@inductiv/node-red-openai-api 6.22.0 → 6.27.0

package/locales/en-US/node.json

@@ -140,6 +140,7 @@
             "compactModelResponse": "compact model response",
             "listInputItems": "list input items",
             "countInputTokens": "count input tokens",
+            "manageModelResponseWebSocket": "manage model response websocket",
             "listSkills": "list skills",
             "createSkill": "create skill",
             "getSkill": "retrieve skill",

package/node.html

@@ -357,6 +357,11 @@
         data-i18n="OpenaiApi.parameters.countInputTokens"
         ></option>
 
+        <option
+        value="manageModelResponseWebSocket"
+        data-i18n="OpenaiApi.parameters.manageModelResponseWebSocket"
+        ></option>
+
       </optgroup>
 
               <optgroup style="font-style: normal;" label="🔄 Runs (Beta)">
@@ -2673,44 +2678,87 @@
 
         <h4 style="font-weight: bolder;"> ⋙ Create Realtime Client Secret</h4>
         <p>Create an ephemeral client secret for browser/mobile realtime sessions.</p>
+        <p>Put Realtime session configuration inside <code>msg.payload.session</code>. Top-level
+            <code>msg.payload</code> is reserved for client-secret options such as
+            <code>expires_after</code>.</p>
         <dl class="message-properties">
             <h4>msg.payload Properties</h4>
             <dt class="optional">
-                model
+                expires_after
+                <span class="property-type">object</span>
+            </dt>
+            <dd>Optional client-secret expiration config, for example
+                <code>{ "anchor": "created_at", "seconds": 600 }</code>.</dd>
+            <dt class="optional">
+                session
+                <span class="property-type">object</span>
+            </dt>
+            <dd>Realtime session configuration to attach to the generated client secret.</dd>
+            <dt>
+                session.type
                 <span class="property-type">string</span>
             </dt>
-            <dd>Realtime model id.</dd>
+            <dd>Must be <code>realtime</code> for a Realtime session.</dd>
             <dt class="optional">
-                instructions
+                session.model
+                <span class="property-type">string</span>
+            </dt>
+            <dd>Realtime model id, including newer SDK-supported ids such as
+                <code>gpt-realtime-1.5</code> and <code>gpt-audio-1.5</code>.</dd>
+            <dt class="optional">
+                session.instructions
                 <span class="property-type">string</span>
             </dt>
             <dd>Default system instructions for the session.</dd>
             <dt class="optional">
-                output_modalities
+                session.include
+                <span class="property-type">array</span>
+            </dt>
+            <dd>Additional server output fields to include, such as
+                <code>item.input_audio_transcription.logprobs</code>.</dd>
+            <dt class="optional">
+                session.max_output_tokens
+                <span class="property-type">integer | string</span>
+            </dt>
+            <dd>Maximum output tokens per assistant response, or <code>inf</code>.</dd>
+            <dt class="optional">
+                session.output_modalities
                 <span class="property-type">array</span>
             </dt>
             <dd>Output modalities such as <code>text</code> or <code>audio</code>.</dd>
             <dt class="optional">
-                audio
+                session.audio
                 <span class="property-type">object</span>
             </dt>
             <dd>Input/output audio config for the session.</dd>
             <dt class="optional">
-                tools
+                session.prompt
+                <span class="property-type">object</span>
+            </dt>
+            <dd>Reusable prompt reference and variables for the session.</dd>
+            <dt class="optional">
+                session.tools
                 <span class="property-type">array</span>
             </dt>
             <dd>Tools available to the realtime model.</dd>
             <dt class="optional">
-                tool_choice
+                session.tool_choice
                 <span class="property-type">string | object</span>
             </dt>
             <dd>How tools are selected.</dd>
             <dt class="optional">
-                tracing
+                session.tracing
                 <span class="property-type">string | object | null</span>
             </dt>
             <dd>Trace configuration for the realtime session.</dd>
+            <dt class="optional">
+                session.truncation
+                <span class="property-type">string | object</span>
+            </dt>
+            <dd>Conversation truncation behavior for the session.</dd>
         </dl>
+        <p>Successful responses are returned in <code>msg.payload</code> with top-level
+            <code>value</code>, <code>expires_at</code>, and <code>session</code> fields.</p>
 
         <h4 style="font-weight: bolder;"> ⋙ Accept Realtime Call</h4>
         <p>Accept an incoming SIP call and attach it to a realtime session.</p>
@@ -2726,16 +2774,43 @@
                 <span class="property-type">string</span>
             </dt>
             <dd>Must be <code>realtime</code>.</dd>
+            <dt class="optional">
+                include
+                <span class="property-type">array</span>
+            </dt>
+            <dd>Additional server output fields to include, such as
+                <code>item.input_audio_transcription.logprobs</code>.</dd>
             <dt class="optional">
                 model
                 <span class="property-type">string</span>
             </dt>
-            <dd>Realtime model id for the accepted call.</dd>
+            <dd>Realtime model id for the accepted call, including newer SDK-supported ids such as
+                <code>gpt-realtime-1.5</code> and <code>gpt-audio-1.5</code>.</dd>
             <dt class="optional">
                 instructions
                 <span class="property-type">string</span>
             </dt>
             <dd>System instructions for the call session.</dd>
+            <dt class="optional">
+                max_output_tokens
+                <span class="property-type">integer | string</span>
+            </dt>
+            <dd>Maximum output tokens per assistant response, or <code>inf</code>.</dd>
+            <dt class="optional">
+                output_modalities
+                <span class="property-type">array</span>
+            </dt>
+            <dd>Output modalities such as <code>text</code> or <code>audio</code>.</dd>
+            <dt class="optional">
+                audio
+                <span class="property-type">object</span>
+            </dt>
+            <dd>Input/output audio config for the accepted call session.</dd>
+            <dt class="optional">
+                prompt
+                <span class="property-type">object</span>
+            </dt>
+            <dd>Reusable prompt reference and variables for the call session.</dd>
             <dt class="optional">
                 tools
                 <span class="property-type">array</span>
@@ -2746,6 +2821,16 @@
                 <span class="property-type">string | object</span>
             </dt>
             <dd>Tool selection strategy.</dd>
+            <dt class="optional">
+                tracing
+                <span class="property-type">string | object | null</span>
+            </dt>
+            <dd>Trace configuration for the accepted call session.</dd>
+            <dt class="optional">
+                truncation
+                <span class="property-type">string | object</span>
+            </dt>
+            <dd>Conversation truncation behavior for the call session.</dd>
         </dl>
 
         <h4 style="font-weight: bolder;"> ⋙ Hangup Realtime Call</h4>
@@ -2809,13 +2894,20 @@
                 input
                 <span class="property-type">string | array</span>
             </dt>
-            <dd>Text, image, or file inputs to the model, used to generate a response.</dd>
+            <dd>
+                Text, image, or file inputs to the model, used to generate a response.
+                For assistant message items in an array input, you may also set
+                <code>phase</code> to <code>commentary</code> or <code>final_answer</code>.
+            </dd>
 
             <dt>
                 model
                 <span class="property-type">string</span>
             </dt>
-            <dd>Model ID used to generate the response.</dd>
+            <dd>
+                Model ID used to generate the response, including newer Responses-compatible
+                model strings such as <code>gpt-5.4</code>.
+            </dd>
 
             <dt class="optional">
                 background
@@ -2950,6 +3042,13 @@
             </dt>
             <dd>
                 An array of tools the model may call while generating a response.
+                This includes Model Context Protocol tools, Tool Search via
+                <code>{ "type": "tool_search" }</code>, and GA computer use via
+                <code>{ "type": "computer" }</code>.
+                Deferred tool loading is supported by setting
+                <code>defer_loading: true</code> on tool definitions.
+                Computer-tool workflows typically require follow-up requests that send
+                <code>computer_call_output</code> items back through <code>input</code>.
                 For Model Context Protocol (MCP) tools, each object in the array should follow <a
                     href="https://platform.openai.com/docs/guides/tools-remote-mcp?lang=javascript&utm_source==%40inductiv%2Fnode-red-openai-api"
                     target="_blank">this
@@ -3183,6 +3282,73 @@
 
         </dl>
 
+        <h4 style="font-weight: bolder;"> ⋙ Manage Model Response WebSocket</h4>
+        <p>
+            Opens, uses, and closes a persistent Responses websocket connection from a single
+            <code>OpenAI API</code> node instance.
+        </p>
+        <p>
+            Use the same node instance for all websocket actions. Set <code>msg.payload.action</code> to
+            <code>connect</code>, <code>send</code>, or <code>close</code>.
+        </p>
+        <p>
+            Incoming server events are emitted asynchronously from this node as new messages with the raw
+            event object in <code>msg.payload</code>. Connection metadata is added under
+            <code>msg.openai</code>.
+        </p>
+        <dl class="message-properties">
+            <h4>msg.payload Properties</h4>
+
+            <dt>
+                action
+                <span class="property-type">string</span>
+            </dt>
+            <dd>
+                Required lifecycle action. Must be <code>connect</code>, <code>send</code>, or
+                <code>close</code>.
+            </dd>
+
+            <dt>
+                connection_id
+                <span class="property-type">string</span>
+            </dt>
+            <dd>
+                Stable identifier for the websocket connection owned by this node instance.
+            </dd>
+
+            <dt class="optional">
+                event
+                <span class="property-type">object</span>
+            </dt>
+            <dd>
+                Required when <code>action</code> is <code>send</code>. Provide a Responses client event,
+                currently <code>{ "type": "response.create", ... }</code>.
+            </dd>
+
+            <dt class="optional">
+                code
+                <span class="property-type">integer</span>
+            </dt>
+            <dd>
+                Optional websocket close code used when <code>action</code> is <code>close</code>.
+                Defaults to <code>1000</code>.
+            </dd>
+
+            <dt class="optional">
+                reason
+                <span class="property-type">string</span>
+            </dt>
+            <dd>
+                Optional websocket close reason used when <code>action</code> is <code>close</code>.
+                Defaults to <code>OK</code>.
+            </dd>
+
+        </dl>
+        <p>
+            The websocket URL is derived from the configured <code>Service Host</code> API Base using the
+            same auth contract as the rest of this node, including custom auth headers and query-string auth.
+        </p>
+
     </details>
 </section>
 

package/node.js

@@ -43,6 +43,10 @@ module.exports = function (RED) {
       let node = this;
       node.service = RED.nodes.getNode(config.service);
       node.config = config;
+      node._cleanupHandlers = [];
+      node.registerCleanupHandler = function (handler) {
+        node._cleanupHandlers.push(handler);
+      };
 
       node.on("input", function (msg) {
         if (!node.service) {
@@ -144,6 +148,12 @@ module.exports = function (RED) {
             node.error(errorMessage, msg);
           });
       });
+
+      node.on("close", function (done) {
+        Promise.all(node._cleanupHandlers.map((handler) => handler())).then(
+          () => done()
+        ).catch(done);
+      });
     }
   }
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@inductiv/node-red-openai-api",
-  "version": "6.22.0",
+  "version": "6.27.0",
   "description": "Enhance your Node-RED projects with advanced AI capabilities.",
   "main": "node.js",
   "engines": {
@@ -33,11 +33,12 @@
     "build:html": "gulp build-html",
     "build:js": "esbuild src/lib.js --bundle --outfile=lib.js --format=cjs --platform=node",
     "build": "npm run build:html && npm run build:js",
-    "test": "node --test test/service-host-node.test.js test/service-host-editor-template.test.js test/openai-node-auth-routing.test.js test/openai-methods-mapping.test.js",
+    "test": "node --test test/service-host-node.test.js test/service-host-editor-template.test.js test/openai-node-auth-routing.test.js test/openai-methods-mapping.test.js test/openai-responses-websocket.test.js",
     "prepare": "npm run build"
   },
   "dependencies": {
-    "openai": "^6.22.0"
+    "openai": "^6.27.0",
+    "ws": "^8.19.0"
   },
   "author": "Allan Bunch",
   "license": "MIT",

package/src/realtime/help.html

@@ -6,44 +6,87 @@
 
         <h4 style="font-weight: bolder;"> ⋙ Create Realtime Client Secret</h4>
         <p>Create an ephemeral client secret for browser/mobile realtime sessions.</p>
+        <p>Put Realtime session configuration inside <code>msg.payload.session</code>. Top-level
+            <code>msg.payload</code> is reserved for client-secret options such as
+            <code>expires_after</code>.</p>
         <dl class="message-properties">
             <h4>msg.payload Properties</h4>
             <dt class="optional">
-                model
+                expires_after
+                <span class="property-type">object</span>
+            </dt>
+            <dd>Optional client-secret expiration config, for example
+                <code>{ "anchor": "created_at", "seconds": 600 }</code>.</dd>
+            <dt class="optional">
+                session
+                <span class="property-type">object</span>
+            </dt>
+            <dd>Realtime session configuration to attach to the generated client secret.</dd>
+            <dt>
+                session.type
                 <span class="property-type">string</span>
             </dt>
-            <dd>Realtime model id.</dd>
+            <dd>Must be <code>realtime</code> for a Realtime session.</dd>
             <dt class="optional">
-                instructions
+                session.model
+                <span class="property-type">string</span>
+            </dt>
+            <dd>Realtime model id, including newer SDK-supported ids such as
+                <code>gpt-realtime-1.5</code> and <code>gpt-audio-1.5</code>.</dd>
+            <dt class="optional">
+                session.instructions
                 <span class="property-type">string</span>
             </dt>
             <dd>Default system instructions for the session.</dd>
             <dt class="optional">
-                output_modalities
+                session.include
+                <span class="property-type">array</span>
+            </dt>
+            <dd>Additional server output fields to include, such as
+                <code>item.input_audio_transcription.logprobs</code>.</dd>
+            <dt class="optional">
+                session.max_output_tokens
+                <span class="property-type">integer | string</span>
+            </dt>
+            <dd>Maximum output tokens per assistant response, or <code>inf</code>.</dd>
+            <dt class="optional">
+                session.output_modalities
                 <span class="property-type">array</span>
             </dt>
             <dd>Output modalities such as <code>text</code> or <code>audio</code>.</dd>
             <dt class="optional">
-                audio
+                session.audio
                 <span class="property-type">object</span>
             </dt>
             <dd>Input/output audio config for the session.</dd>
             <dt class="optional">
-                tools
+                session.prompt
+                <span class="property-type">object</span>
+            </dt>
+            <dd>Reusable prompt reference and variables for the session.</dd>
+            <dt class="optional">
+                session.tools
                 <span class="property-type">array</span>
             </dt>
             <dd>Tools available to the realtime model.</dd>
             <dt class="optional">
-                tool_choice
+                session.tool_choice
                 <span class="property-type">string | object</span>
             </dt>
             <dd>How tools are selected.</dd>
             <dt class="optional">
-                tracing
+                session.tracing
                 <span class="property-type">string | object | null</span>
             </dt>
             <dd>Trace configuration for the realtime session.</dd>
+            <dt class="optional">
+                session.truncation
+                <span class="property-type">string | object</span>
+            </dt>
+            <dd>Conversation truncation behavior for the session.</dd>
         </dl>
+        <p>Successful responses are returned in <code>msg.payload</code> with top-level
+            <code>value</code>, <code>expires_at</code>, and <code>session</code> fields.</p>
 
         <h4 style="font-weight: bolder;"> ⋙ Accept Realtime Call</h4>
         <p>Accept an incoming SIP call and attach it to a realtime session.</p>
@@ -59,16 +102,43 @@
                 <span class="property-type">string</span>
             </dt>
             <dd>Must be <code>realtime</code>.</dd>
+            <dt class="optional">
+                include
+                <span class="property-type">array</span>
+            </dt>
+            <dd>Additional server output fields to include, such as
+                <code>item.input_audio_transcription.logprobs</code>.</dd>
             <dt class="optional">
                 model
                 <span class="property-type">string</span>
             </dt>
-            <dd>Realtime model id for the accepted call.</dd>
+            <dd>Realtime model id for the accepted call, including newer SDK-supported ids such as
+                <code>gpt-realtime-1.5</code> and <code>gpt-audio-1.5</code>.</dd>
             <dt class="optional">
                 instructions
                 <span class="property-type">string</span>
             </dt>
             <dd>System instructions for the call session.</dd>
+            <dt class="optional">
+                max_output_tokens
+                <span class="property-type">integer | string</span>
+            </dt>
+            <dd>Maximum output tokens per assistant response, or <code>inf</code>.</dd>
+            <dt class="optional">
+                output_modalities
+                <span class="property-type">array</span>
+            </dt>
+            <dd>Output modalities such as <code>text</code> or <code>audio</code>.</dd>
+            <dt class="optional">
+                audio
+                <span class="property-type">object</span>
+            </dt>
+            <dd>Input/output audio config for the accepted call session.</dd>
+            <dt class="optional">
+                prompt
+                <span class="property-type">object</span>
+            </dt>
+            <dd>Reusable prompt reference and variables for the call session.</dd>
             <dt class="optional">
                 tools
                 <span class="property-type">array</span>
@@ -79,6 +149,16 @@
                 <span class="property-type">string | object</span>
             </dt>
             <dd>Tool selection strategy.</dd>
+            <dt class="optional">
+                tracing
+                <span class="property-type">string | object | null</span>
+            </dt>
+            <dd>Trace configuration for the accepted call session.</dd>
+            <dt class="optional">
+                truncation
+                <span class="property-type">string | object</span>
+            </dt>
+            <dd>Conversation truncation behavior for the call session.</dd>
         </dl>
 
         <h4 style="font-weight: bolder;"> ⋙ Hangup Realtime Call</h4>

package/src/responses/help.html

@@ -12,13 +12,20 @@
                 input
                 <span class="property-type">string | array</span>
             </dt>
-            <dd>Text, image, or file inputs to the model, used to generate a response.</dd>
+            <dd>
+                Text, image, or file inputs to the model, used to generate a response.
+                For assistant message items in an array input, you may also set
+                <code>phase</code> to <code>commentary</code> or <code>final_answer</code>.
+            </dd>
 
             <dt>
                 model
                 <span class="property-type">string</span>
             </dt>
-            <dd>Model ID used to generate the response.</dd>
+            <dd>
+                Model ID used to generate the response, including newer Responses-compatible
+                model strings such as <code>gpt-5.4</code>.
+            </dd>
 
             <dt class="optional">
                 background
@@ -153,6 +160,13 @@
             </dt>
             <dd>
                 An array of tools the model may call while generating a response.
+                This includes Model Context Protocol tools, Tool Search via
+                <code>{ "type": "tool_search" }</code>, and GA computer use via
+                <code>{ "type": "computer" }</code>.
+                Deferred tool loading is supported by setting
+                <code>defer_loading: true</code> on tool definitions.
+                Computer-tool workflows typically require follow-up requests that send
+                <code>computer_call_output</code> items back through <code>input</code>.
                 For Model Context Protocol (MCP) tools, each object in the array should follow <a
                     href="https://platform.openai.com/docs/guides/tools-remote-mcp?lang=javascript&utm_source==%40inductiv%2Fnode-red-openai-api"
                     target="_blank">this
@@ -386,5 +400,72 @@
 
         </dl>
 
+        <h4 style="font-weight: bolder;"> ⋙ Manage Model Response WebSocket</h4>
+        <p>
+            Opens, uses, and closes a persistent Responses websocket connection from a single
+            <code>OpenAI API</code> node instance.
+        </p>
+        <p>
+            Use the same node instance for all websocket actions. Set <code>msg.payload.action</code> to
+            <code>connect</code>, <code>send</code>, or <code>close</code>.
+        </p>
+        <p>
+            Incoming server events are emitted asynchronously from this node as new messages with the raw
+            event object in <code>msg.payload</code>. Connection metadata is added under
+            <code>msg.openai</code>.
+        </p>
+        <dl class="message-properties">
+            <h4>msg.payload Properties</h4>
+
+            <dt>
+                action
+                <span class="property-type">string</span>
+            </dt>
+            <dd>
+                Required lifecycle action. Must be <code>connect</code>, <code>send</code>, or
+                <code>close</code>.
+            </dd>
+
+            <dt>
+                connection_id
+                <span class="property-type">string</span>
+            </dt>
+            <dd>
+                Stable identifier for the websocket connection owned by this node instance.
+            </dd>
+
+            <dt class="optional">
+                event
+                <span class="property-type">object</span>
+            </dt>
+            <dd>
+                Required when <code>action</code> is <code>send</code>. Provide a Responses client event,
+                currently <code>{ "type": "response.create", ... }</code>.
+            </dd>
+
+            <dt class="optional">
+                code
+                <span class="property-type">integer</span>
+            </dt>
+            <dd>
+                Optional websocket close code used when <code>action</code> is <code>close</code>.
+                Defaults to <code>1000</code>.
+            </dd>
+
+            <dt class="optional">
+                reason
+                <span class="property-type">string</span>
+            </dt>
+            <dd>
+                Optional websocket close reason used when <code>action</code> is <code>close</code>.
+                Defaults to <code>OK</code>.
+            </dd>
+
+        </dl>
+        <p>
+            The websocket URL is derived from the configured <code>Service Host</code> API Base using the
+            same auth contract as the rest of this node, including custom auth headers and query-string auth.
+        </p>
+
     </details>
 </section>