feat: Introduce RAG as context/memory store for Gemini Live API

Google APIs · copybara-github · commit 07ab54bca52b · 2025-05-27T12:48:23.000-07:00
docs: A comment for message `RagFileParsingConfig` is updated.
docs: A comment for field `global_max_embedding_requests_per_min` in message `.google.cloud.aiplatform.v1beta1.ImportRagFilesConfig` is updated.

PiperOrigin-RevId: 763909021
diff --git a/google/cloud/aiplatform/v1beta1/tool.proto b/google/cloud/aiplatform/v1beta1/tool.proto
@@ -315,6 +315,15 @@ message VertexRagStore {
   // Optional. The retrieval config for the Rag query.
   RagRetrievalConfig rag_retrieval_config = 6
       [(google.api.field_behavior) = OPTIONAL];
+
+  // Optional. Currently only supported for Gemini Multimodal Live API.
+  //
+  // In Gemini Multimodal Live API, if `store_context` bool is
+  // true, Gemini will leverage it to automatically memorize the
+  // interactions between the client and Gemini, and retrieve context when
+  // needed to augment the response generation for users' ongoing and future
+  // interactions.
+  bool store_context = 7 [(google.api.field_behavior) = OPTIONAL];
 }
 
 // Retrieve from Vertex AI Search datastore or engine for grounding.
diff --git a/google/cloud/aiplatform/v1beta1/vertex_rag_data.proto b/google/cloud/aiplatform/v1beta1/vertex_rag_data.proto
@@ -295,6 +295,29 @@ message RagCorpus {
     singular: "ragCorpus"
   };
 
+  // The config for the corpus type of the RagCorpus.
+  message CorpusTypeConfig {
+    // Config for the document corpus.
+    message DocumentCorpus {}
+
+    // Config for the memory corpus.
+    message MemoryCorpus {
+      // The LLM parser to use for the memory corpus.
+      RagFileParsingConfig.LlmParser llm_parser = 1;
+    }
+
+    // Optional.
+    // Whether the RagCorpus is used as document store or memory store.
+    oneof corpus_type_config {
+      // Optional. Config for the document corpus.
+      DocumentCorpus document_corpus = 1
+          [(google.api.field_behavior) = OPTIONAL];
+
+      // Optional. Config for the memory corpus.
+      MemoryCorpus memory_corpus = 2 [(google.api.field_behavior) = OPTIONAL];
+    }
+  }
+
   // The backend config of the RagCorpus.
   // It can be data store and/or retrieval engine.
   oneof backend_config {
@@ -358,6 +381,10 @@ message RagCorpus {
     (google.api.field_behavior) = IMMUTABLE,
     (google.api.field_behavior) = OPTIONAL
   ];
+
+  // Optional. The corpus type config of the RagCorpus.
+  CorpusTypeConfig corpus_type_config = 13
+      [(google.api.field_behavior) = OPTIONAL];
 }
 
 // A RagFile contains user data for chunking, embedding and indexing.
@@ -521,7 +548,7 @@ message RagFileParsingConfig {
     int32 global_max_parsing_requests_per_min = 3;
   }
 
-  // Specifies the advanced parsing for RagFiles.
+  // Specifies the LLM parsing for RagFiles.
   message LlmParser {
     // The name of a LLM model used for parsing.
     // Format:
@@ -652,7 +679,7 @@ message ImportRagFilesConfig {
   // Optional. The max number of queries per minute that the indexing pipeline
   // job is allowed to make to the embedding model specified in the project.
   // Please follow the quota usage guideline of the embedding model you use to
-  // set the value properly. If this value is not specified,
+  // set the value properly.If this value is not specified,
   // max_embedding_requests_per_min will be used by indexing pipeline job as the
   // global limit.
   int32 global_max_embedding_requests_per_min = 18
