Documentation
¶
Overview ¶
Package azureaigateway provides a plugin for Azure API Management (APIM) Gateway fronting multiple AI backends (AWS Bedrock, Azure OpenAI, Google Vertex AI).
Each backend only defines what differs: endpoint path, auth header, request/response format. The shared HTTP plumbing lives in the Client.
Package azureaigateway - Azure OpenAI backend for Azure OpenAI using OpenAI Chat Completions API format ¶
Package azureaigateway - Bedrock backend for AWS Bedrock using Anthropic Messages API format ¶
Package azureaigateway - Vertex AI backend for Google Vertex AI using Gemini API format
Index ¶
- type AzureOpenAIBackend
- func (b *AzureOpenAIBackend) AuthHeader() (string, string)
- func (b *AzureOpenAIBackend) BuildEndpoint(baseURL, deploymentName string) string
- func (b *AzureOpenAIBackend) ListModels() ([]string, error)
- func (b *AzureOpenAIBackend) ParseResponse(body []byte) (string, error)
- func (b *AzureOpenAIBackend) PrepareRequest(msgs []*chat.ChatCompletionMessage, opts *domain.ChatOptions) ([]byte, error)
- type Backend
- type BedrockBackend
- func (b *BedrockBackend) AuthHeader() (string, string)
- func (b *BedrockBackend) BuildEndpoint(baseURL, model string) string
- func (b *BedrockBackend) ListModels() ([]string, error)
- func (b *BedrockBackend) ParseResponse(body []byte) (string, error)
- func (b *BedrockBackend) PrepareRequest(msgs []*chat.ChatCompletionMessage, opts *domain.ChatOptions) ([]byte, error)
- type Client
- type VertexAIBackend
- func (b *VertexAIBackend) AuthHeader() (string, string)
- func (b *VertexAIBackend) BuildEndpoint(baseURL, model string) string
- func (b *VertexAIBackend) ListModels() ([]string, error)
- func (b *VertexAIBackend) ParseResponse(body []byte) (string, error)
- func (b *VertexAIBackend) PrepareRequest(msgs []*chat.ChatCompletionMessage, opts *domain.ChatOptions) ([]byte, error)
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
This section is empty.
Types ¶
type AzureOpenAIBackend ¶
type AzureOpenAIBackend struct {
// contains filtered or unexported fields
}
AzureOpenAIBackend implements the Backend interface for Azure OpenAI through Azure APIM Gateway
func NewAzureOpenAIBackend ¶
func NewAzureOpenAIBackend(subscriptionKey, apiVersion string) *AzureOpenAIBackend
NewAzureOpenAIBackend creates a new Azure OpenAI backend handler If apiVersion is empty, defaults to "2025-04-01-preview"
func (*AzureOpenAIBackend) AuthHeader ¶
func (b *AzureOpenAIBackend) AuthHeader() (string, string)
AuthHeader returns the Azure OpenAI auth header
func (*AzureOpenAIBackend) BuildEndpoint ¶
func (b *AzureOpenAIBackend) BuildEndpoint(baseURL, deploymentName string) string
BuildEndpoint constructs the Azure OpenAI API endpoint URL API version reference: https://learn.microsoft.com/azure/ai-services/openai/reference
func (*AzureOpenAIBackend) ListModels ¶
func (b *AzureOpenAIBackend) ListModels() ([]string, error)
ListModels returns the list of models available through Azure OpenAI. These are deployment names that must exist in your Azure OpenAI resource.
func (*AzureOpenAIBackend) ParseResponse ¶
func (b *AzureOpenAIBackend) ParseResponse(body []byte) (string, error)
ParseResponse parses Azure OpenAI API response (OpenAI chat completions format)
func (*AzureOpenAIBackend) PrepareRequest ¶
func (b *AzureOpenAIBackend) PrepareRequest(msgs []*chat.ChatCompletionMessage, opts *domain.ChatOptions) ([]byte, error)
PrepareRequest converts messages to Azure OpenAI (OpenAI-compatible) API format
type Backend ¶
type Backend interface {
// ListModels returns the list of models available for this backend
ListModels() ([]string, error)
// BuildEndpoint constructs the full API endpoint URL for the given model
BuildEndpoint(baseURL, model string) string
// AuthHeader returns the header name and value for authentication.
// Each APIM backend uses a different auth header:
// Bedrock: "Authorization", "Bearer <key>"
// Azure OpenAI: "api-key", "<key>"
// Vertex AI: "x-goog-api-key", "<key>"
AuthHeader() (name, value string)
// PrepareRequest prepares the HTTP request body for this backend's API format
PrepareRequest(msgs []*chat.ChatCompletionMessage, opts *domain.ChatOptions) ([]byte, error)
// ParseResponse parses the HTTP response body into text content
ParseResponse(body []byte) (string, error)
}
Backend defines the interface that all Azure AI Gateway backends must implement. Each backend only provides what is unique to its API format. The shared HTTP mechanics (request execution, error handling, streaming fallback) are handled by the Client.
type BedrockBackend ¶
type BedrockBackend struct {
// contains filtered or unexported fields
}
BedrockBackend implements the Backend interface for AWS Bedrock through Azure APIM Gateway
func NewBedrockBackend ¶
func NewBedrockBackend(subscriptionKey string) *BedrockBackend
NewBedrockBackend creates a new Bedrock backend handler
func (*BedrockBackend) AuthHeader ¶
func (b *BedrockBackend) AuthHeader() (string, string)
AuthHeader returns the Bedrock auth header (Bearer token)
func (*BedrockBackend) BuildEndpoint ¶
func (b *BedrockBackend) BuildEndpoint(baseURL, model string) string
BuildEndpoint constructs the Bedrock API endpoint URL
func (*BedrockBackend) ListModels ¶
func (b *BedrockBackend) ListModels() ([]string, error)
ListModels returns the list of available Bedrock inference profiles
func (*BedrockBackend) ParseResponse ¶
func (b *BedrockBackend) ParseResponse(body []byte) (string, error)
ParseResponse parses Bedrock API response (Anthropic content blocks)
func (*BedrockBackend) PrepareRequest ¶
func (b *BedrockBackend) PrepareRequest(msgs []*chat.ChatCompletionMessage, opts *domain.ChatOptions) ([]byte, error)
PrepareRequest converts messages to Bedrock API format (Anthropic Messages API). System messages are extracted into the top-level "system" field per the Anthropic API spec.
type Client ¶
type Client struct {
*plugins.PluginBase
BackendType *plugins.SetupQuestion
GatewayURL *plugins.SetupQuestion
SubscriptionKey *plugins.SetupQuestion
APIVersion *plugins.SetupQuestion
// contains filtered or unexported fields
}
Client implements the Azure AI Gateway vendor for Fabric. It supports multiple backends (Bedrock, Azure OpenAI, Vertex AI) through a unified Azure APIM Gateway with shared subscription key authentication.
func (*Client) IsConfigured ¶
IsConfigured returns true if both gateway URL and subscription key are configured
func (*Client) ListModels ¶
ListModels delegates to the active backend
func (*Client) Send ¶
func (c *Client) Send(ctx context.Context, msgs []*chat.ChatCompletionMessage, opts *domain.ChatOptions) (string, error)
Send sends a non-streaming request through the APIM gateway. This is the single implementation of HTTP plumbing shared by all backends.
func (*Client) SendStream ¶
func (c *Client) SendStream(msgs []*chat.ChatCompletionMessage, opts *domain.ChatOptions, channel chan domain.StreamUpdate) error
SendStream falls back to non-streaming (APIM gateway doesn't support SSE pass-through).
NOTE: This method uses context.Background() because the ai.Vendor interface does not accept a context parameter for SendStream. If the caller disconnects, this request will continue until the gateway timeout (300s). A future update to the ai.Vendor interface should add context propagation to SendStream.
type VertexAIBackend ¶
type VertexAIBackend struct {
// contains filtered or unexported fields
}
VertexAIBackend implements the Backend interface for Google Vertex AI (Gemini) through Azure APIM Gateway.
func NewVertexAIBackend ¶
func NewVertexAIBackend(subscriptionKey string) *VertexAIBackend
NewVertexAIBackend creates a new Vertex AI backend handler
func (*VertexAIBackend) AuthHeader ¶
func (b *VertexAIBackend) AuthHeader() (string, string)
AuthHeader returns the Vertex AI auth header (Google API key via APIM)
func (*VertexAIBackend) BuildEndpoint ¶
func (b *VertexAIBackend) BuildEndpoint(baseURL, model string) string
BuildEndpoint constructs the Vertex AI API endpoint URL Uses /publishers/google/models/{model}:generateContent path per Azure APIM Gateway routing This is the APIM-specific path that proxies to Google's Vertex AI service (differs from direct Vertex AI API which uses /v1beta/models/{model}:generateContent)
func (*VertexAIBackend) ListModels ¶
func (b *VertexAIBackend) ListModels() ([]string, error)
ListModels returns the list of Gemini models available through Vertex AI
func (*VertexAIBackend) ParseResponse ¶
func (b *VertexAIBackend) ParseResponse(body []byte) (string, error)
ParseResponse parses Gemini API response (candidates/content/parts)
func (*VertexAIBackend) PrepareRequest ¶
func (b *VertexAIBackend) PrepareRequest(msgs []*chat.ChatCompletionMessage, opts *domain.ChatOptions) ([]byte, error)
PrepareRequest converts messages to Gemini API format (contents/parts). System messages are extracted into the top-level "systemInstruction" field per the Gemini API spec.
Source Files