Method: projects.locations.collections.engines.assistants.streamAssist

Assists the user with a query in a streaming fashion.

HTTP request

POST https://discoveryengine.googleapis.com/v1alpha/{name=projects/*/locations/*/collections/*/engines/*/assistants/*}:streamAssist

The URL uses gRPC Transcoding syntax.

Path parameters

Parameters
name

string

Required. The resource name of the Assistant. Format: projects/{project}/locations/{location}/collections/{collection}/engines/{engine}/assistants/{assistant}

Request body

The request body contains data with the following structure:

JSON representation 
{
  "query": {
    object (Query)
  },
  "answerGenerationMode": enum (AnswerGenerationMode),
  "session": string,
  "assistSkippingMode": enum (AssistSkippingMode),
  "fileIds": [
    string
  ],
  "userMetadata": {
    object (AssistUserMetadata)
  },
  "languageCode": string,
  "googleSearchGroundingEnabled": boolean,
  "webGroundingEnabled": boolean,
  "dataStoreSpecs": [
    {
      object (DataStoreSpec)
    }
  ],
  "assistGenerationConfig": {
    object (AssistGenerationConfig)
  },
  "cannedQuery": string,
  "toolsSpec": {
    object (ToolsSpec)
  },
  "generationSpec": {
    object (GenerationSpec)
  }
}
Fields
query

object (Query)

Optional. Current user query.

Empty query is only supported if fileIds are provided. In this case, the answer will be generated based on those context files.
answerGenerationMode

enum (AnswerGenerationMode)

Optional. The answer generation mode. If not specified, it defaults to NORMAL.
session

string

Optional. The session to use for the request. If specified, the assistant has access to the session history, and the query and the answer are stored there.

If - is specified as the session ID, or it is left empty, then a new session is created with an automatically generated id.

Format: projects/{project}/locations/{location}/collections/{collection}/engines/{engine}/sessions/{session}
assistSkippingMode

enum (AssistSkippingMode)

Optional. Determines the cases in which the assistant will not answer the query. If the answer is skipped, the reasons will be listed in AssistAnswer.assist_skipped_reasons within AssistResponse.answer.
fileIds[]

string

Optional. The IDs of the files to be used for answering the request. When uploading a file, its ID is returned in AddContextFileResponse.file_id.
userMetadata

object (AssistUserMetadata)

Optional. Information about the user initiating the query.
languageCode

string

Optional. Language to be used for answering if language detection fails. Also used as the language of error messages created by actions, regardless of language detection results.
googleSearchGroundingEnabled
(deprecated)

boolean

Optional. Deprecated. Please refrain from using this field.
webGroundingEnabled
(deprecated)

boolean

Optional. Deprecated. Please refrain from using this field.
dataStoreSpecs[]
(deprecated)

object (DataStoreSpec)

Optional. Deprecated: Use [dataStoreSpecs][ToolSpec.VertexAiSearchSpec.data_store_specs] instead. Specs defining DataStores to filter on in a search call and configurations for those data stores. This is only considered for Engines with multiple data stores.

Not setting this field will result in using all data stores in the engine, except when ignoreDataStores is set to true.
assistGenerationConfig
(deprecated)

object (AssistGenerationConfig)

Optional. Deprecated. Use generationSpec instead. Request level generation configuration. This is used to override the default generation configuration at the engine level.
cannedQuery

string

Optional. The resource name of the CannedQuery for this request. This parameter can only be set if the session is new, otherwise an INVALID_ARGUMENT error is returned.

A session is considered new if there is a StreamAssistRequest.session provided and the corresponding Session does not have any turns in it, or if there is no StreamAssistRequest.session provided (and a new Session gets created).

If the cannedQuery parameter is given, the corresponding CannedQuery will be taken into account during processing. The exact behaviour is described and documented in the CannedQuery.

Format: projects/{project}/locations/{location}/collections/{collection}/engines/{engine}/assistants/{assistant}/cannedQueries/{cannedQuery}
toolsSpec

object (ToolsSpec)

Optional. Specification of tools that are used to serve the request.
generationSpec

object (GenerationSpec)

Optional. Specification of the generation configuration for the request.

Response body

Response for the AssistantService.StreamAssist method.

If successful, the response body contains data with the following structure:

JSON representation 
{
  "answer": {
    object (AssistAnswer)
  },
  "sessionInfo": {
    object (SessionInfo)
  },
  "statusUpdates": [
    {
      object (StatusUpdate)
    }
  ],
  "assistToken": string,
  "toolResult": [
    {
      object (ToolResult)
    }
  ],
  "finalResultToolInvocationId": string
}
Fields
answer

object (AssistAnswer)

assistants.assist answer resource object containing parts of the assistant's final answer for the user's query.

Not present if the current response doesn't add anything to previously sent AssistAnswer.replies.

Observe AssistAnswer.state to see if more parts are to be expected. While the state is IN_PROGRESS, the AssistAnswer.replies field in each response will contain replies (reply fragments) to be appended to the ones received in previous responses. AssistAnswer.name won't be filled.

If the state is SUCCEEDED, FAILED or SKIPPED, the response is the last response and AssistAnswer.name will have a value.
sessionInfo

object (SessionInfo)

Session information.
statusUpdates[]

object (StatusUpdate)

Optional progress updates.
assistToken

string

A global unique ID that identifies the current pair of request and stream of responses. Used for feedback and support.
toolResult[]

object (ToolResult)

Results of tool invocations.
finalResultToolInvocationId

string

If this field is present, it contains the id of the tool invocation whose result is used as the final answer.

Authorization scopes

Requires the following OAuth scope:

  • https://www.googleapis.com/auth/cloud-platform

For more information, see the Authentication Overview.

IAM Permissions

Requires the following IAM permission on the name resource:

  • discoveryengine.assistants.assist

For more information, see the IAM documentation.

AnswerGenerationMode

Available answer generation modes.

Enums
ANSWER_GENERATION_MODE_UNSPECIFIED Unspecified answer generation mode.
NORMAL Normal answer generation mode. May trigger a few searches and call actions.
RESEARCH Research answer generation mode. Deep dives into the query, triggering many searches for a wide range of topics related to the query.

AssistSkippingMode

Available skipping behaviors for assist requests.

Enums
ASSIST_SKIPPING_MODE_UNSPECIFIED Default value. The answer can be skipped if the query isn't seeking assistance.
REQUEST_ASSIST Request Assistant answer. The answer may still be skipped if the query fails policy checks.

AssistUserMetadata

User metadata of the request.

JSON representation 
{
  "timeZone": string
}
Fields
timeZone

string

Optional. IANA time zone, e.g. Europe/Budapest.

AssistGenerationConfig

Assistant generation configuration to be used for the request.

JSON representation 
{
  "modelId": string
}
Fields
modelId

string

Optional. The Vertex AI modelId used for the generative model. If not set, the default Assistant model will be used.

ToolsSpec

Specification of tools that are used to serve the request.

JSON representation 
{
  "toolRegistry": string,
  "vertexAiSearchSpec": {
    object (VertexAiSearchSpec)
  },
  "webGroundingSpec": {
    object (WebGroundingSpec)
  },
  "imageGenerationSpec": {
    object (ImageGenerationSpec)
  },
  "videoGenerationSpec": {
    object (VideoGenerationSpec)
  }
}
Fields
toolRegistry

string

Optional. The name of the tool registry to use. Format: projects/{project}/locations/{location}/toolRegistries/{toolRegistry}
vertexAiSearchSpec

object (VertexAiSearchSpec)

Optional. Specification of the Vertex AI Search tool.
webGroundingSpec

object (WebGroundingSpec)

Optional. Specification of the web grounding tool. If field is present, enables grounding with web search. Works only if Assistant.web_grounding_type is [WEB_GROUNDING_TYPE_GOOGLE_SEARCH][] or [WEB_GROUNDING_TYPE_ENTERPRISE_WEB_SEARCH][].
imageGenerationSpec

object (ImageGenerationSpec)

Optional. Specification of the image generation tool.
videoGenerationSpec

object (VideoGenerationSpec)

Optional. Specification of the video generation tool.

VertexAiSearchSpec

Specification of the Vertex AI Search tool.

JSON representation 
{
  "disabled": boolean,
  "dataStoreSpecs": [
    {
      object (DataStoreSpec)
    }
  ],
  "filter": string
}
Fields
disabled
(deprecated)

boolean

Optional. Deprecated. Please refrain from using this field. Whether the Vertex AI Search tool is disabled. Default value is false, the tool is enabled by default.
dataStoreSpecs[]

object (DataStoreSpec)

Optional. Specs defining DataStores to filter on in a search call and configurations for those data stores. This is only considered for Engines with multiple data stores.
filter

string

Optional. The filter syntax consists of an expression language for constructing a predicate from one or more fields of the documents being filtered. Filter expression is case-sensitive.

If this field is unrecognizable, an INVALID_ARGUMENT is returned.

Filtering in Vertex AI Search is done by mapping the LHS filter key to a key property defined in the Vertex AI Search backend -- this mapping is defined by the customer in their schema. For example a media customer might have a field 'name' in their schema. In this case the filter would look like this: filter --> name:'ANY("king kong")'

For more information about filtering including syntax and filter operators, see Filter

WebGroundingSpec

Specification of the web grounding tool.

JSON representation 
{
  "enabled": boolean
}
Fields
enabled
(deprecated)

boolean

Optional. Deprecated. Please refrain from using this field. Whether the web grounding tool is enabled.

ImageGenerationSpec

This type has no fields.

Specification of the image generation tool.

VideoGenerationSpec

This type has no fields.

Specification of the video generation tool.

GenerationSpec

Assistant generation specification for the request. This allows to override the default generation configuration at the engine level.

JSON representation 
{
  "modelId": string
}
Fields
modelId

string

Optional. The Vertex AI modelId used for the generative model. If not set, the default Assistant model will be used.

SessionInfo

Information about the session.

JSON representation 
{
  "session": string,
  "queryId": string
}
Fields
session

string

Name of the newly generated or continued session.

Format: projects/{project}/locations/{location}/collections/{collection}/engines/{engine}/sessions/{session}.
queryId

string

Query ID that corresponds to this assist API call. One session can have multiple turns, each with a unique query ID.

By specifying the session name and this query id in the assistants.assist API call, the assist request happens in the context of the search results from this search call.

StatusUpdate

Update about how the assist call is progressing.

JSON representation 
{
  "updateText": string,
  "invocationId": string
}
Fields
updateText

string

Free text status status update, e.g., 'Starting internal search in HR docs for benefits', or 'Web search for "Moon landing" got 12 results'.
invocationId

string

Unique identifier of the invocation that generates the update.

ToolResult

(Partial) result of a tool invocation.

JSON representation 
{
  "toolInvocationId": string,
  "result": {
    object (Reply)
  }
}
Fields
toolInvocationId

string

Unique identifier of the tool invocation that generates the result.
result

object (Reply)

Partial content of the result.