Package cloud.google.com/go/vertexai/genai (v0.13.1)

Package genai is a client for the generative VertexAI model.

Functions

func Ptr

func Ptr(t T) *T

Ptr returns a pointer to its argument. It can be used to initialize pointer fields:

model.Temperature = genai.Ptr[float32](0.1)

func WithClientInfo

func WithClientInfo(key, value string) option.ClientOption

WithClientInfo is an option that sets request information identifying the product that is calling this client.

func WithREST

func WithREST() option.ClientOption

WithREST is an option that enables REST transport for the client. The default transport (if this option isn't provided) is gRPC.

Blob

type Blob struct {
	// Required. The IANA standard MIME type of the source data.
	MIMEType string
	// Required. Raw bytes.
	Data []byte
}

Blob contains binary data like images. Use [Text] for text.

func ImageData

func ImageData(format string, data []byte) Blob

ImageData is a convenience function for creating an image Blob for input to a model. The format should be the second part of the MIME type, after "image/". For example, for a PNG image, pass "png".

BlockedError

type BlockedError struct {
	// If non-nil, the model's response was blocked.
	// Consult the Candidate and SafetyRatings fields for details.
	Candidate *Candidate

	// If non-nil, there was a problem with the prompt.
	PromptFeedback *PromptFeedback
}

A BlockedError indicates that the model's response was blocked. There can be two underlying causes: the prompt or a candidate response.

func (*BlockedError) Error

func (e *BlockedError) Error() string

BlockedReason

type BlockedReason int32

BlockedReason is blocked reason enumeration.

BlockedReasonUnspecified, BlockedReasonSafety, BlockedReasonOther, BlockedReasonBlocklist, BlockedReasonProhibitedContent

const (
	// BlockedReasonUnspecified means unspecified blocked reason.
	BlockedReasonUnspecified BlockedReason = 0
	// BlockedReasonSafety means candidates blocked due to safety.
	BlockedReasonSafety BlockedReason = 1
	// BlockedReasonOther means candidates blocked due to other reason.
	BlockedReasonOther BlockedReason = 2
	// BlockedReasonBlocklist means candidates blocked due to the terms which are included from the
	// terminology blocklist.
	BlockedReasonBlocklist BlockedReason = 3
	// BlockedReasonProhibitedContent means candidates blocked due to prohibited content.
	BlockedReasonProhibitedContent BlockedReason = 4
)

func (BlockedReason) String

func (v BlockedReason) String() string

CachedContent

type CachedContent struct {
	// Expiration time of the cached content.
	//
	// Types that are assignable to Expiration:
	//
	//	*CachedContent_ExpireTime
	//	*CachedContent_Ttl
	Expiration ExpireTimeOrTTL
	// Immutable. Identifier. The server-generated resource name of the cached
	// content Format:
	// projects/{project}/locations/{location}/cachedContents/{cached_content}
	Name string
	// Immutable. The name of the publisher model to use for cached content.
	// Format:
	// projects/{project}/locations/{location}/publishers/{publisher}/models/{model}
	Model string
	// Optional. Input only. Immutable. Developer set system instruction.
	// Currently, text only
	SystemInstruction *Content
	// Optional. Input only. Immutable. The content to cache
	Contents []*Content
	// Optional. Input only. Immutable. A list of `Tools` the model may use to
	// generate the next response
	Tools []*Tool
	// Optional. Input only. Immutable. Tool config. This config is shared for all
	// tools
	ToolConfig *ToolConfig
	// Output only. Creatation time of the cache entry.
	CreateTime time.Time
	// Output only. When the cache entry was last updated in UTC time.
	UpdateTime time.Time
}

CachedContent is a resource used in LLM queries for users to explicitly specify what to cache and how to cache.

CachedContentIterator

type CachedContentIterator struct {
	// contains filtered or unexported fields
}

A CachedContentIterator iterates over CachedContents.

func (*CachedContentIterator) Next

func (it *CachedContentIterator) Next() (*CachedContent, error)

Next returns the next result. Its second return value is iterator.Done if there are no more results. Once Next returns Done, all subsequent calls will return Done.

func (*CachedContentIterator) PageInfo

func (it *CachedContentIterator) PageInfo() *iterator.PageInfo

PageInfo supports pagination. See the google.golang.org/api/iterator package for details.

CachedContentToUpdate

type CachedContentToUpdate struct {
	// If non-nil, update the expire time or TTL.
	Expiration *ExpireTimeOrTTL
}

CachedContentToUpdate specifies which fields of a CachedContent to modify in a call to [Client.UpdateCachedContent].

Candidate

type Candidate struct {
	// Output only. Index of the candidate.
	Index int32
	// Output only. Content parts of the candidate.
	Content *Content
	// Output only. The reason why the model stopped generating tokens.
	// If empty, the model has not stopped generating the tokens.
	FinishReason FinishReason
	// Output only. List of ratings for the safety of a response candidate.
	//
	// There is at most one rating per category.
	SafetyRatings []*SafetyRating
	// Output only. Describes the reason the mode stopped generating tokens in
	// more detail. This is only filled when `finish_reason` is set.
	FinishMessage string
	// Output only. Source attribution of the generated content.
	CitationMetadata *CitationMetadata
}

Candidate is a response candidate generated from the model.

func (*Candidate) FunctionCalls

func (c *Candidate) FunctionCalls() []FunctionCall

FunctionCalls return all the FunctionCall parts in the candidate.

ChatSession

type ChatSession struct {
	History []*Content
	// contains filtered or unexported fields
}

A ChatSession provides interactive chat.

Example

package main

import (
	"context"
	"fmt"
	"log"

	"cloud.google.com/go/vertexai/genai"

	"google.golang.org/api/iterator"
)

// Your GCP project
const projectID = "your-project"

// A GCP location like "us-central1"; if you're using standard Google-published
// models (like untuned Gemini models), you can keep location blank ("").
const location = "some-gcp-location"

// A model name like "gemini-1.0-pro"
// For custom models from different publishers, prepent the full publisher
// prefix for the model, e.g.:
//
//	modelName = publishers/some-publisher/models/some-model-name
const modelName = "some-model"

func main() {
	ctx := context.Background()
	client, err := genai.NewClient(ctx, projectID, location)
	if err != nil {
		log.Fatal(err)
	}
	defer client.Close()
	model := client.GenerativeModel(modelName)
	cs := model.StartChat()

	send := func(msg string) *genai.GenerateContentResponse {
		fmt.Printf("== Me: %s\n== Model:\n", msg)
		res, err := cs.SendMessage(ctx, genai.Text(msg))
		if err != nil {
			log.Fatal(err)
		}
		return res
	}

	res := send("Can you name some brands of air fryer?")
	printResponse(res)
	iter := cs.SendMessageStream(ctx, genai.Text("Which one of those do you recommend?"))
	for {
		res, err := iter.Next()
		if err == iterator.Done {
			break
		}
		if err != nil {
			log.Fatal(err)
		}
		printResponse(res)
	}

	for i, c := range cs.History {
		log.Printf("    %d: %+v", i, c)
	}
	res = send("Why do you like the Philips?")
	if err != nil {
		log.Fatal(err)
	}
	printResponse(res)
}

func printResponse(resp *genai.GenerateContentResponse) {
	for _, cand := range resp.Candidates {
		for _, part := range cand.Content.Parts {
			fmt.Println(part)
		}
	}
	fmt.Println("---")
}

func (*ChatSession) SendMessage

func (cs *ChatSession) SendMessage(ctx context.Context, parts ...Part) (*GenerateContentResponse, error)

SendMessage sends a request to the model as part of a chat session.

func (*ChatSession) SendMessageStream

func (cs *ChatSession) SendMessageStream(ctx context.Context, parts ...Part) *GenerateContentResponseIterator

SendMessageStream is like SendMessage, but with a streaming request.

Citation

type Citation struct {
	// Output only. Start index into the content.
	StartIndex int32
	// Output only. End index into the content.
	EndIndex int32
	// Output only. Url reference of the attribution.
	URI string
	// Output only. Title of the attribution.
	Title string
	// Output only. License of the attribution.
	License string
	// Output only. Publication date of the attribution.
	PublicationDate civil.Date
}

Citation contains source attributions for content.

CitationMetadata

type CitationMetadata struct {
	// Output only. List of citations.
	Citations []*Citation
}

CitationMetadata is a collection of source attributions for a piece of content.

Client

type Client struct {
	// contains filtered or unexported fields
}

A Client is a Google Vertex AI client.

Example

Example

cachedContent

cachedContent

package main

import (
	"context"
	"log"
	"os"

	"cloud.google.com/go/vertexai/genai"
)

// Your GCP project
const projectID = "your-project"

// A GCP location like "us-central1"; if you're using standard Google-published
// models (like untuned Gemini models), you can keep location blank ("").
const location = "some-gcp-location"

// A model name like "gemini-1.0-pro"
// For custom models from different publishers, prepent the full publisher
// prefix for the model, e.g.:
//
//	modelName = publishers/some-publisher/models/some-model-name
const modelName = "some-model"

func main() {
	ctx := context.Background()
	client, err := genai.NewClient(ctx, projectID, location)
	if err != nil {
		log.Fatal(err)
	}
	defer client.Close()
	file := genai.FileData{MIMEType: "application/pdf", FileURI: "gs://my-bucket/my-doc.pdf"}
	cc, err := client.CreateCachedContent(ctx, &genai.CachedContent{
		Model:    modelName,
		Contents: []*genai.Content{genai.NewUserContent(file)},
	})
	model := client.GenerativeModelFromCachedContent(cc)
	// Work with the model as usual in this program.
	_ = model

	// Store the CachedContent name for later use.
	if err := os.WriteFile("my-cached-content-name", []byte(cc.Name), 0o644); err != nil {
		log.Fatal(err)
	}

	///////////////////////////////
	// Later, in another process...

	bytes, err := os.ReadFile("my-cached-content-name")
	if err != nil {
		log.Fatal(err)
	}
	ccName := string(bytes)

	// No need to call [Client.GetCachedContent]; the name is sufficient.
	model = client.GenerativeModel(modelName)
	model.CachedContentName = ccName
	// Proceed as usual.
}

func NewClient

func NewClient(ctx context.Context, projectID, location string, opts ...option.ClientOption) (*Client, error)

NewClient creates a new Google Vertex AI client.

Clients should be reused instead of created as needed. The methods of Client are safe for concurrent use by multiple goroutines. projectID is your GCP project; location is GCP region/location per https://cloud.google.com/vertex-ai/docs/general/locations If location is empty, this function attempts to infer it from environment variables and falls back to a default location if unsuccessful.

You may configure the client by passing in options from the [google.golang.org/api/option] package. You may also use options defined in this package, such as [WithREST].

func (*Client) Close

func (c *Client) Close() error

Close closes the client.

func (*Client) CreateCachedContent

func (c *Client) CreateCachedContent(ctx context.Context, cc *CachedContent) (*CachedContent, error)

CreateCachedContent creates a new CachedContent. The argument should contain a model name and some data to be cached, which can include contents, a system instruction, tools and/or tool configuration. It can also include an expiration time or TTL. But it should not include a name; the system will generate one.

The return value will contain the name, which should be used to refer to the CachedContent in other API calls. It will also hold various metadata like expiration and creation time. It will not contain any of the actual content provided as input.

You can use the return value to create a model with [Client.GenerativeModelFromCachedContent]. Or you can set [GenerativeModel.CachedContentName] to the name of the CachedContent, in which case you must ensure that the model provided in this call matches the name in the [GenerativeModel].

func (*Client) DeleteCachedContent

func (c *Client) DeleteCachedContent(ctx context.Context, name string) error

DeleteCachedContent deletes the CachedContent with the given name.

func (*Client) GenerativeModel

func (c *Client) GenerativeModel(name string) *GenerativeModel

GenerativeModel creates a new instance of the named model. name is a string model name like "gemini-1.0-pro" or "models/gemini-1.0-pro" for Google-published models. See https://cloud.google.com/vertex-ai/generative-ai/docs/learn/model-versioning for details on model naming and versioning, and https://cloud.google.com/vertex-ai/generative-ai/docs/model-garden/explore-models for providing model garden names. The SDK isn't familiar with custom model garden models, and will pass your model name to the backend API server.

func (*Client) GenerativeModelFromCachedContent

func (c *Client) GenerativeModelFromCachedContent(cc *CachedContent) *GenerativeModel

GenerativeModelFromCachedContent returns a [GenerativeModel] that uses the given [CachedContent]. The argument should come from a call to [Client.CreateCachedContent] or [Client.GetCachedContent].

func (*Client) GetCachedContent

func (c *Client) GetCachedContent(ctx context.Context, name string) (*CachedContent, error)

GetCachedContent retrieves the CachedContent with the given name.

func (*Client) ListCachedContents

func (c *Client) ListCachedContents(ctx context.Context) *CachedContentIterator

ListCachedContents lists all the CachedContents associated with the project and location.

func (*Client) UpdateCachedContent

func (c *Client) UpdateCachedContent(ctx context.Context, cc *CachedContent, ccu *CachedContentToUpdate) (*CachedContent, error)

UpdateCachedContent modifies the [CachedContent] according to the values of the [CachedContentToUpdate] struct. It returns the modified CachedContent.

The argument CachedContent must have its Name field populated. If its UpdateTime field is non-zero, it will be compared with the update time of the stored CachedContent and the call will fail if they differ. This avoids a race condition when two updates are attempted concurrently. All other fields of the argument CachedContent are ignored.

Content

type Content struct {
	// Optional. The producer of the content. Must be either 'user' or 'model'.
	//
	// Useful to set for multi-turn conversations, otherwise can be left blank
	// or unset.
	Role string
	// Required. Ordered `Parts` that constitute a single message. Parts may have
	// different IANA MIME types.
	Parts []Part
}

Content is the base structured datatype containing multi-part content of a message.

A Content includes a role field designating the producer of the Content and a parts field containing multi-part data that contains the content of the message turn.

func NewUserContent

func NewUserContent(parts ...Part) *Content

NewUserContent returns a [Content] with a "user" role set and one or more parts.

CountTokensResponse

type CountTokensResponse struct {
	// The total number of tokens counted across all instances from the request.
	TotalTokens int32
	// The total number of billable characters counted across all instances from
	// the request.
	TotalBillableCharacters int32
}

CountTokensResponse is response message for [PredictionService.CountTokens][google.cloud.aiplatform.v1beta1.PredictionService.CountTokens].

ExpireTimeOrTTL

type ExpireTimeOrTTL struct {
	ExpireTime time.Time
	TTL        time.Duration
}

ExpireTimeOrTTL describes the time when a resource expires. If ExpireTime is non-zero, it is the expiration time. Otherwise, the expiration time is the value of TTL ("time to live") added to the current time.

FileData

type FileData struct {
	// Required. The IANA standard MIME type of the source data.
	MIMEType string
	// Required. URI.
	FileURI string
}

FileData is URI based data.

FinishReason

type FinishReason int32

FinishReason is the reason why the model stopped generating tokens. If empty, the model has not stopped generating the tokens.

FinishReasonUnspecified, FinishReasonStop, FinishReasonMaxTokens, FinishReasonSafety, FinishReasonRecitation, FinishReasonOther, FinishReasonBlocklist, FinishReasonProhibitedContent, FinishReasonSpii, FinishReasonMalformedFunctionCall

const (
	// FinishReasonUnspecified means the finish reason is unspecified.
	FinishReasonUnspecified FinishReason = 0
	// FinishReasonStop means natural stop point of the model or provided stop sequence.
	FinishReasonStop FinishReason = 1
	// FinishReasonMaxTokens means the maximum number of tokens as specified in the request was reached.
	FinishReasonMaxTokens FinishReason = 2
	// FinishReasonSafety means the token generation was stopped as the response was flagged for safety
	// reasons. NOTE: When streaming the Candidate.content will be empty if
	// content filters blocked the output.
	FinishReasonSafety FinishReason = 3
	// FinishReasonRecitation means the token generation was stopped as the response was flagged for
	// unauthorized citations.
	FinishReasonRecitation FinishReason = 4
	// FinishReasonOther means all other reasons that stopped the token generation
	FinishReasonOther FinishReason = 5
	// FinishReasonBlocklist means the token generation was stopped as the response was flagged for the
	// terms which are included from the terminology blocklist.
	FinishReasonBlocklist FinishReason = 6
	// FinishReasonProhibitedContent means the token generation was stopped as the response was flagged for
	// the prohibited contents.
	FinishReasonProhibitedContent FinishReason = 7
	// FinishReasonSpii means the token generation was stopped as the response was flagged for
	// Sensitive Personally Identifiable Information (SPII) contents.
	FinishReasonSpii FinishReason = 8
	// FinishReasonMalformedFunctionCall means the function call generated by the model is invalid.
	FinishReasonMalformedFunctionCall FinishReason = 9
)

func (FinishReason) String

func (v FinishReason) String() string

FunctionCall

type FunctionCall struct {
	// Required. The name of the function to call.
	// Matches [FunctionDeclaration.name].
	Name string
	// Optional. Required. The function parameters and values in JSON object
	// format. See [FunctionDeclaration.parameters] for parameter details.
	Args map[string]any
}

FunctionCall is a predicted [FunctionCall] returned from the model that contains a string representing the [FunctionDeclaration.name] and a structured JSON object containing the parameters and their values.

FunctionCallingConfig

type FunctionCallingConfig struct {
	// Optional. Function calling mode.
	Mode FunctionCallingMode
	// Optional. Function names to call. Only set when the Mode is ANY. Function
	// names should match [FunctionDeclaration.name]. With mode set to ANY, model
	// will predict a function call from the set of function names provided.
	AllowedFunctionNames []string
}

FunctionCallingConfig holds configuration for function calling.

FunctionCallingMode

type FunctionCallingMode int32

FunctionCallingMode is function calling mode.

FunctionCallingUnspecified, FunctionCallingAuto, FunctionCallingAny, FunctionCallingNone

const (
	// FunctionCallingUnspecified means unspecified function calling mode. This value should not be used.
	FunctionCallingUnspecified FunctionCallingMode = 0
	// FunctionCallingAuto means default model behavior, model decides to predict either a function call
	// or a natural language repspose.
	FunctionCallingAuto FunctionCallingMode = 1
	// FunctionCallingAny means model is constrained to always predicting a function call only.
	// If "allowed_function_names" are set, the predicted function call will be
	// limited to any one of "allowed_function_names", else the predicted
	// function call will be any one of the provided "function_declarations".
	FunctionCallingAny FunctionCallingMode = 2
	// FunctionCallingNone means model will not predict any function call. Model behavior is same as when
	// not passing any function declarations.
	FunctionCallingNone FunctionCallingMode = 3
)

func (FunctionCallingMode) String

func (v FunctionCallingMode) String() string

FunctionDeclaration

type FunctionDeclaration struct {
	// Required. The name of the function to call.
	// Must start with a letter or an underscore.
	// Must be a-z, A-Z, 0-9, or contain underscores, dots and dashes, with a
	// maximum length of 64.
	Name string
	// Optional. Description and purpose of the function.
	// Model uses it to decide how and whether to call the function.
	Description string
	// Optional. Describes the parameters to this function in JSON Schema Object
	// format. Reflects the Open API 3.03 Parameter Object. string Key: the name
	// of the parameter. Parameter names are case sensitive. Schema Value: the
	// Schema defining the type used for the parameter. For function with no
	// parameters, this can be left unset. Parameter names must start with a
	// letter or an underscore and must only contain chars a-z, A-Z, 0-9, or
	// underscores with a maximum length of 64. Example with 1 required and 1
	// optional parameter: type: OBJECT properties:
	//
	//	param1:
	//	  type: STRING
	//	param2:
	//	  type: INTEGER
	//
	// required:
	//   - param1
	Parameters *Schema
	// Optional. Describes the output from this function in JSON Schema format.
	// Reflects the Open API 3.03 Response Object. The Schema defines the type
	// used for the response value of the function.
	Response *Schema
}

FunctionDeclaration is structured representation of a function declaration as defined by the OpenAPI 3.0 specification. Included in this declaration are the function name and parameters. This FunctionDeclaration is a representation of a block of code that can be used as a Tool by the model and executed by the client.

FunctionResponse

type FunctionResponse struct {
	// Required. The name of the function to call.
	// Matches [FunctionDeclaration.name] and [FunctionCall.name].
	Name string
	// Required. The function response in JSON object format.
	Response map[string]any
}

FunctionResponse is the result output from a [FunctionCall] that contains a string representing the [FunctionDeclaration.name] and a structured JSON object containing any output from the function is used as context to the model. This should contain the result of a [FunctionCall] made based on model prediction.

GenerateContentResponse

type GenerateContentResponse struct {
	// Output only. Generated candidates.
	Candidates []*Candidate
	// Output only. Content filter results for a prompt sent in the request.
	// Note: Sent only in the first stream chunk.
	// Only happens when no candidates were generated due to content violations.
	PromptFeedback *PromptFeedback
	// Usage metadata about the response(s).
	UsageMetadata *UsageMetadata
}

GenerateContentResponse is the response from a GenerateContent or GenerateContentStream call.

GenerateContentResponseIterator

type GenerateContentResponseIterator struct {
	// contains filtered or unexported fields
}

GenerateContentResponseIterator is an iterator over GnerateContentResponse.

func (*GenerateContentResponseIterator) MergedResponse

MergedResponse returns the result of combining all the streamed responses seen so far. After iteration completes, the merged response should match the response obtained without streaming (that is, if [GenerativeModel.GenerateContent] were called).

func (*GenerateContentResponseIterator) Next

Next returns the next response.

GenerationConfig

type GenerationConfig struct {
	// Optional. Controls the randomness of predictions.
	Temperature *float32
	// Optional. If specified, nucleus sampling will be used.
	TopP *float32
	// Optional. If specified, top-k sampling will be used.
	TopK *int32
	// Optional. Number of candidates to generate.
	CandidateCount *int32
	// Optional. The maximum number of output tokens to generate per message.
	MaxOutputTokens *int32
	// Optional. Stop sequences.
	StopSequences []string
	// Optional. Positive penalties.
	PresencePenalty *float32
	// Optional. Frequency penalties.
	FrequencyPenalty *float32
	// Optional. Output response mimetype of the generated candidate text.
	// Supported mimetype:
	// - `text/plain`: (default) Text output.
	// - `application/json`: JSON response in the candidates.
	// The model needs to be prompted to output the appropriate response type,
	// otherwise the behavior is undefined.
	// This is a preview feature.
	ResponseMIMEType string
	// Optional. The `Schema` object allows the definition of input and output
	// data types. These types can be objects, but also primitives and arrays.
	// Represents a select subset of an [OpenAPI 3.0 schema
	// object](https://spec.openapis.org/oas/v3.0.3#schema).
	// If set, a compatible response_mime_type must also be set.
	// Compatible mimetypes:
	// `application/json`: Schema for JSON response.
	ResponseSchema *Schema
}

GenerationConfig is generation config.

func (*GenerationConfig) SetCandidateCount

func (c *GenerationConfig) SetCandidateCount(x int32)

SetCandidateCount sets the CandidateCount field.

func (*GenerationConfig) SetMaxOutputTokens

func (c *GenerationConfig) SetMaxOutputTokens(x int32)

SetMaxOutputTokens sets the MaxOutputTokens field.

func (*GenerationConfig) SetTemperature

func (c *GenerationConfig) SetTemperature(x float32)

SetTemperature sets the Temperature field.

func (*GenerationConfig) SetTopK

func (c *GenerationConfig) SetTopK(x int32)

SetTopK sets the TopK field.

func (*GenerationConfig) SetTopP

func (c *GenerationConfig) SetTopP(x float32)

SetTopP sets the TopP field.

GenerativeModel

type GenerativeModel struct {
	GenerationConfig
	SafetySettings    []*SafetySetting
	Tools             []*Tool
	ToolConfig        *ToolConfig // configuration for tools
	SystemInstruction *Content
	// The name of the CachedContent to use.
	// Must have already been created with [Client.CreateCachedContent].
	CachedContentName string
	// contains filtered or unexported fields
}

GenerativeModel is a model that can generate text. Create one with [Client.GenerativeModel], then configure it by setting the exported fields.

The model holds all the config for a GenerateContentRequest, so the GenerateContent method can use a vararg for the content.

func (*GenerativeModel) CountTokens

func (m *GenerativeModel) CountTokens(ctx context.Context, parts ...Part) (*CountTokensResponse, error)

CountTokens counts the number of tokens in the content.

Example

package main

import (
	"context"
	"fmt"
	"log"

	"cloud.google.com/go/vertexai/genai"
)

// Your GCP project
const projectID = "your-project"

// A GCP location like "us-central1"; if you're using standard Google-published
// models (like untuned Gemini models), you can keep location blank ("").
const location = "some-gcp-location"

// A model name like "gemini-1.0-pro"
// For custom models from different publishers, prepent the full publisher
// prefix for the model, e.g.:
//
//	modelName = publishers/some-publisher/models/some-model-name
const modelName = "some-model"

func main() {
	ctx := context.Background()
	client, err := genai.NewClient(ctx, projectID, location)
	if err != nil {
		log.Fatal(err)
	}
	defer client.Close()

	model := client.GenerativeModel(modelName)

	resp, err := model.CountTokens(ctx, genai.Text("What kind of fish is this?"))
	if err != nil {
		log.Fatal(err)
	}

	fmt.Println("Num tokens:", resp.TotalTokens)
}

func (*GenerativeModel) GenerateContent

func (m *GenerativeModel) GenerateContent(ctx context.Context, parts ...Part) (*GenerateContentResponse, error)

GenerateContent produces a single request and response.

Examples

package main

import (
	"context"
	"fmt"
	"log"

	"cloud.google.com/go/vertexai/genai"
)

// Your GCP project
const projectID = "your-project"

// A GCP location like "us-central1"; if you're using standard Google-published
// models (like untuned Gemini models), you can keep location blank ("").
const location = "some-gcp-location"

// A model name like "gemini-1.0-pro"
// For custom models from different publishers, prepent the full publisher
// prefix for the model, e.g.:
//
//	modelName = publishers/some-publisher/models/some-model-name
const modelName = "some-model"

func main() {
	ctx := context.Background()
	client, err := genai.NewClient(ctx, projectID, location)
	if err != nil {
		log.Fatal(err)
	}
	defer client.Close()

	model := client.GenerativeModel(modelName)
	model.SetTemperature(0.9)
	resp, err := model.GenerateContent(ctx, genai.Text("What is the average size of a swallow?"))
	if err != nil {
		log.Fatal(err)
	}

	printResponse(resp)
}

func printResponse(resp *genai.GenerateContentResponse) {
	for _, cand := range resp.Candidates {
		for _, part := range cand.Content.Parts {
			fmt.Println(part)
		}
	}
	fmt.Println("---")
}
config
package main

import (
	"context"
	"fmt"
	"log"

	"cloud.google.com/go/vertexai/genai"
)

func main() {
	ctx := context.Background()
	const projectID = "YOUR PROJECT ID"
	const location = "GCP LOCATION"
	client, err := genai.NewClient(ctx, projectID, location)
	if err != nil {
		log.Fatal(err)
	}
	defer client.Close()

	model := client.GenerativeModel("gemini-1.0-pro")
	model.SetTemperature(0.9)
	model.SetTopP(0.5)
	model.SetTopK(20)
	model.SetMaxOutputTokens(100)
	model.SystemInstruction = genai.NewUserContent(genai.Text("You are Yoda from Star Wars."))
	resp, err := model.GenerateContent(ctx, genai.Text("What is the average size of a swallow?"))
	if err != nil {
		log.Fatal(err)
	}
	printResponse(resp)
}

func printResponse(resp *genai.GenerateContentResponse) {
	for _, cand := range resp.Candidates {
		for _, part := range cand.Content.Parts {
			fmt.Println(part)
		}
	}
	fmt.Println("---")
}

func (*GenerativeModel) GenerateContentStream

func (m *GenerativeModel) GenerateContentStream(ctx context.Context, parts ...Part) *GenerateContentResponseIterator

GenerateContentStream returns an iterator that enumerates responses.

Example

package main

import (
	"context"
	"fmt"
	"log"

	"cloud.google.com/go/vertexai/genai"

	"google.golang.org/api/iterator"
)

// Your GCP project
const projectID = "your-project"

// A GCP location like "us-central1"; if you're using standard Google-published
// models (like untuned Gemini models), you can keep location blank ("").
const location = "some-gcp-location"

// A model name like "gemini-1.0-pro"
// For custom models from different publishers, prepent the full publisher
// prefix for the model, e.g.:
//
//	modelName = publishers/some-publisher/models/some-model-name
const modelName = "some-model"

func main() {
	ctx := context.Background()
	client, err := genai.NewClient(ctx, projectID, location)
	if err != nil {
		log.Fatal(err)
	}
	defer client.Close()

	model := client.GenerativeModel(modelName)

	iter := model.GenerateContentStream(ctx, genai.Text("Tell me a story about a lumberjack and his giant ox. Keep it very short."))
	for {
		resp, err := iter.Next()
		if err == iterator.Done {
			break
		}
		if err != nil {
			log.Fatal(err)
		}
		printResponse(resp)
	}
}

func printResponse(resp *genai.GenerateContentResponse) {
	for _, cand := range resp.Candidates {
		for _, part := range cand.Content.Parts {
			fmt.Println(part)
		}
	}
	fmt.Println("---")
}

func (*GenerativeModel) Name

func (m *GenerativeModel) Name() string

Name returns the name of the model.

func (*GenerativeModel) StartChat

func (m *GenerativeModel) StartChat() *ChatSession

StartChat starts a chat session.

HarmBlockMethod

type HarmBlockMethod int32

HarmBlockMethod determines how harm blocking is done.

HarmBlockMethodUnspecified, HarmBlockMethodSeverity, HarmBlockMethodProbability

const (
	// HarmBlockMethodUnspecified means the harm block method is unspecified.
	HarmBlockMethodUnspecified HarmBlockMethod = 0
	// HarmBlockMethodSeverity means the harm block method uses both probability and severity scores.
	HarmBlockMethodSeverity HarmBlockMethod = 1
	// HarmBlockMethodProbability means the harm block method uses the probability score.
	HarmBlockMethodProbability HarmBlockMethod = 2
)

func (HarmBlockMethod) String

func (v HarmBlockMethod) String() string

HarmBlockThreshold

type HarmBlockThreshold int32

HarmBlockThreshold specifies probability based thresholds levels for blocking.

HarmBlockUnspecified, HarmBlockLowAndAbove, HarmBlockMediumAndAbove, HarmBlockOnlyHigh, HarmBlockNone

const (
	// HarmBlockUnspecified means unspecified harm block threshold.
	HarmBlockUnspecified HarmBlockThreshold = 0
	// HarmBlockLowAndAbove means block low threshold and above (i.e. block more).
	HarmBlockLowAndAbove HarmBlockThreshold = 1
	// HarmBlockMediumAndAbove means block medium threshold and above.
	HarmBlockMediumAndAbove HarmBlockThreshold = 2
	// HarmBlockOnlyHigh means block only high threshold (i.e. block less).
	HarmBlockOnlyHigh HarmBlockThreshold = 3
	// HarmBlockNone means block none.
	HarmBlockNone HarmBlockThreshold = 4
)

func (HarmBlockThreshold) String

func (v HarmBlockThreshold) String() string

HarmCategory

type HarmCategory int32

HarmCategory specifies harm categories that will block the content.

HarmCategoryUnspecified, HarmCategoryHateSpeech, HarmCategoryDangerousContent, HarmCategoryHarassment, HarmCategorySexuallyExplicit

const (
	// HarmCategoryUnspecified means the harm category is unspecified.
	HarmCategoryUnspecified HarmCategory = 0
	// HarmCategoryHateSpeech means the harm category is hate speech.
	HarmCategoryHateSpeech HarmCategory = 1
	// HarmCategoryDangerousContent means the harm category is dangerous content.
	HarmCategoryDangerousContent HarmCategory = 2
	// HarmCategoryHarassment means the harm category is harassment.
	HarmCategoryHarassment HarmCategory = 3
	// HarmCategorySexuallyExplicit means the harm category is sexually explicit content.
	HarmCategorySexuallyExplicit HarmCategory = 4
)

func (HarmCategory) String

func (v HarmCategory) String() string

HarmProbability

type HarmProbability int32

HarmProbability specifies harm probability levels in the content.

HarmProbabilityUnspecified, HarmProbabilityNegligible, HarmProbabilityLow, HarmProbabilityMedium, HarmProbabilityHigh

const (
	// HarmProbabilityUnspecified means harm probability unspecified.
	HarmProbabilityUnspecified HarmProbability = 0
	// HarmProbabilityNegligible means negligible level of harm.
	HarmProbabilityNegligible HarmProbability = 1
	// HarmProbabilityLow means low level of harm.
	HarmProbabilityLow HarmProbability = 2
	// HarmProbabilityMedium means medium level of harm.
	HarmProbabilityMedium HarmProbability = 3
	// HarmProbabilityHigh means high level of harm.
	HarmProbabilityHigh HarmProbability = 4
)

func (HarmProbability) String

func (v HarmProbability) String() string

HarmSeverity

type HarmSeverity int32

HarmSeverity specifies harm severity levels.

HarmSeverityUnspecified, HarmSeverityNegligible, HarmSeverityLow, HarmSeverityMedium, HarmSeverityHigh

const (
	// HarmSeverityUnspecified means harm severity unspecified.
	HarmSeverityUnspecified HarmSeverity = 0
	// HarmSeverityNegligible means negligible level of harm severity.
	HarmSeverityNegligible HarmSeverity = 1
	// HarmSeverityLow means low level of harm severity.
	HarmSeverityLow HarmSeverity = 2
	// HarmSeverityMedium means medium level of harm severity.
	HarmSeverityMedium HarmSeverity = 3
	// HarmSeverityHigh means high level of harm severity.
	HarmSeverityHigh HarmSeverity = 4
)

func (HarmSeverity) String

func (v HarmSeverity) String() string

Part

type Part interface {
	// contains filtered or unexported methods
}

A Part is either a Text, a Blob, or a FileData.

PromptFeedback

type PromptFeedback struct {
	// Output only. Blocked reason.
	BlockReason BlockedReason
	// Output only. Safety ratings.
	SafetyRatings []*SafetyRating
	// Output only. A readable block reason message.
	BlockReasonMessage string
}

PromptFeedback contains content filter results for a prompt sent in the request.

SafetyRating

type SafetyRating struct {
	// Output only. Harm category.
	Category HarmCategory
	// Output only. Harm probability levels in the content.
	Probability HarmProbability
	// Output only. Harm probability score.
	ProbabilityScore float32
	// Output only. Harm severity levels in the content.
	Severity HarmSeverity
	// Output only. Harm severity score.
	SeverityScore float32
	// Output only. Indicates whether the content was filtered out because of this
	// rating.
	Blocked bool
}

SafetyRating is the safety rating corresponding to the generated content.

SafetySetting

type SafetySetting struct {
	// Required. Harm category.
	Category HarmCategory
	// Required. The harm block threshold.
	Threshold HarmBlockThreshold
	// Optional. Specify if the threshold is used for probability or severity
	// score. If not specified, the threshold is used for probability score.
	Method HarmBlockMethod
}

SafetySetting is safety settings.

Schema

type Schema struct {
	// Optional. The type of the data.
	Type Type
	// Optional. The format of the data.
	// Supported formats:
	//
	//	for NUMBER type: "float", "double"
	//	for INTEGER type: "int32", "int64"
	//	for STRING type: "email", "byte", etc
	Format string
	// Optional. The title of the Schema.
	Title string
	// Optional. The description of the data.
	Description string
	// Optional. Indicates if the value may be null.
	Nullable bool
	// Optional. SCHEMA FIELDS FOR TYPE ARRAY
	// Schema of the elements of Type.ARRAY.
	Items *Schema
	// Optional. Minimum number of the elements for Type.ARRAY.
	MinItems int64
	// Optional. Maximum number of the elements for Type.ARRAY.
	MaxItems int64
	// Optional. Possible values of the element of Type.STRING with enum format.
	// For example we can define an Enum Direction as :
	// {type:STRING, format:enum, enum:["EAST", NORTH", "SOUTH", "WEST"]}
	Enum []string
	// Optional. SCHEMA FIELDS FOR TYPE OBJECT
	// Properties of Type.OBJECT.
	Properties map[string]*Schema
	// Optional. Required properties of Type.OBJECT.
	Required []string
	// Optional. Minimum number of the properties for Type.OBJECT.
	MinProperties int64
	// Optional. Maximum number of the properties for Type.OBJECT.
	MaxProperties int64
	// Optional. SCHEMA FIELDS FOR TYPE INTEGER and NUMBER
	// Minimum value of the Type.INTEGER and Type.NUMBER
	Minimum float64
	// Optional. Maximum value of the Type.INTEGER and Type.NUMBER
	Maximum float64
	// Optional. SCHEMA FIELDS FOR TYPE STRING
	// Minimum length of the Type.STRING
	MinLength int64
	// Optional. Maximum length of the Type.STRING
	MaxLength int64
	// Optional. Pattern of the Type.STRING to restrict a string to a regular
	// expression.
	Pattern string
}

Schema is used to define the format of input/output data. Represents a select subset of an OpenAPI 3.0 schema object. More fields may be added in the future as needed.

Text

type Text string

A Text is a piece of text, like a question or phrase.

Tool

type Tool struct {
	// Optional. Function tool type.
	// One or more function declarations to be passed to the model along with the
	// current user query. Model may decide to call a subset of these functions
	// by populating [FunctionCall][content.part.function_call] in the response.
	// User should provide a [FunctionResponse][content.part.function_response]
	// for each function call in the next turn. Based on the function responses,
	// Model will generate the final response back to the user.
	// Maximum 64 function declarations can be provided.
	FunctionDeclarations []*FunctionDeclaration
}

Tool details that the model may use to generate response.

A Tool is a piece of code that enables the system to interact with external systems to perform an action, or set of actions, outside of knowledge and scope of the model. A Tool object should contain exactly one type of Tool (e.g FunctionDeclaration, Retrieval or GoogleSearchRetrieval).

Example

package main

import (
	"context"
	"fmt"
	"log"

	"cloud.google.com/go/vertexai/genai"
)

// Your GCP project
const projectID = "your-project"

// A GCP location like "us-central1"; if you're using standard Google-published
// models (like untuned Gemini models), you can keep location blank ("").
const location = "some-gcp-location"

func main() {
	ctx := context.Background()
	client, err := genai.NewClient(ctx, projectID, location)
	if err != nil {
		log.Fatal(err)
	}
	defer client.Close()

	currentWeather := func(city string) string {
		switch city {
		case "New York, NY":
			return "cold"
		case "Miami, FL":
			return "warm"
		default:
			return "unknown"
		}
	}

	// To use functions / tools, we have to first define a schema that describes
	// the function to the model. The schema is similar to OpenAPI 3.0.
	//
	// In this example, we create a single function that provides the model with
	// a weather forecast in a given location.
	schema := &genai.Schema{
		Type: genai.TypeObject,
		Properties: map[string]*genai.Schema{
			"location": {
				Type:        genai.TypeString,
				Description: "The city and state, e.g. San Francisco, CA",
			},
			"unit": {
				Type: genai.TypeString,
				Enum: []string{"celsius", "fahrenheit"},
			},
		},
		Required: []string{"location"},
	}

	weatherTool := &genai.Tool{
		FunctionDeclarations: []*genai.FunctionDeclaration{{
			Name:        "CurrentWeather",
			Description: "Get the current weather in a given location",
			Parameters:  schema,
		}},
	}

	model := client.GenerativeModel("gemini-1.0-pro")

	// Before initiating a conversation, we tell the model which tools it has
	// at its disposal.
	model.Tools = []*genai.Tool{weatherTool}

	// For using tools, the chat mode is useful because it provides the required
	// chat context. A model needs to have tools supplied to it in the chat
	// history so it can use them in subsequent conversations.
	//
	// The flow of message expected here is:
	//
	// 1. We send a question to the model
	// 2. The model recognizes that it needs to use a tool to answer the question,
	//    an returns a FunctionCall response asking to use the CurrentWeather
	//    tool.
	// 3. We send a FunctionResponse message, simulating the return value of
	//    CurrentWeather for the model's query.
	// 4. The model provides its text answer in response to this message.
	session := model.StartChat()

	res, err := session.SendMessage(ctx, genai.Text("What is the weather like in New York?"))
	if err != nil {
		log.Fatal(err)
	}

	part := res.Candidates[0].Content.Parts[0]
	funcall, ok := part.(genai.FunctionCall)
	if !ok {
		log.Fatalf("expected FunctionCall: %v", part)
	}

	if funcall.Name != "CurrentWeather" {
		log.Fatalf("expected CurrentWeather: %v", funcall.Name)
	}

	// Expect the model to pass a proper string "location" argument to the tool.
	locArg, ok := funcall.Args["location"].(string)
	if !ok {
		log.Fatalf("expected string: %v", funcall.Args["location"])
	}

	weatherData := currentWeather(locArg)
	res, err = session.SendMessage(ctx, genai.FunctionResponse{
		Name: weatherTool.FunctionDeclarations[0].Name,
		Response: map[string]any{
			"weather": weatherData,
		},
	})
	if err != nil {
		log.Fatal(err)
	}

	printResponse(res)
}

func printResponse(resp *genai.GenerateContentResponse) {
	for _, cand := range resp.Candidates {
		for _, part := range cand.Content.Parts {
			fmt.Println(part)
		}
	}
	fmt.Println("---")
}

ToolConfig

type ToolConfig struct {
	// Optional. Function calling config.
	FunctionCallingConfig *FunctionCallingConfig
}

ToolConfig configures tools.

Type

type Type int32

Type contains the list of OpenAPI data types as defined by https://swagger.io/docs/specification/data-models/data-types/

TypeUnspecified, TypeString, TypeNumber, TypeInteger, TypeBoolean, TypeArray, TypeObject

const (
	// TypeUnspecified means not specified, should not be used.
	TypeUnspecified Type = 0
	// TypeString means openAPI string type
	TypeString Type = 1
	// TypeNumber means openAPI number type
	TypeNumber Type = 2
	// TypeInteger means openAPI integer type
	TypeInteger Type = 3
	// TypeBoolean means openAPI boolean type
	TypeBoolean Type = 4
	// TypeArray means openAPI array type
	TypeArray Type = 5
	// TypeObject means openAPI object type
	TypeObject Type = 6
)

func (Type) String

func (v Type) String() string

UsageMetadata

type UsageMetadata struct {
	// Number of tokens in the request.
	PromptTokenCount int32
	// Number of tokens in the response(s).
	CandidatesTokenCount int32
	TotalTokenCount      int32
}

UsageMetadata is usage metadata about response(s).