💭 fix: Preserve Custom Endpoint Reasoning Params

api/server/services/Config/loadCustomConfig.js

@@ -177,7 +177,8 @@ https://www.librechat.ai/docs/configuration/stt_tts`);
 
 // Validate and fill out missing values for custom parameters
 function parseCustomParams(endpointName, customParams) {
-  const paramEndpoint = customParams.defaultParamsEndpoint;
+  const paramEndpoint = customParams.defaultParamsEndpoint ?? 'custom';
+  customParams.defaultParamsEndpoint = paramEndpoint;
   customParams.paramDefinitions = customParams.paramDefinitions || [];
 
   // Checks if `defaultParamsEndpoint` is a key in `paramSettings`.

api/server/services/Config/loadCustomConfig.spec.js

@@ -11,7 +11,7 @@ jest.mock('librechat-data-provider', () => {
     paramSettings: {
       foo: {},
       bar: {},
-      custom: {},
+      custom: [],
       openrouter: [
         {
           key: 'promptCache',
@@ -59,6 +59,7 @@ jest.mock('@librechat/data-schemas', () => {
 const axios = require('axios');
 const { loadYaml } = require('@librechat/api');
 const { logger } = require('@librechat/data-schemas');
+const { ReasoningParameterFormat, ReasoningResponseKey } = require('librechat-data-provider');
 const loadCustomConfig = require('./loadCustomConfig');
 
 describe('loadCustomConfig', () => {
@@ -307,11 +308,28 @@ describe('loadCustomConfig', () => {
       );
     });
 
-    it('throws an error when defaultParamsEndpoint is not provided', async () => {
-      const malformedCustomParams = { defaultParamsEndpoint: undefined };
-      await expect(loadCustomParams(malformedCustomParams)).rejects.toThrow(
-        'defaultParamsEndpoint of "Google" endpoint is invalid. Valid options are foo, bar, custom, openrouter, google',
-      );
+    it('defaults defaultParamsEndpoint when only reasoningFormat is provided', async () => {
+      const parsedConfig = await loadCustomParams({
+        reasoningFormat: ReasoningParameterFormat.reasoningObject,
+      });
+
+      expect(parsedConfig.endpoints.custom[0].customParams).toEqual({
+        defaultParamsEndpoint: 'custom',
+        reasoningFormat: ReasoningParameterFormat.reasoningObject,
+        paramDefinitions: [],
+      });
+    });
+
+    it('defaults defaultParamsEndpoint when only reasoningKey is provided', async () => {
+      const parsedConfig = await loadCustomParams({
+        reasoningKey: ReasoningResponseKey.reasoning,
+      });
+
+      expect(parsedConfig.endpoints.custom[0].customParams).toEqual({
+        defaultParamsEndpoint: 'custom',
+        reasoningKey: ReasoningResponseKey.reasoning,
+        paramDefinitions: [],
+      });
     });
 
     it('fills the paramDefinitions with missing values', async () => {

packages/api/src/agents/initialize.ts

@@ -15,6 +15,7 @@ import type {
   AgentToolResources,
   AgentToolOptions,
   TEndpointOption,
+  ReasoningResponseKey,
   TFile,
   Agent,
   TUser,
@@ -169,6 +170,8 @@ export type InitializedAgent = Agent & {
   actionsEnabled?: boolean;
   /** Maximum characters allowed in a single tool result before truncation. */
   maxToolResultChars?: number;
+  /** Response field to read model reasoning from for custom OpenAI-compatible endpoints. */
+  reasoningKey?: ReasoningResponseKey;
   /**
    * Whether the code-execution environment is available *for this agent*.
    * Narrower than the incoming `params.codeEnvAvailable` admin flag — this
@@ -1035,6 +1038,7 @@ export async function initializeAgent(
     actionsEnabled,
     baseContextTokens,
     codeEnvAvailable: effectiveCodeEnvAvailable,
+    reasoningKey: customEndpointConfig?.customParams?.reasoningKey,
     skillCount,
     accessibleSkillIds: executableSkillIds,
     activeSkillNames,

packages/api/src/agents/run.spec.ts

@@ -1,5 +1,6 @@
 import { Providers } from '@librechat/agents';
 import { ToolMessage, AIMessage, HumanMessage } from '@librechat/agents/langchain/messages';
+import { ReasoningResponseKey } from 'librechat-data-provider';
 
 import {
   extractDiscoveredToolsFromHistory,
@@ -150,6 +151,56 @@ describe('getReasoningKey', () => {
 
     expect(reasoningKey).toBe('reasoning');
   });
+
+  it('keeps Vercel AI Gateway on ChatOpenAI normalized reasoning_content', () => {
+    const llmConfig = {
+      configuration: {
+        baseURL: 'https://ai-gateway.vercel.sh/v1',
+      },
+    } as Parameters<typeof getReasoningKey>[1];
+
+    const reasoningKey = getReasoningKey(Providers.OPENAI, llmConfig);
+
+    expect(reasoningKey).toBe('reasoning_content');
+  });
+
+  it('keeps Vercel custom endpoint names on ChatOpenAI normalized reasoning_content', () => {
+    const llmConfig = {} as Parameters<typeof getReasoningKey>[1];
+
+    const reasoningKey = getReasoningKey(Providers.OPENAI, llmConfig, 'Vercel');
+
+    expect(reasoningKey).toBe('reasoning_content');
+  });
+
+  it('uses explicit reasoning response keys for Vercel when configured', () => {
+    const llmConfig = {
+      configuration: {
+        baseURL: 'https://ai-gateway.vercel.sh/v1',
+      },
+    } as Parameters<typeof getReasoningKey>[1];
+
+    const reasoningKey = getReasoningKey(
+      Providers.OPENAI,
+      llmConfig,
+      'Vercel',
+      ReasoningResponseKey.reasoning,
+    );
+
+    expect(reasoningKey).toBe('reasoning');
+  });
+
+  it('uses explicit reasoning response keys for otherwise default OpenAI-compatible endpoints', () => {
+    const llmConfig = {} as Parameters<typeof getReasoningKey>[1];
+
+    const reasoningKey = getReasoningKey(
+      Providers.OPENAI,
+      llmConfig,
+      'Company Gateway',
+      ReasoningResponseKey.reasoning,
+    );
+
+    expect(reasoningKey).toBe('reasoning');
+  });
 });
 
 describe('isDeepSeekReasoningProvider', () => {

packages/api/src/agents/run.ts

@@ -26,6 +26,7 @@ import type {
   Agent,
   AgentModelParameters,
   AgentSubagentsConfig,
+  ReasoningResponseKey,
   SummarizationConfig,
 } from 'librechat-data-provider';
 import type { BaseMessage } from '@librechat/agents/langchain/messages';
@@ -212,6 +213,8 @@ const customProviders = new Set([
   KnownEndpoints.ollama,
 ]);
 
+type AgentReasoningKey = 'reasoning_content' | 'reasoning';
+
 function includesOpenRouter(value?: string | null): boolean {
   return typeof value === 'string' && value.toLowerCase().includes(KnownEndpoints.openrouter);
 }
@@ -220,8 +223,13 @@ export function getReasoningKey(
   provider: Providers,
   llmConfig: t.RunLLMConfig,
   agentEndpoint?: string | null,
-): 'reasoning_content' | 'reasoning' {
-  let reasoningKey: 'reasoning_content' | 'reasoning' = 'reasoning_content';
+  customReasoningKey?: ReasoningResponseKey,
+): AgentReasoningKey {
+  if (customReasoningKey) {
+    return customReasoningKey as AgentReasoningKey;
+  }
+
+  let reasoningKey: AgentReasoningKey = 'reasoning_content';
   if (provider === Providers.GOOGLE) {
     reasoningKey = 'reasoning';
   } else if (
@@ -293,6 +301,8 @@ type RunAgent = Omit<Agent, 'tools'> & {
   codeEnvAvailable?: boolean;
   /** Optional per-agent summarization overrides */
   summarization?: SummarizationConfig;
+  /** Response field to read model reasoning from for custom OpenAI-compatible endpoints. */
+  reasoningKey?: ReasoningResponseKey;
   /**
    * Maximum characters allowed in a single tool result before truncation.
    * Overrides the default computed from maxContextTokens.
@@ -946,7 +956,7 @@ export async function createRun({
       agent.maxContextTokens,
     );
 
-    const reasoningKey = getReasoningKey(provider, llmConfig, agent.endpoint);
+    const reasoningKey = getReasoningKey(provider, llmConfig, agent.endpoint, agent.reasoningKey);
     return {
       provider,
       reasoningKey,

packages/api/src/endpoints/openai/config.backward-compat.spec.ts

@@ -10,7 +10,7 @@ describe('getOpenAIConfig - Backward Compatibility', () => {
   describe('OpenAI endpoint', () => {
     it('should handle GPT-5 model with reasoning and web search', () => {
       const apiKey = 'sk-proj-somekey';
-      const endpoint = undefined;
+      const endpoint = EModelEndpoint.openAI;
       const options = {
         modelOptions: {
           model: 'gpt-5-nano',
@@ -138,7 +138,7 @@ describe('getOpenAIConfig - Backward Compatibility', () => {
 
     it('should handle Azure OpenAI with Responses API and reasoning', () => {
       const apiKey = 'some_azure_key';
-      const endpoint = undefined;
+      const endpoint = EModelEndpoint.azureOpenAI;
       const options = {
         modelOptions: {
           model: 'gpt-5',
@@ -395,6 +395,7 @@ describe('getOpenAIConfig - Backward Compatibility', () => {
         modelOptions: {
           model: '@cf/deepseek-ai/deepseek-r1-distill-qwen-32b',
           user: 'some-user',
+          reasoning_effort: ReasoningEffort.high,
         },
         reverseProxyUrl:
           'https://gateway.ai.cloudflare.com/v1/${CF_ACCOUNT_ID}/${CF_GATEWAY_ID}/workers-ai/v1',
@@ -419,6 +420,9 @@ describe('getOpenAIConfig - Backward Compatibility', () => {
           user: 'some-user',
           disableStreaming: true,
           apiKey: 'someKey',
+          modelKwargs: {
+            reasoning_effort: ReasoningEffort.high,
+          },
         },
         configOptions: {
           baseURL: