +313,6 @@ module.exports = {
   getCacheMultiplier,
   defaultRate,
   cacheTokenValues,
+  setCustomTokenOverrides,
+  setCustomCacheOverrides,
 };

api/server/services/AppService.js

@@ -12,6 +12,7 @@ const { agentsConfigSetup } = require('./start/agents');
 const { initializeRoles } = require('~/models/Role');
 const { getMCPManager } = require('~/config');
 const paths = require('~/config/paths');
+const { loadTokenRatesConfig } = require('./Config/loadTokenRatesConfig');
 
 /**
  *
@@ -21,9 +22,10 @@ const paths = require('~/config/paths');
  */
 const AppService = async (app) => {
   await initializeRoles();
-  /** @type {TCustomConfig}*/
+  /** @type {TCustomConfig} */
   const config = (await loadCustomConfig()) ?? {};
   const configDefaults = getConfigDefaults();
+  loadTokenRatesConfig(config, configDefaults);
 
   const filteredTools = config.filteredTools;
   const includedTools = config.includedTools;

api/server/services/Config/loadTokenRatesConfig.js

@@ -0,0 +1,71 @@
+const { removeNullishValues } = require('librechat-data-provider');
+const { logger } = require('~/config');
+const { setCustomTokenOverrides, setCustomCacheOverrides } = require('~/models/tx');
+
+/**
+ * Loads token rates from the user's configuration, merging with default token rates if available.
+ *
+ * @param {TCustomConfig | undefined} config - The loaded custom configuration.
+ * @param {TConfigDefaults} [configDefaults] - Optional default configuration values.
+ * @returns {TCustomConfig['tokenRates']} - The final token rates configuration.
+ */
+function loadTokenRatesConfig(config, configDefaults) {
+  const userTokenRates = removeNullishValues(config?.tokenRates ?? {});
+
+  if (!configDefaults?.tokenRates) {
+    logger.info(`User tokenRates configuration:\n${JSON.stringify(userTokenRates, null, 2)}`);
+    // Apply custom token rates even if there are no defaults
+    applyCustomTokenRates(userTokenRates);
+    return userTokenRates;
+  }
+
+  /** @type {TCustomConfig['tokenRates']} */
+  const defaultTokenRates = removeNullishValues(configDefaults.tokenRates);
+  const merged = { ...defaultTokenRates, ...userTokenRates };
+
+  // Apply custom token rates configuration
+  applyCustomTokenRates(merged);
+
+  logger.info(`Merged tokenRates configuration:\n${JSON.stringify(merged, null, 2)}`);
+  return merged;
+}
+
+/**
+ * Processes the token rates configuration to set up custom overrides for each model.
+ *
+ * The configuration is expected to be specified per model:
+ *
+ * For each model in the tokenRates configuration, this function will call the tx.js
+ * override functions to apply the custom token and cache multipliers.
+ *
+ * @param {TModelTokenRates} tokenRates - The token rates configuration mapping models to token costs.
+ */
+function applyCustomTokenRates(tokenRates) {
+  // Iterate over each model in the tokenRates configuration.
+  Object.keys(tokenRates).forEach((model) => {
+    const rate = tokenRates[model];
+    // If token multipliers are provided, set custom token overrides.
+    if (rate.prompt != null || rate.completion != null) {
+      setCustomTokenOverrides({
+        [model]: {
+          prompt: rate.prompt,
+          completion: rate.completion,
+        },
+      });
+    }
+    // Check for cache overrides.
+    const cacheOverrides = rate.cache;
+    if (cacheOverrides && (cacheOverrides.write != null || cacheOverrides.read != null)) {
+      setCustomCacheOverrides({
+        [model]: {
+          cache: {
+            write: cacheOverrides.write,
+            read: cacheOverrides.read,
+          },
+        },
+      });
+    }
+  });
+}
+
+module.exports = { loadTokenRatesConfig };

librechat.example.yaml

@@ -68,6 +68,17 @@ interface:
   multiConvo: true
   agents: true
 
+# Example Custom Token Rates (optional)
+#tokenRates:
+#  gpt-4o-mini:
+#    prompt: 200.0
+#    completion: 400.0
+#  claude-3.7-sonnet:
+#    Cache:
+#      read: 200.0
+#      write: 400.0
+
+
 # Example Registration Object Structure (optional)
 registration:
   socialLogins: ['github', 'google', 'discord', 'openid', 'facebook', 'apple']

packages/data-provider/src/config.ts

@@ -526,12 +526,37 @@ export type TStartupConfig = {
   helpAndFaqURL: string;
   customFooter?: string;
   modelSpecs?: TSpecsConfig;
+  tokenRates?: TModelTokenRates;
   sharedLinksEnabled: boolean;
   publicSharedLinksEnabled: boolean;
   analyticsGtmId?: string;
   instanceProjectId: string;
 };
 
+// Token cost schema type
+export type TTokenCost = {
+  prompt?: number;
+  completion?: number;
+  cache?: {
+    write?: number;
+    read?: number;
+  };
+};
+
+// Endpoint token rates schema type
+export type TModelTokenRates = Record<string, TTokenCost>;
+
+const tokenCostSchema = z.object({
+  prompt: z.number().optional(),     // e.g. 1.5 => $1.50 / 1M tokens
+  completion: z.number().optional(), // e.g. 2.0 => $2.00 / 1M tokens
+  cache: z
+    .object({
+      write: z.number().optional(),
+      read: z.number().optional(),
+    })
+    .optional(),
+});
+
 export const configSchema = z.object({
   version: z.string(),
   cache: z.boolean().default(true),
@@ -563,6 +588,7 @@ export const configSchema = z.object({
   rateLimits: rateLimitSchema.optional(),
   fileConfig: fileConfigSchema.optional(),
   modelSpecs: specsConfigSchema.optional(),
+  tokenRates: tokenCostSchema.optional(),
   endpoints: z
     .object({
       all: baseEndpointSchema.optional(),