Documentation
¶
Overview ¶
Package google provides a Google Gemini language model implementation for GoAI.
It uses the Gemini REST API with SSE streaming and a unique wire format.
Usage:
model := google.Chat("gemini-2.5-flash", google.WithAPIKey("..."))
result, err := goai.GenerateText(ctx, model, goai.WithPrompt("Hello"))
Index ¶
Constants ¶
This section is empty.
Variables ¶
var Tools = struct { // GoogleSearch enables grounding with Google Search. // The model decides when to search based on the prompt. // Returns sources via groundingMetadata in response. GoogleSearch func(opts ...GoogleSearchOption) provider.ToolDefinition // URLContext enables URL context tool that gives Gemini access to web content. // The model uses URLs from the prompt to fetch and process content. // Requires Gemini 2.0+. URLContext func() provider.ToolDefinition // CodeExecution enables the model to generate and run Python code. // The model can write code, execute it in a sandboxed environment, and // use the output to formulate its response. // Requires Gemini 2.0+. CodeExecution func() provider.ToolDefinition }{ GoogleSearch: googleSearchTool, URLContext: urlContextTool, CodeExecution: codeExecutionTool, }
Tools provides factory functions for Google provider-defined tools. These tools use Gemini's built-in capabilities (Google Search grounding, URL context, code execution). Requires Gemini 2.0+. Matches Vercel AI SDK's google.tools.
Functions ¶
func Chat ¶
func Chat(modelID string, opts ...Option) provider.LanguageModel
Chat creates a Google Gemini language model for the given model ID.
func Embedding ¶
func Embedding(modelID string, opts ...Option) provider.EmbeddingModel
Embedding creates a Google Gemini embedding model for the given model ID. Supports text-embedding-004, embedding-001, etc.
func Image ¶
func Image(modelID string, opts ...Option) provider.ImageModel
Image creates a Google Generative AI image model.
For Imagen models (imagen-4.0-generate-001, etc.), uses the :predict endpoint with instances/parameters format -- same wire format as Vertex AI but with Gemini API auth (x-goog-api-key).
For Gemini image models (gemini-2.5-flash-image, etc.), uses the generateContent endpoint with responseModalities: ["IMAGE"] -- Nano Banana pattern.
Provider options (under "google" key in ImageParams.ProviderOptions):
Imagen models:
- personGeneration (string): "dont_allow", "allow_adult", "allow_all"
- aspectRatio (string): overrides params.AspectRatio
Gemini image models:
- imageConfig (map): {"aspectRatio": "1:1", "imageSize": "1K"}
Types ¶
type GoogleSearchOption ¶
type GoogleSearchOption func(*googleSearchConfig)
GoogleSearchOption configures the Google Search grounding tool.
func WithImageSearch ¶
func WithImageSearch() GoogleSearchOption
WithImageSearch enables image search results.
func WithTimeRange ¶
func WithTimeRange(startTime, endTime string) GoogleSearchOption
WithTimeRange restricts search results to a specific time range (RFC3339 format).
func WithWebSearch ¶
func WithWebSearch() GoogleSearchOption
WithWebSearch enables web search results.
type Option ¶
type Option func(*options)
Option configures the Google provider.
func WithAPIKey ¶
WithAPIKey sets a static API key for authentication.
func WithBaseURL ¶
WithBaseURL overrides the default Gemini API base URL.
func WithHTTPClient ¶
WithHTTPClient sets a custom HTTP client for all requests.
func WithHeaders ¶
WithHeaders sets additional HTTP headers sent with every request.
func WithTokenSource ¶
func WithTokenSource(ts provider.TokenSource) Option
WithTokenSource sets a dynamic token source for authentication.
Source Files