Runs, gRPC: RunService.Attach

Статья создана

Обновлена 21 февраля 2025 г.

gRPC request
AttachRunRequest
ToolResultList
ToolResult
FunctionResult
StreamEvent
StreamCursor
Error
MessageContent
ContentPart
Text
Message
Author
Citation
Source
FileChunk
SearchIndex
ExpirationConfig
TextSearchIndex
ChunkingStrategy
StaticChunkingStrategy
NgramTokenizer
VectorSearchIndex
HybridSearchIndex
CombinationStrategy
MeanCombinationStrategy
ReciprocalRankFusionCombinationStrategy
File
ChunkContent
ToolCallList
ToolCall
FunctionCall

Bi-directional streaming method to interact with a specific run.
Like Listen, Attach streams events from the run, but also allows clients to send events back.
For example, it can be used to submit function call results when the run is waiting for user input.

gRPC request

rpc Attach (stream AttachRunRequest) returns (stream StreamEvent)

AttachRunRequest

{
  "run_id": "string",
  "events_start_idx": "google.protobuf.Int64Value",
  // Includes only one of the fields `tool_result_list`
  "tool_result_list": {
    "tool_results": [
      {
        // Includes only one of the fields `function_result`
        "function_result": {
          "name": "string",
          // Includes only one of the fields `content`
          "content": "string"
          // end of the list of possible fields
        }
        // end of the list of possible fields
      }
    ]
  }
  // end of the list of possible fields
}

Field	Description
run_id	string Required field. ID of the run to listen to.
events_start_idx	google.protobuf.Int64Value Starting index for events. If provided, listening will start from this event.
tool_result_list	ToolResultList A list of tool results to submit to the run, such as the output of a function call. Includes only one of the fields `tool_result_list`. Event to submit to the run.

ToolResultList

Represents a list of tool results.

Field

Description

tool_results[]

ToolResult

A list of tool results.

ToolResult

Represents the result of a tool call.

Field

Description

function_result

FunctionResult

Represents the result of a function call.

Includes only one of the fields function_result.

FunctionResult

Represents the result of a function call.

Field

Description

name

string

The name of the function that was executed.

content

string

The result of the function call, represented as a string.
This field can be used to store the output of the function.

Includes only one of the fields content.

StreamEvent

{
  "event_type": "EventType",
  "stream_cursor": {
    "current_event_idx": "int64",
    "num_user_events_received": "int64"
  },
  // Includes only one of the fields `error`, `partial_message`, `completed_message`, `tool_call_list`
  "error": {
    "code": "int64",
    "message": "string"
  },
  "partial_message": {
    "content": [
      {
        // Includes only one of the fields `text`
        "text": {
          "content": "string"
        }
        // end of the list of possible fields
      }
    ]
  },
  "completed_message": {
    "id": "string",
    "thread_id": "string",
    "created_by": "string",
    "created_at": "google.protobuf.Timestamp",
    "author": {
      "id": "string",
      "role": "string"
    },
    "labels": "map<string, string>",
    "content": {
      "content": [
        {
          // Includes only one of the fields `text`
          "text": {
            "content": "string"
          }
          // end of the list of possible fields
        }
      ]
    },
    "status": "MessageStatus",
    "citations": [
      {
        "sources": [
          {
            // Includes only one of the fields `chunk`
            "chunk": {
              "search_index": {
                "id": "string",
                "folder_id": "string",
                "name": "string",
                "description": "string",
                "created_by": "string",
                "created_at": "google.protobuf.Timestamp",
                "updated_by": "string",
                "updated_at": "google.protobuf.Timestamp",
                "expiration_config": {
                  "expiration_policy": "ExpirationPolicy",
                  "ttl_days": "int64"
                },
                "expires_at": "google.protobuf.Timestamp",
                "labels": "map<string, string>",
                // Includes only one of the fields `text_search_index`, `vector_search_index`, `hybrid_search_index`
                "text_search_index": {
                  "chunking_strategy": {
                    // Includes only one of the fields `static_strategy`
                    "static_strategy": {
                      "max_chunk_size_tokens": "int64",
                      "chunk_overlap_tokens": "int64"
                    }
                    // end of the list of possible fields
                  },
                  // Includes only one of the fields `ngram_tokenizer`
                  "ngram_tokenizer": {
                    "min_gram": "google.protobuf.Int64Value",
                    "max_gram": "google.protobuf.Int64Value"
                  }
                  // end of the list of possible fields
                },
                "vector_search_index": {
                  "doc_embedder_uri": "string",
                  "query_embedder_uri": "string",
                  "chunking_strategy": {
                    // Includes only one of the fields `static_strategy`
                    "static_strategy": {
                      "max_chunk_size_tokens": "int64",
                      "chunk_overlap_tokens": "int64"
                    }
                    // end of the list of possible fields
                  }
                },
                "hybrid_search_index": {
                  "text_search_index": {
                    "chunking_strategy": {
                      // Includes only one of the fields `static_strategy`
                      "static_strategy": {
                        "max_chunk_size_tokens": "int64",
                        "chunk_overlap_tokens": "int64"
                      }
                      // end of the list of possible fields
                    },
                    // Includes only one of the fields `ngram_tokenizer`
                    "ngram_tokenizer": {
                      "min_gram": "google.protobuf.Int64Value",
                      "max_gram": "google.protobuf.Int64Value"
                    }
                    // end of the list of possible fields
                  },
                  "vector_search_index": {
                    "doc_embedder_uri": "string",
                    "query_embedder_uri": "string",
                    "chunking_strategy": {
                      // Includes only one of the fields `static_strategy`
                      "static_strategy": {
                        "max_chunk_size_tokens": "int64",
                        "chunk_overlap_tokens": "int64"
                      }
                      // end of the list of possible fields
                    }
                  },
                  "chunking_strategy": {
                    // Includes only one of the fields `static_strategy`
                    "static_strategy": {
                      "max_chunk_size_tokens": "int64",
                      "chunk_overlap_tokens": "int64"
                    }
                    // end of the list of possible fields
                  },
                  "normalization_strategy": "NormalizationStrategy",
                  "combination_strategy": {
                    // Includes only one of the fields `mean_combination`, `rrf_combination`
                    "mean_combination": {
                      "mean_evaluation_technique": "MeanEvaluationTechnique",
                      "weights": [
                        "double"
                      ]
                    },
                    "rrf_combination": {
                      "k": "google.protobuf.Int64Value"
                    }
                    // end of the list of possible fields
                  }
                }
                // end of the list of possible fields
              },
              "source_file": {
                "id": "string",
                "folder_id": "string",
                "name": "string",
                "description": "string",
                "mime_type": "string",
                "created_by": "string",
                "created_at": "google.protobuf.Timestamp",
                "updated_by": "string",
                "updated_at": "google.protobuf.Timestamp",
                "expiration_config": {
                  "expiration_policy": "ExpirationPolicy",
                  "ttl_days": "int64"
                },
                "expires_at": "google.protobuf.Timestamp",
                "labels": "map<string, string>"
              },
              "content": {
                "content": [
                  {
                    // Includes only one of the fields `text`
                    "text": {
                      "content": "string"
                    }
                    // end of the list of possible fields
                  }
                ]
              }
            }
            // end of the list of possible fields
          }
        ]
      }
    ]
  },
  "tool_call_list": {
    "tool_calls": [
      {
        // Includes only one of the fields `function_call`
        "function_call": {
          "name": "string",
          "arguments": "google.protobuf.Struct"
        }
        // end of the list of possible fields
      }
    ]
  }
  // end of the list of possible fields
}

Represents an event in the stream of a run.

Field	Description
event_type	enum EventType The type of event. `EVENT_TYPE_UNSPECIFIED`: Unspecified event type. `PARTIAL_MESSAGE`: Partial message is available. `ERROR`: Run has failed due to an error. `DONE`: The run has completed. `TOOL_CALLS`: The run is waiting for tool calls to be executed and their results to be submitted.
stream_cursor	StreamCursor The current position in the stream.
error	Error Error information if the run has failed. Includes only one of the fields `error`, `partial_message`, `completed_message`, `tool_call_list`. Event data, which can be one of several types.
partial_message	MessageContent Partially generated message. Includes only one of the fields `error`, `partial_message`, `completed_message`, `tool_call_list`. Event data, which can be one of several types.
completed_message	Message Final message generated by an assistant if a run has completed successfully. Includes only one of the fields `error`, `partial_message`, `completed_message`, `tool_call_list`. Event data, which can be one of several types.
tool_call_list	ToolCallList A list of tool calls requested by the assistant. Includes only one of the fields `error`, `partial_message`, `completed_message`, `tool_call_list`. Event data, which can be one of several types.

StreamCursor

Represents the cursor position in a stream of events.

Field

Description

current_event_idx

int64

Index of the current event in the stream.

num_user_events_received

int64

The number of user events received so far.

Error

Field	Description
code	int64
message	string

MessageContent

Represents the content of a message, which can consist of multiple parts.

Field

Description

content[]

ContentPart

A list of content parts that make up the message.

ContentPart

ContentPart represents an individual part of the message content, which can be of various types.

Field

Description

text

Text

Text content of the message part.

Includes only one of the fields text.

Specifies the type of content that the part contains.

Text

Text represents a textual content part of a message.

Field

Description

content

string

Text content of the message.

Message

Field	Description
id	string Unique identifier of the message.
thread_id	string ID of the thread that this message belongs to.
created_by	string Identifier of the subject who created this message.
created_at	google.protobuf.Timestamp Timestamp representing when the message was created.
author	Author Author of the message, containing details about the message's creator.
labels	object (map<string, string>) Set of key-value pairs that can be used to organize and categorize the message.
content	MessageContent Content of the message.
status	enum MessageStatus Status of the message. `MESSAGE_STATUS_UNSPECIFIED`: Unspecified message status. `COMPLETED`: Message was successfully created by a user or generated by an assistant. `TRUNCATED`: Message generation was truncated due to reaching the maximum allowed number of tokens. `FILTERED_CONTENT`: Message generation was stopped because potentially sensitive content was detected either in the prompt or in the generated response.
citations[]	Citation List of citations used to generate the message.

Author

Author of the message, containing details about the message's creator.

Field

Description

string

Unique identifier of the author. This could be either the user's ID or the assistant's ID, depending on the role.

role

string

Role of the author, indicating whether the message was created by a "user" or an "assistant".

Citation

Represents a citation used for generating a message.

Field

Description

sources[]

Source

List of sources for citation.

Source

Represents a source used for generating a message citation.

Field

Description

chunk

FileChunk

File chunk source.

Includes only one of the fields chunk.

Specifies the type of source type.

FileChunk

FileChunk represents a chunk of a file used as a source.

Field	Description
search_index	SearchIndex Search index associated with the file chunk.
source_file	File The original file from which the chunk is derived.
content	ChunkContent Content of the file chunk.

SearchIndex

Represents a search index used to store and query data, either using traditional keyword-based text search or vector-based search mechanisms.

Field	Description
id	string Unique identifier of the search index.
folder_id	string ID of the folder that the search index belongs to.
name	string Name of the search index.
description	string Description of the search index.
created_by	string Identifier of the subject who created this search index.
created_at	google.protobuf.Timestamp Timestamp representing when the search index was created.
updated_by	string Identifier of the subject who last updated this search index.
updated_at	google.protobuf.Timestamp Timestamp representing the last time this search index was updated.
expiration_config	ExpirationConfig Configuration for the expiration of the search index, defining when and how the search index will expire.
expires_at	google.protobuf.Timestamp Timestamp representing when the search index will expire.
labels	object (map<string, string>) Set of key-value pairs that can be used to organize and categorize the search index.
text_search_index	TextSearchIndex Keyword-based text search index configuration. This type of index is used for traditional text search, where documents are indexed based on their keywords. Includes only one of the fields `text_search_index`, `vector_search_index`, `hybrid_search_index`. Type of the search index. It can be either a traditional keyword-based text search or a vector-based search.
vector_search_index	VectorSearchIndex Vector-based search index configuration. This type is used for vector search, where documents are indexed using vector embeddings. Includes only one of the fields `text_search_index`, `vector_search_index`, `hybrid_search_index`. Type of the search index. It can be either a traditional keyword-based text search or a vector-based search.
hybrid_search_index	HybridSearchIndex Hybrid (vector-based + keyword-based) search index configuration This type is used for hybrid search, where documents are indexed using both keyword-based and vector-based search mechanisms. Includes only one of the fields `text_search_index`, `vector_search_index`, `hybrid_search_index`. Type of the search index. It can be either a traditional keyword-based text search or a vector-based search.

ExpirationConfig

Field

Description

expiration_policy

enum ExpirationPolicy

EXPIRATION_POLICY_UNSPECIFIED
STATIC
SINCE_LAST_ACTIVE

ttl_days

int64

TextSearchIndex

Defines the configuration for a traditional keyword-based text search index.

Field

Description

chunking_strategy

ChunkingStrategy

Chunking strategy used to split text into smaller chunks before indexing.
In the case of text search, tokens are individual text characters.

ngram_tokenizer

NgramTokenizer

Tokenizer that generates n-grams.

Includes only one of the fields ngram_tokenizer.

Tokenizer type used for text search. The tokenizer determines how the
input text is broken down into tokens before indexing.
If not specified, the default tokenizer configuration is applied.

ChunkingStrategy

Defines a general strategy for chunking text into smaller segments.
Currently, only StaticChunkingStrategy is supported.

Field

Description

static_strategy

StaticChunkingStrategy

Includes only one of the fields static_strategy.

StaticChunkingStrategy

Defines a chunking strategy where chunks are created with a fixed maximum chunk size and an overlap between consecutive chunks.

Field

Description

max_chunk_size_tokens

int64

The maximum number of tokens allowed in a single chunk.
Constraints: must be within the range [100, 2048].
Default value: 800

chunk_overlap_tokens

int64

The number of tokens that should overlap between consecutive chunks.
This allows for some context from the previous chunk to be included in the next chunk.
Constraints: must be less than or equal to half of max_chunk_size_tokens.
Default value: 400

NgramTokenizer

Configuration for the NgramTokenizer, which splits text into overlapping character sequences (n-grams) of specified lengths.

Example:
Input text: hello
min_gram = 2, max_gram = 3

Generated tokens:

For n = 2 (2-character n-grams): he, el, ll, lo
For n = 3 (3-character n-grams): hel, ell, llo

Final tokens: [he, el, ll, lo, hel, ell, llo]

Field

Description

min_gram

google.protobuf.Int64Value

Minimum length of characters in a gram. Defaults to 3

max_gram

google.protobuf.Int64Value

Maximum length of characters in a gram. Defaults to 4

VectorSearchIndex

Defines the configuration for a vector-based search index. This type uses embeddings to represent documents and queries.

Field	Description
doc_embedder_uri	string The ID of the model to be used for obtaining document text embeddings.
query_embedder_uri	string The ID of the model to be used for obtaining query text embeddings.
chunking_strategy	ChunkingStrategy Chunking strategy used to split text into smaller chunks before indexing. In the case of vector search, tokens are produced by the tokenizer from the embedding model.

HybridSearchIndex

Defines the configuration for a hybrid (vector-based + keyword-based) search index. This type uses both embeddings and keyword-based search to represent documents and queries.

Field	Description
text_search_index	TextSearchIndex Configuration for a traditional keyword-based text search index.
vector_search_index	VectorSearchIndex Configuration for a vector-based search index.
chunking_strategy	ChunkingStrategy Common chunking strategy that applies to both text and vector search indexes. If provided, it overrides the individual chunking strategies in both `text_search_index` and `vector_search_index`. In this case, both text and vector search will use token-based chunking, where tokens are produced by the tokenizer of the embedding model.
normalization_strategy	enum NormalizationStrategy Normalization strategy for relevance scores from different indices. Default is MIN_MAX_STRATEGY `NORMALIZATION_STRATEGY_UNSPECIFIED` `MIN_MAX`: https://en.wikipedia.org/wiki/Feature_scaling#Rescaling_(min-max_normalization) `L2`: https://en.wikipedia.org/wiki/Cosine_similarity#L2-normalized_Euclidean_distance
combination_strategy	CombinationStrategy Combination strategy for merging rankings from different indices. Default is arithmetic mean

CombinationStrategy

Combination strategy for merging rankings from different indices

Field

Description

mean_combination

MeanCombinationStrategy

Includes only one of the fields mean_combination, rrf_combination.

rrf_combination

ReciprocalRankFusionCombinationStrategy

Includes only one of the fields mean_combination, rrf_combination.

MeanCombinationStrategy

Field

Description

mean_evaluation_technique

enum MeanEvaluationTechnique

Technique for averaging relevance scores from different indices. Default is ARITHMETIC

MEAN_EVALUATION_TECHNIQUE_UNSPECIFIED
ARITHMETIC: https://en.wikipedia.org/wiki/Arithmetic_mean
GEOMETRIC: https://en.wikipedia.org/wiki/Geometric_mean
HARMONIC: https://en.wikipedia.org/wiki/Harmonic_mean

weights[]

double

Weights used for evaluating the weighted mean of relevance scores. The sum of the values must equal 1.0
If not provided, all scores are given equal weight

ReciprocalRankFusionCombinationStrategy

https://plg.uwaterloo.ca/~gvcormac/cormacksigir09-rrf.pdf

Field

Description

google.protobuf.Int64Value

The parameter k for RRFscore. Default is 60

File

Field	Description
id	string Unique identifier of the file.
folder_id	string ID of the folder that the file belongs to.
name	string Name of the file.
description	string Description of the file.
mime_type	string MIME type of the file, indicating the file's format (e.g., "application/pdf")
created_by	string Identifier of the subject who created this file.
created_at	google.protobuf.Timestamp Timestamp representing when the file was created.
updated_by	string Identifier of the subject who last updated this file.
updated_at	google.protobuf.Timestamp Timestamp representing the last time this file was updated.
expiration_config	ExpirationConfig Configuration for the expiration of the file, defining when and how the file will expire.
expires_at	google.protobuf.Timestamp Timestamp representing when the file will expire.
labels	object (map<string, string>) Set of key-value pairs that can be used to organize and categorize the file.

ChunkContent

Represents the content of a file chunk.

Field

Description

content[]

ContentPart

A list of content parts that make up the chunk.

ToolCallList

Represents a list of tool calls.

Field

Description

tool_calls[]

ToolCall

A list of tool calls to be executed.

ToolCall

Represents a call to a tool.

Field

Description

function_call

FunctionCall

Represents a call to a function.

Includes only one of the fields function_call.

FunctionCall

Represents the invocation of a function with specific arguments.

Field

Description

name

string

The name of the function being called.

arguments

google.protobuf.Struct

The structured arguments passed to the function.
These arguments must adhere to the JSON Schema defined in the corresponding function's parameters.

Runs, gRPC: RunService.Attach

gRPC requestgRPC request

AttachRunRequestAttachRunRequest

ToolResultListToolResultList

ToolResultToolResult

FunctionResultFunctionResult

StreamEventStreamEvent

StreamCursorStreamCursor

ErrorError

MessageContentMessageContent

ContentPartContentPart

TextText

MessageMessage

AuthorAuthor

CitationCitation

SourceSource

FileChunkFileChunk

SearchIndexSearchIndex

ExpirationConfigExpirationConfig

TextSearchIndexTextSearchIndex

ChunkingStrategyChunkingStrategy

StaticChunkingStrategyStaticChunkingStrategy

NgramTokenizerNgramTokenizer

VectorSearchIndexVectorSearchIndex

HybridSearchIndexHybridSearchIndex

CombinationStrategyCombinationStrategy

MeanCombinationStrategyMeanCombinationStrategy

ReciprocalRankFusionCombinationStrategyReciprocalRankFusionCombinationStrategy

FileFile

ChunkContentChunkContent

ToolCallListToolCallList

ToolCallToolCall

FunctionCallFunctionCall

Была ли статья полезна?

gRPC request

AttachRunRequest

ToolResultList

ToolResult

FunctionResult

StreamEvent

StreamCursor

Error

MessageContent

ContentPart

Text

Message

Author

Citation

Source

FileChunk

SearchIndex

ExpirationConfig

TextSearchIndex

ChunkingStrategy

StaticChunkingStrategy

NgramTokenizer

VectorSearchIndex

HybridSearchIndex

CombinationStrategy

MeanCombinationStrategy

ReciprocalRankFusionCombinationStrategy

File

ChunkContent

ToolCallList

ToolCall

FunctionCall