feat: add serviceTier parameter to Responses API LLM (#1206)

Author: piyush-gambhir

.changeset/add-service-tier-param.md

@@ -0,0 +1,5 @@
+---
+'@livekit/agents-plugin-openai': patch
+---
+
+add serviceTier parameter to Responses API LLM for parity with Chat Completions LLM

plugins/openai/src/responses/llm.ts

@@ -26,6 +26,8 @@ export interface LLMOptions {
   store?: boolean;
   metadata?: Record<string, string>;
   strictToolSchema?: boolean;
+  /** Specifies the processing tier (e.g. 'auto', 'default', 'priority', 'flex'). */
+  serviceTier?: string;
 
   /**
    * Whether to use the WebSocket API.
@@ -114,6 +116,10 @@ class ResponsesHttpLLM extends llm.LLM {
       modelOptions.metadata = this.#opts.metadata;
     }
 
+    if (this.#opts.serviceTier) {
+      modelOptions.service_tier = this.#opts.serviceTier;
+    }
+
     return new ResponsesHttpLLMStream(this, {
       model: this.#opts.model,
       client: this.#client,

plugins/openai/src/ws/llm.ts

@@ -141,6 +141,8 @@ export interface WSLLMOptions {
   store?: boolean;
   metadata?: Record<string, string>;
   strictToolSchema?: boolean;
+  /** Specifies the processing tier (e.g. 'auto', 'default', 'priority', 'flex'). */
+  serviceTier?: string;
 }
 
 const defaultLLMOptions: WSLLMOptions = {
@@ -266,6 +268,10 @@ export class WSLLM extends llm.LLM {
       modelOptions.metadata = this.#opts.metadata;
     }
 
+    if (this.#opts.serviceTier) {
+      modelOptions.service_tier = this.#opts.serviceTier;
+    }
+
     let inputChatCtx = chatCtx;
     let prevResponseId: string | undefined;
     const canUseStoredResponse = modelOptions.store !== false;