[AI Infra] Add new temperature parameter to Inference AI, and OpenAI, Bedrock, Gemini connectors (#239806)

## Summary

Follow up of #239626. This PR
addresses #239181.

If a user defines/sets a temperature at the connector level, the
connector's temperature will override the Kibana's default temperature
for each connector. If temperature is not explicitly set by a user, then
we respect the the temperature set by the request.

AI connector
<img width="866" height="880" alt="Screenshot 2025-10-28 at 09 16 29"
src="https://github.com/user-attachments/assets/e5d28754-0b4d-46df-99b7-43c26bb9ab5a"
/>

<img width="866" height="880" alt="Screenshot 2025-10-28 at 09 20 18"
src="https://github.com/user-attachments/assets/71e3a125-e8a3-4bd9-a04d-c34b29e0b7a1"
/>
<img width="866" height="880" alt="Screenshot 2025-10-28 at 09 18 18"
src="https://github.com/user-attachments/assets/6fa62eb0-24dc-4245-bc3f-a84fa10b34c4"
/>
<img width="866" height="880" alt="Screenshot 2025-10-28 at 09 18 03"
src="https://github.com/user-attachments/assets/b60391fd-f23f-40c5-b8a8-87c419f5b2c6"
/>


Before:
<img width="1896" height="1386" alt="image"
src="https://github.com/user-attachments/assets/6213ee44-d53e-440e-95c3-ac1abaded3c0"
/>

After:



### Checklist

Check the PR satisfies following conditions. 

Reviewers should verify this PR satisfies this list as well.

- [ ] Any text added follows [EUI's writing
guidelines](https://elastic.github.io/eui/#/guidelines/writing), uses
sentence case text and includes [i18n
support](https://github.com/elastic/kibana/blob/main/src/platform/packages/shared/kbn-i18n/README.md)
- [ ]
[Documentation](https://www.elastic.co/guide/en/kibana/master/development-documentation.html)
was added for features that require explanation or tutorials
- [ ] [Unit or functional
tests](https://www.elastic.co/guide/en/kibana/master/development-tests.html)
were updated or added to match the most common scenarios
- [ ] If a plugin configuration key changed, check if it needs to be
allowlisted in the cloud and added to the [docker
list](https://github.com/elastic/kibana/blob/main/src/dev/build/tasks/os_packages/docker_generator/resources/base/bin/kibana-docker)
- [ ] This was checked for breaking HTTP API changes, and any breaking
changes have been approved by the breaking-change committee. The
`release_note:breaking` label should be applied in these situations.
- [ ] [Flaky Test
Runner](https://ci-stats.kibana.dev/trigger_flaky_test_runner/1) was
used on any tests changed
- [ ] The PR description includes the appropriate Release Notes section,
and the correct `release_note:*` label is applied per the
[guidelines](https://www.elastic.co/guide/en/kibana/master/contributing.html#kibana-release-notes-process)
- [ ] Review the [backport
guidelines](https://docs.google.com/document/d/1VyN5k91e5OVumlc0Gb9RPa3h1ewuPE705nRtioPiTvY/edit?usp=sharing)
and apply applicable `backport:*` labels.

### Identify risks

Does this PR introduce any risks? For example, consider risks like hard
to test bugs, performance regression, potential of data loss.

Describe the risk, its severity, and mitigation for each identified
risk. Invite stakeholders and evaluate how to proceed before merging.

- [ ] [See some risk
examples](https://github.com/elastic/kibana/blob/main/RISK_MATRIX.mdx)
- [ ] ...

---------

Co-authored-by: Elastic Machine <[email protected]>
Co-authored-by: Melissa <[email protected]>

Author: 3 people

x-pack/platform/packages/shared/kbn-inference-endpoint-ui-common/src/components/additional_options_fields.tsx

@@ -56,6 +56,7 @@ interface AdditionalOptionsFieldsProps {
   taskTypeOptions: TaskTypeOption[];
   isEdit?: boolean;
   allowContextWindowLength?: boolean;
+  allowTemperature?: boolean;
 }
 
 export const AdditionalOptionsFields: React.FC<AdditionalOptionsFieldsProps> = ({
@@ -65,6 +66,7 @@ export const AdditionalOptionsFields: React.FC<AdditionalOptionsFieldsProps> = (
   onTaskTypeOptionsSelect,
   isEdit,
   allowContextWindowLength,
+  allowTemperature,
 }) => {
   const { euiTheme } = useEuiTheme();
   const { setFieldValue } = useFormContext();
@@ -175,6 +177,100 @@ export const AdditionalOptionsFields: React.FC<AdditionalOptionsFieldsProps> = (
     ]
   );
 
+  const temperatureSettings = useMemo(
+    () =>
+      (selectedTaskType === CHAT_COMPLETION_TASK_TYPE || selectedTaskType === DEFAULT_TASK_TYPE) &&
+      allowTemperature ? (
+        <>
+          <EuiTitle size="xxs" data-test-subj="temperature-details-label">
+            <EuiFlexGroup alignItems="center" justifyContent="spaceBetween">
+              <EuiFlexItem grow={false}>
+                <h4>
+                  <FormattedMessage
+                    id="xpack.inferenceEndpointUICommon.components.additionalInfo.temperatureLabel"
+                    defaultMessage="Temperature"
+                  />
+                </h4>
+              </EuiFlexItem>
+              <EuiFlexItem grow={false}>
+                <EuiText color="subdued" size="xs">
+                  {LABELS.OPTIONALTEXT}
+                </EuiText>
+              </EuiFlexItem>
+            </EuiFlexGroup>
+          </EuiTitle>
+          <EuiText size="xs" color="subdued">
+            <FormattedMessage
+              id="xpack.inferenceEndpointUICommon.components.additionalInfo.temperatureHelpInfo"
+              defaultMessage="Controls the randomness of the model's output. Changing the temperature can affect the general performance of AI Assistant and AI-driven features in Kibana, and we recommend keeping the default value."
+            />
+          </EuiText>
+          <EuiSpacer size="m" />
+          <UseField
+            path="config.temperature"
+            config={{
+              validations: [
+                {
+                  validator: ({ value, path }) => {
+                    if (value !== undefined && value !== null && value !== '') {
+                      const numValue = Number(value);
+                      if (isNaN(numValue) || numValue < 0) {
+                        return {
+                          code: 'ERR_FIELD_INVALID',
+                          path,
+                          message: LABELS.TEMPERATURE_VALIDATION_MESSAGE,
+                        };
+                      }
+                    }
+                  },
+                  isBlocking: false,
+                },
+              ],
+            }}
+          >
+            {(field) => {
+              const { isInvalid, errorMessage } = getFieldValidityAndErrorMessage(field);
+              return (
+                <EuiFormRow
+                  id="temperatureSettings"
+                  label={LABELS.TEMPERATURE_LABEL}
+                  fullWidth
+                  isInvalid={isInvalid}
+                  error={errorMessage}
+                  data-test-subj={'configuration-formrow-temperatureSettings'}
+                >
+                  <EuiFormControlLayout
+                    fullWidth
+                    clear={{
+                      onClick: (e) => {
+                        setFieldValue('config.temperature', undefined);
+                      },
+                    }}
+                  >
+                    <EuiFieldNumber
+                      min={0}
+                      max={1}
+                      step={0.1}
+                      fullWidth
+                      data-test-subj={'temperatureSettingsNumber'}
+                      value={config.temperature ?? ''}
+                      isInvalid={isInvalid}
+                      onChange={(e) => {
+                        const value = e.target.value;
+                        setFieldValue('config.temperature', value === '' ? undefined : value);
+                      }}
+                    />
+                  </EuiFormControlLayout>
+                </EuiFormRow>
+              );
+            }}
+          </UseField>
+          <EuiSpacer size="m" />
+        </>
+      ) : null,
+    [setFieldValue, config.temperature, selectedTaskType, allowTemperature]
+  );
+
   const taskTypeSettings = useMemo(
     () =>
       selectedTaskType || config.taskType?.length ? (
@@ -289,6 +385,12 @@ export const AdditionalOptionsFields: React.FC<AdditionalOptionsFieldsProps> = (
             return (
               <EuiFormRow
                 id="inferenceId"
+                label={
+                  <FormattedMessage
+                    id="xpack.inferenceEndpointUICommon.components.additionalInfo.inferenceIdLabel"
+                    defaultMessage="Inference ID"
+                  />
+                }
                 isInvalid={isInvalid}
                 error={errorMessage}
                 fullWidth
@@ -338,6 +440,7 @@ export const AdditionalOptionsFields: React.FC<AdditionalOptionsFieldsProps> = (
         </UseField>
         <EuiSpacer size="m" />
         {contextWindowLengthSettings}
+        {temperatureSettings}
       </EuiPanel>
     </EuiAccordion>
   );

x-pack/platform/packages/shared/kbn-inference-endpoint-ui-common/src/components/inference_flyout_wrapper.tsx

@@ -126,6 +126,7 @@ export const InferenceFlyoutWrapper: React.FC<InferenceFlyoutWrapperProps> = ({
         provider: inferenceEndpoint?.config.provider ?? '',
         providerConfig: inferenceEndpoint?.config.providerConfig,
         contextWindowLength: inferenceEndpoint?.config.contextWindowLength ?? undefined,
+        temperature: inferenceEndpoint?.config.temperature ?? undefined,
       },
       secrets: {
         providerSecrets: {},

x-pack/platform/packages/shared/kbn-inference-endpoint-ui-common/src/components/inference_service_form_fields.tsx

@@ -99,6 +99,7 @@ interface InferenceServicesProps {
     isPreconfigured?: boolean;
     allowContextWindowLength?: boolean;
     reenterSecretsOnEdit?: boolean;
+    allowTemperature?: boolean;
   };
   http: HttpSetup;
   toasts: IToasts;
@@ -109,6 +110,7 @@ export const InferenceServiceFormFields: React.FC<InferenceServicesProps> = ({
   toasts,
   config: {
     allowContextWindowLength,
+    allowTemperature,
     isEdit,
     enforceAdaptiveAllocations,
     isPreconfigured,
@@ -140,6 +142,7 @@ export const InferenceServiceFormFields: React.FC<InferenceServicesProps> = ({
       'config.taskType',
       'config.inferenceId',
       'config.contextWindowLength',
+      'config.temperature',
       'config.provider',
       'config.providerConfig',
     ],
@@ -585,6 +588,7 @@ export const InferenceServiceFormFields: React.FC<InferenceServicesProps> = ({
             selectedTaskType={selectedTaskType}
             isEdit={isEdit}
             allowContextWindowLength={allowContextWindowLength}
+            allowTemperature={allowTemperature}
           />
           {/* HIDDEN VALIDATION */}
           <ProviderSecretHiddenField

x-pack/platform/packages/shared/kbn-inference-endpoint-ui-common/src/translations.ts

@@ -223,3 +223,17 @@ export const CONTEXT_WINDOW_TASK_TYPE_VALIDATION_MESSAGE = i18n.translate(
     defaultMessage: 'Context window length is only applicable for chat completion tasks.',
   }
 );
+
+export const TEMPERATURE_LABEL = i18n.translate(
+  'xpack.inferenceEndpointUICommon.components.temperatureTextFieldLabel',
+  {
+    defaultMessage: 'Temperature',
+  }
+);
+
+export const TEMPERATURE_VALIDATION_MESSAGE = i18n.translate(
+  'xpack.inferenceEndpointUICommon.components.temperatureValidationMessage',
+  {
+    defaultMessage: 'Temperature must be a number between 0 and 1.',
+  }
+);

x-pack/platform/packages/shared/kbn-inference-endpoint-ui-common/src/types/types.ts

@@ -28,6 +28,7 @@ export interface Config {
   provider: string;
   providerConfig?: Record<string, unknown>;
   contextWindowLength?: number;
+  temperature?: number;
 }
 
 export interface Secrets {

x-pack/platform/plugins/shared/inference/server/chat_complete/adapters/bedrock/bedrock_claude_adapter.test.ts

@@ -9,15 +9,19 @@ import { PassThrough } from 'stream';
 import { loggerMock } from '@kbn/logging-mocks';
 import { noop } from 'rxjs';
 import type { InferenceExecutor } from '../../utils/inference_executor';
-import { MessageRole, ToolChoiceType } from '@kbn/inference-common';
+import { MessageRole, ToolChoiceType, InferenceConnectorType } from '@kbn/inference-common';
 import { bedrockClaudeAdapter } from './bedrock_claude_adapter';
 import { addNoToolUsageDirective } from './prompts';
 import { lastValueFrom, toArray } from 'rxjs';
 describe('bedrockClaudeAdapter', () => {
   const logger = loggerMock.create();
   const executorMock = {
     invoke: jest.fn(),
-  } as InferenceExecutor & { invoke: jest.MockedFn<InferenceExecutor['invoke']> };
+    getConnector: jest.fn(),
+  } as InferenceExecutor & {
+    invoke: jest.MockedFn<InferenceExecutor['invoke']>;
+    getConnector: jest.MockedFn<InferenceExecutor['getConnector']>;
+  };
 
   beforeEach(() => {
     executorMock.invoke.mockReset();
@@ -31,6 +35,14 @@ describe('bedrockClaudeAdapter', () => {
         },
       };
     });
+    executorMock.getConnector.mockReset();
+    executorMock.getConnector.mockReturnValue({
+      type: InferenceConnectorType.Bedrock,
+      name: 'bedrock-connector',
+      connectorId: 'test-connector-id',
+      config: {},
+      capabilities: {},
+    });
   });
 
   function getCallParams() {

x-pack/platform/plugins/shared/inference/server/chat_complete/adapters/bedrock/bedrock_claude_adapter.ts

@@ -24,6 +24,7 @@ import type { ConverseCompletionChunk } from './process_completion_chunks';
 import { processConverseCompletionChunks } from './process_completion_chunks';
 import { addNoToolUsageDirective } from './prompts';
 import { toolChoiceToConverse, toolsToConverseBedrock } from './convert_tools';
+import { getTemperatureIfValid } from '../../utils/get_temperature';
 
 export const bedrockClaudeAdapter: InferenceConnectorAdapter = {
   chatComplete: ({
@@ -47,13 +48,14 @@ export const bedrockClaudeAdapter: InferenceConnectorAdapter = {
       ? [{ text: addNoToolUsageDirective(system) }]
       : [{ text: system }];
     const bedRockTools = noToolUsage ? [] : toolsToConverseBedrock(tools, messages);
+    const connector = executor.getConnector();
 
     const subActionParams = {
       system: systemMessage,
       messages: converseMessages,
       tools: bedRockTools?.length ? bedRockTools : undefined,
       toolChoice: toolChoiceToConverse(toolChoice),
-      temperature,
+      ...getTemperatureIfValid(temperature, { connector, modelName }),
       model: modelName,
       stopSequences: ['\n\nHuman:'],
       signal: abortSignal,

x-pack/platform/plugins/shared/inference/server/chat_complete/adapters/gemini/gemini_adapter.test.ts

@@ -11,17 +11,29 @@ import { noop, tap, lastValueFrom, toArray, of } from 'rxjs';
 import { loggerMock } from '@kbn/logging-mocks';
 import type { InferenceExecutor } from '../../utils/inference_executor';
 import { observableIntoEventSourceStream } from '../../../util/observable_into_event_source_stream';
-import { MessageRole, ToolChoiceType } from '@kbn/inference-common';
+import { MessageRole, ToolChoiceType, InferenceConnectorType } from '@kbn/inference-common';
 import { geminiAdapter } from './gemini_adapter';
 
 describe('geminiAdapter', () => {
   const logger = loggerMock.create();
   const executorMock = {
     invoke: jest.fn(),
-  } as InferenceExecutor & { invoke: jest.MockedFn<InferenceExecutor['invoke']> };
+    getConnector: jest.fn(),
+  } as InferenceExecutor & {
+    invoke: jest.MockedFn<InferenceExecutor['invoke']>;
+    getConnector: jest.MockedFn<InferenceExecutor['getConnector']>;
+  };
 
   beforeEach(() => {
     executorMock.invoke.mockReset();
+    executorMock.getConnector.mockReset();
+    executorMock.getConnector.mockReturnValue({
+      type: InferenceConnectorType.Gemini,
+      name: 'gemini-connector',
+      connectorId: 'test-connector-id',
+      config: {},
+      capabilities: {},
+    });
     processVertexStreamMock.mockReset().mockImplementation(() => tap(noop));
   });
 

x-pack/platform/plugins/shared/inference/server/chat_complete/adapters/gemini/gemini_adapter.ts

@@ -14,6 +14,7 @@ import { handleConnectorResponse } from '../../utils';
 import { eventSourceStreamIntoObservable } from '../../../util/event_source_stream_into_observable';
 import { processVertexStream } from './process_vertex_stream';
 import type { GenerateContentResponseChunk, GeminiMessage, GeminiToolConfig } from './types';
+import { getTemperatureIfValid } from '../../utils/get_temperature';
 
 export const geminiAdapter: InferenceConnectorAdapter = {
   chatComplete: ({
@@ -27,6 +28,7 @@ export const geminiAdapter: InferenceConnectorAdapter = {
     abortSignal,
     metadata,
   }) => {
+    const connector = executor.getConnector();
     return defer(() => {
       return executor.invoke({
         subAction: 'invokeStream',
@@ -35,7 +37,7 @@ export const geminiAdapter: InferenceConnectorAdapter = {
           systemInstruction: system,
           tools: toolsToGemini(tools),
           toolConfig: toolChoiceToConfig(toolChoice),
-          temperature,
+          ...getTemperatureIfValid(temperature, { connector, modelName }),
           model: modelName,
           signal: abortSignal,
           stopSequences: ['\n\nHuman:'],

x-pack/platform/plugins/shared/inference/server/chat_complete/utils/get_temperature.ts

@@ -13,8 +13,14 @@ export const getTemperatureIfValid = (
   temperature?: number,
   { connector, modelName }: { connector?: InferenceConnector; modelName?: string } = {}
 ) => {
-  if (temperature === undefined) return {};
+  // If user sets temperature in the connector config, use it by default
+  if (connector?.config?.temperature) {
+    return { temperature: connector.config.temperature };
+  }
+
+  if (temperature === undefined || temperature < 0) return {};
 
+  // Else, use the temperature from the request
   const model =
     modelName ?? connector?.config?.providerConfig?.model_id ?? connector?.config?.defaultModel;
 
@@ -24,6 +30,7 @@ export const getTemperatureIfValid = (
     model
   ) {
     const normalizedModelName = model.toLowerCase();
+
     const shouldExcludeTemperature = OPENAI_MODELS_WITHOUT_TEMPERATURE.some(
       //  e.g openai/gpt-5 or gpt-5-xxx
       (m) => normalizedModelName.startsWith(m) || normalizedModelName.endsWith(m)