Files
CLIProxyAPI/internal/runtime/executor/gemini_vertex_executor.go
Luis Pater d9451cb65d feat(vertex): support partner and open models via endpoints/openapi route
- Route Vertex partner/open MaaS models and explicit openapi-enabled credentials through the OpenAI-compatible endpoints/openapi endpoint while reusing service-account OAuth token acquisition and supporting openapi_location overrides.
- Register Vertex service-account MaaS models and aliases from credential metadata and OAuth alias rules so partner models are discoverable and routable.
- Normalize Vertex OpenAPI model identifiers and reject static API keys against official Google Vertex OpenAPI hosts that require OAuth tokens.

Closes: #6763
2026-10-19 12:56:25 +08:00

1899 lines
72 KiB
Go

// Package executor provides runtime execution capabilities for various AI service providers.
// This file implements the Vertex AI Gemini executor that talks to Google Vertex AI
// endpoints using service account credentials or API keys.
package executor
import (
"bufio"
"bytes"
"context"
"encoding/json"
"errors"
"fmt"
"io"
"net/http"
"strings"
"time"
vertexauth "github.com/router-for-me/CLIProxyAPI/v8/internal/auth/vertex"
"github.com/router-for-me/CLIProxyAPI/v8/internal/config"
"github.com/router-for-me/CLIProxyAPI/v8/internal/runtime/executor/helps"
internalsignature "github.com/router-for-me/CLIProxyAPI/v8/internal/signature"
"github.com/router-for-me/CLIProxyAPI/v8/internal/thinking"
"github.com/router-for-me/CLIProxyAPI/v8/internal/util"
cliproxyauth "github.com/router-for-me/CLIProxyAPI/v8/sdk/cliproxy/auth"
cliproxyexecutor "github.com/router-for-me/CLIProxyAPI/v8/sdk/cliproxy/executor"
sdktranslator "github.com/router-for-me/CLIProxyAPI/v8/sdk/translator"
log "github.com/sirupsen/logrus"
"github.com/tidwall/gjson"
"github.com/tidwall/sjson"
"golang.org/x/oauth2"
"golang.org/x/oauth2/google"
)
const (
// vertexAPIVersion aligns with current public Vertex Generative AI API.
vertexAPIVersion = "v1"
)
// isImagenModel checks if the model name is an Imagen image generation model.
// Imagen models use the :predict action instead of :generateContent.
func isImagenModel(model string) bool {
lowerModel := strings.ToLower(model)
return strings.Contains(lowerModel, "imagen")
}
// getVertexAction returns the appropriate action for the given model.
// Imagen models use "predict", while Gemini models use "generateContent".
func getVertexAction(model string, isStream bool) string {
if isImagenModel(model) {
return "predict"
}
if isStream {
return "streamGenerateContent"
}
return "generateContent"
}
func isGeminiVertexTerminalStreamChunk(chunk []byte) bool {
for _, line := range bytes.Split(chunk, []byte("\n")) {
trimmed := bytes.TrimSpace(line)
if bytes.Equal(trimmed, []byte("data: [DONE]")) || bytes.Equal(trimmed, []byte("[DONE]")) {
return true
}
payload := helps.JSONPayload(line)
if len(payload) == 0 {
continue
}
eventType := gjson.GetBytes(payload, "type").String()
if eventType == "response.completed" || eventType == "response.incomplete" || eventType == "response.done" || eventType == "message_stop" {
return true
}
for _, path := range []string{
"choices.0.finish_reason",
"response.choices.0.finish_reason",
"candidates.0.finishReason",
"response.candidates.0.finishReason",
} {
if finishReason := gjson.GetBytes(payload, path); finishReason.Exists() && strings.TrimSpace(finishReason.String()) != "" {
return true
}
}
}
return false
}
// convertImagenToGeminiResponse converts Imagen API response to Gemini format
// so it can be processed by the standard translation pipeline.
// This ensures Imagen models return responses in the same format as gemini-3-pro-image-preview.
func convertImagenToGeminiResponse(data []byte, model string) []byte {
predictions := gjson.GetBytes(data, "predictions")
if !predictions.Exists() || !predictions.IsArray() {
return data
}
// Build Gemini-compatible response with inlineData
parts := make([]map[string]any, 0)
for _, pred := range predictions.Array() {
imageData := pred.Get("bytesBase64Encoded").String()
mimeType := pred.Get("mimeType").String()
if mimeType == "" {
mimeType = "image/png"
}
if imageData != "" {
parts = append(parts, map[string]any{
"inlineData": map[string]any{
"mimeType": mimeType,
"data": imageData,
},
})
}
}
// Generate unique response ID using timestamp
responseId := fmt.Sprintf("imagen-%d", time.Now().UnixNano())
response := map[string]any{
"candidates": []map[string]any{{
"content": map[string]any{
"parts": parts,
"role": "model",
},
"finishReason": "STOP",
}},
"responseId": responseId,
"modelVersion": model,
// Imagen API doesn't return token counts, set to 0 for tracking purposes
"usageMetadata": map[string]any{
"promptTokenCount": 0,
"candidatesTokenCount": 0,
"totalTokenCount": 0,
},
}
result, err := json.Marshal(response)
if err != nil {
return data
}
return result
}
// convertToImagenRequest converts a Gemini-style request to Imagen API format.
// Imagen API uses a different structure: instances[].prompt instead of contents[].
func convertToImagenRequest(payload []byte) ([]byte, error) {
// Extract prompt from Gemini-style contents
prompt := ""
// Try to get prompt from contents[0].parts[0].text
contentsText := gjson.GetBytes(payload, "contents.0.parts.0.text")
if contentsText.Exists() {
prompt = contentsText.String()
}
// If no contents, try messages format (OpenAI-compatible)
if prompt == "" {
messagesText := gjson.GetBytes(payload, "messages.#.content")
if messagesText.Exists() && messagesText.IsArray() {
for _, msg := range messagesText.Array() {
if msg.String() != "" {
prompt = msg.String()
break
}
}
}
}
// If still no prompt, try direct prompt field
if prompt == "" {
directPrompt := gjson.GetBytes(payload, "prompt")
if directPrompt.Exists() {
prompt = directPrompt.String()
}
}
if prompt == "" {
return nil, fmt.Errorf("imagen: no prompt found in request")
}
// Build Imagen API request
imagenReq := map[string]any{
"instances": []map[string]any{
{
"prompt": prompt,
},
},
"parameters": map[string]any{
"sampleCount": 1,
},
}
// Extract optional parameters
if aspectRatio := gjson.GetBytes(payload, "aspectRatio"); aspectRatio.Exists() {
imagenReq["parameters"].(map[string]any)["aspectRatio"] = aspectRatio.String()
}
if sampleCount := gjson.GetBytes(payload, "sampleCount"); sampleCount.Exists() {
imagenReq["parameters"].(map[string]any)["sampleCount"] = int(sampleCount.Int())
}
if negativePrompt := gjson.GetBytes(payload, "negativePrompt"); negativePrompt.Exists() {
imagenReq["instances"].([]map[string]any)[0]["negativePrompt"] = negativePrompt.String()
}
return json.Marshal(imagenReq)
}
// GeminiVertexExecutor sends requests to Vertex AI Gemini endpoints using service account credentials.
type GeminiVertexExecutor struct {
cfg *config.Config
}
// NewGeminiVertexExecutor creates a new Vertex AI Gemini executor instance.
//
// Parameters:
// - cfg: The application configuration
//
// Returns:
// - *GeminiVertexExecutor: A new Vertex AI Gemini executor instance
func NewGeminiVertexExecutor(cfg *config.Config) *GeminiVertexExecutor {
return &GeminiVertexExecutor{cfg: cfg}
}
// Identifier returns the executor identifier.
func (e *GeminiVertexExecutor) Identifier() string { return "vertex" }
// PrepareRequest injects Vertex credentials into the outgoing HTTP request.
func (e *GeminiVertexExecutor) PrepareRequest(req *http.Request, auth *cliproxyauth.Auth) error {
if req == nil {
return nil
}
apiKey, _ := vertexAPICreds(auth)
if strings.TrimSpace(apiKey) != "" {
req.Header.Set("x-goog-api-key", apiKey)
req.Header.Del("Authorization")
return nil
}
_, _, saJSON, errCreds := vertexCreds(auth)
if errCreds != nil {
return errCreds
}
token, errToken := vertexAccessToken(req.Context(), e.cfg, auth, saJSON)
if errToken != nil {
return errToken
}
if strings.TrimSpace(token) == "" {
return statusErr{code: http.StatusUnauthorized, msg: "missing access token"}
}
req.Header.Set("Authorization", "Bearer "+token)
req.Header.Del("x-goog-api-key")
return nil
}
// HttpRequest injects Vertex credentials into the request and executes it.
func (e *GeminiVertexExecutor) HttpRequest(ctx context.Context, auth *cliproxyauth.Auth, req *http.Request) (*http.Response, error) {
if req == nil {
return nil, fmt.Errorf("vertex executor: request is nil")
}
if ctx == nil {
ctx = req.Context()
}
httpReq := req.WithContext(ctx)
if err := e.PrepareRequest(httpReq, auth); err != nil {
return nil, err
}
httpClient := helps.NewProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
return httpClient.Do(httpReq)
}
// Execute performs a non-streaming request to the Vertex AI API.
func (e *GeminiVertexExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
ctx = helps.EnsureSessionContext(ctx, opts, req.Payload)
if opts.Alt == "responses/compact" {
return resp, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
}
if helps.ShouldExecuteVertexOpenAPI(auth, req.Model) {
return e.executeOpenAPI(ctx, auth, req, opts)
}
if shouldExecuteVertexInteractions(auth, opts) {
return e.executeInteractions(ctx, auth, req, opts)
}
// Try API key authentication first
apiKey, baseURL := vertexAPICreds(auth)
// If no API key found, fall back to service account authentication
if apiKey == "" {
projectID, location, saJSON, errCreds := vertexCreds(auth)
if errCreds != nil {
return resp, errCreds
}
return e.executeWithServiceAccount(ctx, auth, req, opts, projectID, location, saJSON)
}
// Use API key authentication
return e.executeWithAPIKey(ctx, auth, req, opts, apiKey, baseURL)
}
// ExecuteStream performs a streaming request to the Vertex AI API.
func (e *GeminiVertexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (*cliproxyexecutor.StreamResult, error) {
ctx = helps.EnsureSessionContext(ctx, opts, req.Payload)
if opts.Alt == "responses/compact" {
return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
}
if helps.ShouldExecuteVertexOpenAPI(auth, req.Model) {
return e.executeOpenAPIStream(ctx, auth, req, opts)
}
if shouldExecuteVertexInteractions(auth, opts) {
return e.executeInteractionsStream(ctx, auth, req, opts)
}
// Try API key authentication first
apiKey, baseURL := vertexAPICreds(auth)
// If no API key found, fall back to service account authentication
if apiKey == "" {
projectID, location, saJSON, errCreds := vertexCreds(auth)
if errCreds != nil {
return nil, errCreds
}
return e.executeStreamWithServiceAccount(ctx, auth, req, opts, projectID, location, saJSON)
}
// Use API key authentication
return e.executeStreamWithAPIKey(ctx, auth, req, opts, apiKey, baseURL)
}
// CountTokens counts tokens for the given request using the Vertex AI API.
func (e *GeminiVertexExecutor) CountTokens(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Response, error) {
if helps.ShouldExecuteVertexOpenAPI(auth, req.Model) {
normalizedReq, normalizedOpts := normalizeVertexOpenAPIRequest(req, opts)
return NewOpenAICompatExecutor(e.Identifier(), e.cfg).CountTokens(ctx, auth, normalizedReq, normalizedOpts)
}
// Try API key authentication first
apiKey, baseURL := vertexAPICreds(auth)
// If no API key found, fall back to service account authentication
if apiKey == "" {
projectID, location, saJSON, errCreds := vertexCreds(auth)
if errCreds != nil {
return cliproxyexecutor.Response{}, errCreds
}
return e.countTokensWithServiceAccount(ctx, auth, req, opts, projectID, location, saJSON)
}
// Use API key authentication
return e.countTokensWithAPIKey(ctx, auth, req, opts, apiKey, baseURL)
}
func normalizeVertexOpenAPIRequest(req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (cliproxyexecutor.Request, cliproxyexecutor.Options) {
normalizedModel := helps.NormalizeVertexOpenAPIModel(req.Model)
if normalizedModel == "" || normalizedModel == req.Model {
return req, opts
}
if len(opts.OriginalRequest) == 0 && len(req.Payload) > 0 {
opts.OriginalRequest = req.Payload
}
req.Model = normalizedModel
return req, opts
}
func hasVertexServiceAccount(auth *cliproxyauth.Auth) bool {
if auth == nil || auth.Metadata == nil {
return false
}
sa, ok := auth.Metadata["service_account"].(map[string]any)
return ok && len(sa) > 0
}
func (e *GeminiVertexExecutor) resolveOpenAPICredentials(ctx context.Context, auth *cliproxyauth.Auth) (string, string, error) {
apiKey, baseURL := vertexAPICreds(auth)
if apiKey == "" || hasVertexServiceAccount(auth) {
projectID, location, saJSON, errCreds := vertexCreds(auth)
if errCreds != nil {
return "", "", errCreds
}
token, errTok := vertexAccessToken(ctx, e.cfg, auth, saJSON)
if errTok != nil {
log.Errorf("vertex executor: access token error: %v", errTok)
return "", "", statusErr{code: 500, msg: "internal server error"}
}
if strings.TrimSpace(token) == "" {
return "", "", statusErr{code: http.StatusUnauthorized, msg: "missing access token"}
}
openAPILocation := helps.VertexOpenAPILocation(auth, location)
openAPIBaseURL := helps.VertexServiceAccountOpenAPIBaseURL(helps.VertexAuthBaseURL(auth), vertexAPIVersion, projectID, openAPILocation)
return openAPIBaseURL, token, nil
}
if helps.IsGoogleVertexHost(baseURL) {
return "", "", statusErr{code: http.StatusBadRequest, msg: "vertex endpoints/openapi on aiplatform.googleapis.com requires service account credentials"}
}
openAPILocation := helps.VertexOpenAPILocation(auth, "")
openAPIBaseURL := helps.VertexAPIKeyOpenAPIBaseURL(baseURL, vertexAPIVersion, helps.VertexAuthProjectID(auth), openAPILocation)
return openAPIBaseURL, apiKey, nil
}
func (e *GeminiVertexExecutor) executeOpenAPI(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
req, opts = normalizeVertexOpenAPIRequest(req, opts)
openAPIBaseURL, token, errCreds := e.resolveOpenAPICredentials(ctx, auth)
if errCreds != nil {
baseModel := thinking.ParseSuffix(req.Model).ModelName
reporter := helps.NewExecutorUsageReporter(ctx, e, baseModel, auth)
reporter.TrackFailure(ctx, &errCreds)
return resp, errCreds
}
return NewOpenAICompatExecutor(e.Identifier(), e.cfg).executeWithCredentials(ctx, auth, req, opts, openAPIBaseURL, token, nil)
}
func (e *GeminiVertexExecutor) executeOpenAPIStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (*cliproxyexecutor.StreamResult, error) {
req, opts = normalizeVertexOpenAPIRequest(req, opts)
openAPIBaseURL, token, errCreds := e.resolveOpenAPICredentials(ctx, auth)
if errCreds != nil {
baseModel := thinking.ParseSuffix(req.Model).ModelName
reporter := helps.NewExecutorUsageReporter(ctx, e, baseModel, auth)
reporter.TrackFailure(ctx, &errCreds)
return nil, errCreds
}
return NewOpenAICompatExecutor(e.Identifier(), e.cfg).executeStreamWithCredentials(ctx, auth, req, opts, openAPIBaseURL, token, nil)
}
// Refresh refreshes the authentication credentials (no-op for Vertex).
func (e *GeminiVertexExecutor) Refresh(ctx context.Context, auth *cliproxyauth.Auth) (*cliproxyauth.Auth, error) {
if refreshed, handled, err := helps.RefreshAuthViaHome(ctx, e.cfg, auth); handled {
return refreshed, err
}
return auth, nil
}
// executeWithServiceAccount handles authentication using service account credentials.
// This method contains the original service account authentication logic.
func (e *GeminiVertexExecutor) executeWithServiceAccount(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options, projectID, location string, saJSON []byte) (resp cliproxyexecutor.Response, err error) {
baseModel := thinking.ParseSuffix(req.Model).ModelName
reporter := helps.NewExecutorUsageReporter(ctx, e, baseModel, auth)
defer reporter.TrackFailure(ctx, &err)
var body []byte
var originalTranslated []byte
// Handle Imagen models with special request format
if isImagenModel(baseModel) {
imagenBody, errImagen := convertToImagenRequest(req.Payload)
if errImagen != nil {
return resp, errImagen
}
body = imagenBody
originalTranslated = append([]byte(nil), imagenBody...)
if len(opts.OriginalRequest) > 0 {
originalTranslated, errImagen = convertToImagenRequest(opts.OriginalRequest)
if errImagen != nil {
return resp, errImagen
}
}
} else {
// Standard Gemini translation flow
from := opts.SourceFormat
to := sdktranslator.FromString("gemini")
originalPayloadSource := req.Payload
if len(opts.OriginalRequest) > 0 {
originalPayloadSource = opts.OriginalRequest
}
originalPayload := originalPayloadSource
originalTranslated = helps.TranslateRequestWithCodexMultiAgentV2(ctx, opts.Headers, e.cfg, from, to, baseModel, originalPayload, false)
body, err = helps.TranslateRequestReturningError(ctx, opts.Headers, e.cfg, from, to, baseModel, req.Payload, false, false)
if err != nil {
return resp, err
}
body, err = helps.ApplyRequestThinking(body, req, opts, from.String(), to.String(), e.Identifier())
if err != nil {
return resp, err
}
body = fixGeminiImageAspectRatio(baseModel, body)
body = helps.SetStringIfDifferent(body, "model", baseModel)
body = helps.StripVertexOpenAIResponsesToolCallIDs(body, from.String())
body = internalsignature.SanitizeGeminiRequestThoughtSignatures(body, "contents")
}
action := getVertexAction(baseModel, false)
if req.Metadata != nil {
if a, _ := req.Metadata["action"].(string); a == "countTokens" {
action = "countTokens"
}
}
body = helps.EnsureGeminiLeadingUserContent(body, "contents")
if action != "countTokens" {
body = helps.DropGeminiTrailingThoughtOnlyContent(body, "contents")
body = helps.EnsureGeminiTrailingUserContent(body, "contents")
}
baseURL := vertexBaseURL(location)
url := fmt.Sprintf("%s/%s/projects/%s/locations/%s/publishers/google/models/%s:%s", baseURL, vertexAPIVersion, projectID, location, baseModel, action)
if opts.Alt != "" && action != "countTokens" {
url = url + fmt.Sprintf("?$alt=%s", opts.Alt)
}
body, _ = sjson.DeleteBytes(body, "session_id")
reporter.SetTranslatedReasoningEffort(body, "gemini")
body = helps.ApplyPayloadConfigWithRequest(e.cfg, baseModel, "gemini", opts.SourceFormat.String(), "", body, originalTranslated, helps.PayloadRequestedModel(opts, req.Model), helps.PayloadRequestPath(opts), opts.Headers)
httpReq, errNewReq := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
if errNewReq != nil {
return resp, errNewReq
}
httpReq.Header.Set("Content-Type", "application/json")
if token, errTok := vertexAccessToken(ctx, e.cfg, auth, saJSON); errTok == nil && token != "" {
httpReq.Header.Set("Authorization", "Bearer "+token)
} else if errTok != nil {
log.Errorf("vertex executor: access token error: %v", errTok)
return resp, statusErr{code: 500, msg: "internal server error"}
}
applyGeminiHeaders(httpReq, auth, opts.Headers)
var attrs map[string]string
if auth != nil {
attrs = auth.Attributes
}
util.ApplyCustomHeadersFromAttrs(httpReq, attrs, opts.Headers)
var authID, authLabel, authType, authValue string
if auth != nil {
authID = auth.ID
authLabel = auth.Label
authType, authValue = auth.AccountInfo()
}
helps.RecordAPIRequest(ctx, e.cfg, helps.UpstreamRequestLog{
URL: url,
Method: http.MethodPost,
Headers: httpReq.Header.Clone(),
Body: body,
Provider: e.Identifier(),
AuthID: authID,
AuthLabel: authLabel,
AuthType: authType,
AuthValue: authValue,
})
httpClient := helps.NewProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
httpClient = reporter.TrackHTTPClient(httpClient)
httpResp, errDo := httpClient.Do(httpReq)
if errDo != nil {
helps.RecordAPIResponseError(ctx, e.cfg, errDo)
return resp, errDo
}
defer func() {
if errClose := httpResp.Body.Close(); errClose != nil {
log.Errorf("vertex executor: close response body error: %v", errClose)
}
}()
helps.RecordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
b, _ := io.ReadAll(httpResp.Body)
helps.AppendAPIResponseChunk(ctx, e.cfg, b)
helps.LogWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, helps.SummarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
err = statusErr{code: httpResp.StatusCode, msg: string(b)}
return resp, err
}
data, errRead := io.ReadAll(httpResp.Body)
if errRead != nil {
helps.RecordAPIResponseError(ctx, e.cfg, errRead)
return resp, errRead
}
helps.AppendAPIResponseChunk(ctx, e.cfg, data)
reporter.ObserveResponseModel(data)
// For Imagen models, convert response to Gemini format before translation
// This ensures Imagen responses use the same format as gemini-3-pro-image-preview
if isImagenModel(baseModel) {
data = convertImagenToGeminiResponse(data, baseModel)
}
// Standard Gemini translation (works for both Gemini and converted Imagen responses)
responseFormat := cliproxyexecutor.ResponseFormatOrSource(opts)
to := sdktranslator.FromString("gemini")
var param any
out := sdktranslator.TranslateNonStream(ctx, to, responseFormat, req.Model, helps.ApplyPatchOriginalRequest(req, opts), body, data, &param)
if helps.ApplyPatchTranslationError(param) != nil || len(out) == 0 {
return cliproxyexecutor.Response{}, statusErr{code: http.StatusBadGateway, msg: helps.ApplyPatchUpstreamErrorMessage}
}
if errPayload, okErr := helps.TranslatedClaudeErrorPayload(responseFormat, out); okErr {
err = statusErr{code: http.StatusBadGateway, msg: errPayload}
reporter.PublishFailureWithDetail(ctx, helps.ParseGeminiUsage(data), err)
return cliproxyexecutor.Response{}, err
}
reporter.Publish(ctx, helps.ParseGeminiUsage(data))
if responseFormat == sdktranslator.FormatOpenAIResponse {
out = helps.EnsureResponsesUsageDetails(out)
}
resp = cliproxyexecutor.Response{Payload: out, Headers: httpResp.Header.Clone()}
return resp, nil
}
// executeWithAPIKey handles authentication using API key credentials.
func (e *GeminiVertexExecutor) executeWithAPIKey(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options, apiKey, baseURL string) (resp cliproxyexecutor.Response, err error) {
baseModel := thinking.ParseSuffix(req.Model).ModelName
reporter := helps.NewExecutorUsageReporter(ctx, e, baseModel, auth)
defer reporter.TrackFailure(ctx, &err)
from := opts.SourceFormat
responseFormat := cliproxyexecutor.ResponseFormatOrSource(opts)
to := sdktranslator.FromString("gemini")
originalPayloadSource := req.Payload
if len(opts.OriginalRequest) > 0 {
originalPayloadSource = opts.OriginalRequest
}
originalPayload := originalPayloadSource
originalTranslated := helps.TranslateRequestWithCodexMultiAgentV2(ctx, opts.Headers, e.cfg, from, to, baseModel, originalPayload, false)
body, err := helps.TranslateRequestReturningError(ctx, opts.Headers, e.cfg, from, to, baseModel, req.Payload, false, false)
if err != nil {
return resp, err
}
body, err = helps.ApplyRequestThinking(body, req, opts, from.String(), to.String(), e.Identifier())
if err != nil {
return resp, err
}
body = fixGeminiImageAspectRatio(baseModel, body)
requestedModel := helps.PayloadRequestedModel(opts, req.Model)
requestPath := helps.PayloadRequestPath(opts)
body = helps.SetStringIfDifferent(body, "model", baseModel)
body = helps.StripVertexOpenAIResponsesToolCallIDs(body, from.String())
body = internalsignature.SanitizeGeminiRequestThoughtSignatures(body, "contents")
action := getVertexAction(baseModel, false)
if req.Metadata != nil {
if a, _ := req.Metadata["action"].(string); a == "countTokens" {
action = "countTokens"
}
}
body = helps.EnsureGeminiLeadingUserContent(body, "contents")
if action != "countTokens" {
body = helps.DropGeminiTrailingThoughtOnlyContent(body, "contents")
body = helps.EnsureGeminiTrailingUserContent(body, "contents")
}
// For API key auth, use simpler URL format without project/location.
// The helper honors base URLs that already encode "/publishers/..." so
// configurations like ".../v1/projects/P/locations/L/publishers/google/models"
// keep working without a duplicated "/v1" segment.
url := vertexAPIKeyModelsURL(baseURL, baseModel, action)
if opts.Alt != "" && action != "countTokens" {
url = url + fmt.Sprintf("?$alt=%s", opts.Alt)
}
body, _ = sjson.DeleteBytes(body, "session_id")
reporter.SetTranslatedReasoningEffort(body, to.String())
body = helps.ApplyPayloadConfigWithRequest(e.cfg, baseModel, to.String(), from.String(), "", body, originalTranslated, requestedModel, requestPath, opts.Headers)
httpReq, errNewReq := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
if errNewReq != nil {
return resp, errNewReq
}
httpReq.Header.Set("Content-Type", "application/json")
if apiKey != "" {
httpReq.Header.Set("x-goog-api-key", apiKey)
}
applyGeminiHeaders(httpReq, auth, opts.Headers)
var attrs map[string]string
if auth != nil {
attrs = auth.Attributes
}
util.ApplyCustomHeadersFromAttrs(httpReq, attrs, opts.Headers)
var authID, authLabel, authType, authValue string
if auth != nil {
authID = auth.ID
authLabel = auth.Label
authType, authValue = auth.AccountInfo()
}
helps.RecordAPIRequest(ctx, e.cfg, helps.UpstreamRequestLog{
URL: url,
Method: http.MethodPost,
Headers: httpReq.Header.Clone(),
Body: body,
Provider: e.Identifier(),
AuthID: authID,
AuthLabel: authLabel,
AuthType: authType,
AuthValue: authValue,
})
httpClient := helps.NewProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
httpClient = reporter.TrackHTTPClient(httpClient)
httpResp, errDo := httpClient.Do(httpReq)
if errDo != nil {
helps.RecordAPIResponseError(ctx, e.cfg, errDo)
return resp, errDo
}
defer func() {
if errClose := httpResp.Body.Close(); errClose != nil {
log.Errorf("vertex executor: close response body error: %v", errClose)
}
}()
helps.RecordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
b, _ := io.ReadAll(httpResp.Body)
helps.AppendAPIResponseChunk(ctx, e.cfg, b)
helps.LogWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, helps.SummarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
err = statusErr{code: httpResp.StatusCode, msg: string(b)}
return resp, err
}
data, errRead := io.ReadAll(httpResp.Body)
if errRead != nil {
helps.RecordAPIResponseError(ctx, e.cfg, errRead)
return resp, errRead
}
helps.AppendAPIResponseChunk(ctx, e.cfg, data)
reporter.ObserveResponseModel(data)
var param any
out := sdktranslator.TranslateNonStream(ctx, to, responseFormat, req.Model, helps.ApplyPatchOriginalRequest(req, opts), body, data, &param)
if helps.ApplyPatchTranslationError(param) != nil || len(out) == 0 {
return cliproxyexecutor.Response{}, statusErr{code: http.StatusBadGateway, msg: helps.ApplyPatchUpstreamErrorMessage}
}
if errPayload, okErr := helps.TranslatedClaudeErrorPayload(responseFormat, out); okErr {
err = statusErr{code: http.StatusBadGateway, msg: errPayload}
reporter.PublishFailureWithDetail(ctx, helps.ParseGeminiUsage(data), err)
return cliproxyexecutor.Response{}, err
}
reporter.Publish(ctx, helps.ParseGeminiUsage(data))
if responseFormat == sdktranslator.FormatOpenAIResponse {
out = helps.EnsureResponsesUsageDetails(out)
}
resp = cliproxyexecutor.Response{Payload: out, Headers: httpResp.Header.Clone()}
return resp, nil
}
// executeStreamWithServiceAccount handles streaming authentication using service account credentials.
func (e *GeminiVertexExecutor) executeStreamWithServiceAccount(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options, projectID, location string, saJSON []byte) (_ *cliproxyexecutor.StreamResult, err error) {
baseModel := thinking.ParseSuffix(req.Model).ModelName
reporter := helps.NewExecutorUsageReporter(ctx, e, baseModel, auth)
defer reporter.TrackFailure(ctx, &err)
from := opts.SourceFormat
responseFormat := cliproxyexecutor.ResponseFormatOrSource(opts)
to := sdktranslator.FromString("gemini")
originalPayloadSource := req.Payload
if len(opts.OriginalRequest) > 0 {
originalPayloadSource = opts.OriginalRequest
}
originalPayload := originalPayloadSource
originalTranslated := helps.TranslateRequestWithCodexMultiAgentV2(ctx, opts.Headers, e.cfg, from, to, baseModel, originalPayload, true)
body, err := helps.TranslateRequestReturningError(ctx, opts.Headers, e.cfg, from, to, baseModel, req.Payload, true, false)
if err != nil {
return nil, err
}
body, err = helps.ApplyRequestThinking(body, req, opts, from.String(), to.String(), e.Identifier())
if err != nil {
return nil, err
}
body = fixGeminiImageAspectRatio(baseModel, body)
requestedModel := helps.PayloadRequestedModel(opts, req.Model)
requestPath := helps.PayloadRequestPath(opts)
body = helps.SetStringIfDifferent(body, "model", baseModel)
body = helps.StripVertexOpenAIResponsesToolCallIDs(body, from.String())
body = internalsignature.SanitizeGeminiRequestThoughtSignatures(body, "contents")
action := getVertexAction(baseModel, true)
body = helps.DropGeminiTrailingThoughtOnlyContent(body, "contents")
body = helps.EnsureGeminiBoundaryUserContent(body, "contents")
baseURL := vertexBaseURL(location)
url := fmt.Sprintf("%s/%s/projects/%s/locations/%s/publishers/google/models/%s:%s", baseURL, vertexAPIVersion, projectID, location, baseModel, action)
// Imagen models don't support streaming, skip SSE params
if !isImagenModel(baseModel) {
if opts.Alt == "" {
url = url + "?alt=sse"
} else {
url = url + fmt.Sprintf("?$alt=%s", opts.Alt)
}
}
body, _ = sjson.DeleteBytes(body, "session_id")
reporter.SetTranslatedReasoningEffort(body, to.String())
body = helps.ApplyPayloadConfigWithRequest(e.cfg, baseModel, to.String(), from.String(), "", body, originalTranslated, requestedModel, requestPath, opts.Headers)
httpReq, errNewReq := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
if errNewReq != nil {
return nil, errNewReq
}
httpReq.Header.Set("Content-Type", "application/json")
if token, errTok := vertexAccessToken(ctx, e.cfg, auth, saJSON); errTok == nil && token != "" {
httpReq.Header.Set("Authorization", "Bearer "+token)
} else if errTok != nil {
log.Errorf("vertex executor: access token error: %v", errTok)
return nil, statusErr{code: 500, msg: "internal server error"}
}
applyGeminiHeaders(httpReq, auth, opts.Headers)
var attrs map[string]string
if auth != nil {
attrs = auth.Attributes
}
util.ApplyCustomHeadersFromAttrs(httpReq, attrs, opts.Headers)
var authID, authLabel, authType, authValue string
if auth != nil {
authID = auth.ID
authLabel = auth.Label
authType, authValue = auth.AccountInfo()
}
helps.RecordAPIRequest(ctx, e.cfg, helps.UpstreamRequestLog{
URL: url,
Method: http.MethodPost,
Headers: httpReq.Header.Clone(),
Body: body,
Provider: e.Identifier(),
AuthID: authID,
AuthLabel: authLabel,
AuthType: authType,
AuthValue: authValue,
})
httpClient := helps.NewProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
httpClient = reporter.TrackHTTPClient(httpClient)
httpResp, errDo := httpClient.Do(httpReq)
if errDo != nil {
helps.RecordAPIResponseError(ctx, e.cfg, errDo)
return nil, errDo
}
helps.RecordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
b, _ := io.ReadAll(httpResp.Body)
helps.AppendAPIResponseChunk(ctx, e.cfg, b)
helps.LogWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, helps.SummarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
if errClose := httpResp.Body.Close(); errClose != nil {
log.Errorf("vertex executor: close response body error: %v", errClose)
}
return nil, statusErr{code: httpResp.StatusCode, msg: string(b)}
}
out := make(chan cliproxyexecutor.StreamChunk, 1)
go func() {
defer close(out)
defer reporter.EnsurePublished(ctx)
defer func() {
if errClose := httpResp.Body.Close(); errClose != nil {
log.Errorf("vertex executor: close response body error: %v", errClose)
}
}()
scanner := bufio.NewScanner(httpResp.Body)
scanner.Buffer(nil, streamScannerBuffer)
claudeInputTokens := helps.NewClaudeInputTokenState(from, to, responseFormat, originalPayload)
var streamUsage helps.StreamUsageBuffer
defer streamUsage.Publish(ctx, reporter)
var streamErr error
defer func() {
_ = helps.StopTranslatedClaudeStream(ctx, &streamErr, reporter, &streamUsage, out)
}()
var terminalDelivered bool
var sawFinishReason bool
var param any
helps.InitializeApplyPatchStream(ctx, to, responseFormat, req.Model, helps.ApplyPatchOriginalRequest(req, opts), body, &param)
for scanner.Scan() {
line := scanner.Bytes()
helps.AppendAPIResponseChunk(ctx, e.cfg, line)
reporter.ObserveResponseModel(line)
if !sawFinishReason && geminiStreamHasFinishSignal(helps.JSONPayload(line)) {
sawFinishReason = true
}
if detail, ok := helps.ParseGeminiStreamUsage(line); ok {
streamUsage.Observe(detail, true)
}
lines := helps.TranslateStreamWithClaudeInputTokens(ctx, to, responseFormat, req.Model, helps.ApplyPatchOriginalRequest(req, opts), body, bytes.Clone(line), &param, claudeInputTokens)
helps.RecordApplyPatchStreamFailure(ctx, param, reporter, statusErr{code: http.StatusBadGateway, msg: helps.ApplyPatchUpstreamErrorMessage})
helps.CaptureTranslatedClaudeStreamError(responseFormat, lines, &streamErr, func(msg string) error {
return statusErr{code: http.StatusBadGateway, msg: msg}
})
for i := range lines {
select {
case out <- cliproxyexecutor.StreamChunk{Payload: lines[i]}:
if isGeminiVertexTerminalStreamChunk(lines[i]) {
terminalDelivered = true
}
case <-ctx.Done():
return
}
}
if helps.StopApplyPatchStream(ctx, param, reporter, out, statusErr{code: http.StatusBadGateway, msg: helps.ApplyPatchUpstreamErrorMessage}) {
return
}
}
if helps.EndApplyPatchStream(ctx, param, reporter, out, statusErr{code: http.StatusBadGateway, msg: helps.ApplyPatchUpstreamErrorMessage}) {
return
}
if helps.StopTranslatedClaudeStream(ctx, &streamErr, reporter, &streamUsage, out) {
return
}
if errScan := scanner.Err(); errScan != nil {
if terminalDelivered && errors.Is(errScan, context.Canceled) {
return
}
helps.RecordAPIResponseError(ctx, e.cfg, errScan)
reporter.PublishFailure(ctx, errScan)
select {
case out <- cliproxyexecutor.StreamChunk{Err: errScan}:
case <-ctx.Done():
}
return
}
if ctx.Err() != nil {
return
}
if responseFormat == sdktranslator.FormatClaude && !sawFinishReason {
truncErr := statusErr{code: http.StatusBadGateway, msg: "upstream stream closed before finishReason"}
helps.RecordAPIResponseError(ctx, e.cfg, truncErr)
if !streamUsage.PublishFailure(ctx, reporter, truncErr) {
reporter.PublishFailure(ctx, truncErr)
}
select {
case out <- cliproxyexecutor.StreamChunk{Err: truncErr}:
case <-ctx.Done():
}
return
}
lines := helps.TranslateStreamWithClaudeInputTokens(ctx, to, responseFormat, req.Model, helps.ApplyPatchOriginalRequest(req, opts), body, []byte("[DONE]"), &param, claudeInputTokens)
helps.RecordApplyPatchStreamFailure(ctx, param, reporter, statusErr{code: http.StatusBadGateway, msg: helps.ApplyPatchUpstreamErrorMessage})
helps.CaptureTranslatedClaudeStreamError(responseFormat, lines, &streamErr, func(msg string) error {
return statusErr{code: http.StatusBadGateway, msg: msg}
})
for i := range lines {
select {
case out <- cliproxyexecutor.StreamChunk{Payload: lines[i]}:
case <-ctx.Done():
return
}
}
if helps.StopApplyPatchStream(ctx, param, reporter, out, statusErr{code: http.StatusBadGateway, msg: helps.ApplyPatchUpstreamErrorMessage}) {
return
}
if helps.StopTranslatedClaudeStream(ctx, &streamErr, reporter, &streamUsage, out) {
return
}
}()
return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
}
// executeStreamWithAPIKey handles streaming authentication using API key credentials.
func (e *GeminiVertexExecutor) executeStreamWithAPIKey(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options, apiKey, baseURL string) (_ *cliproxyexecutor.StreamResult, err error) {
baseModel := thinking.ParseSuffix(req.Model).ModelName
reporter := helps.NewExecutorUsageReporter(ctx, e, baseModel, auth)
defer reporter.TrackFailure(ctx, &err)
from := opts.SourceFormat
responseFormat := cliproxyexecutor.ResponseFormatOrSource(opts)
to := sdktranslator.FromString("gemini")
originalPayloadSource := req.Payload
if len(opts.OriginalRequest) > 0 {
originalPayloadSource = opts.OriginalRequest
}
originalPayload := originalPayloadSource
originalTranslated := helps.TranslateRequestWithCodexMultiAgentV2(ctx, opts.Headers, e.cfg, from, to, baseModel, originalPayload, true)
body, err := helps.TranslateRequestReturningError(ctx, opts.Headers, e.cfg, from, to, baseModel, req.Payload, true, false)
if err != nil {
return nil, err
}
body, err = helps.ApplyRequestThinking(body, req, opts, from.String(), to.String(), e.Identifier())
if err != nil {
return nil, err
}
body = fixGeminiImageAspectRatio(baseModel, body)
requestedModel := helps.PayloadRequestedModel(opts, req.Model)
requestPath := helps.PayloadRequestPath(opts)
body = helps.SetStringIfDifferent(body, "model", baseModel)
body = helps.StripVertexOpenAIResponsesToolCallIDs(body, from.String())
body = internalsignature.SanitizeGeminiRequestThoughtSignatures(body, "contents")
action := getVertexAction(baseModel, true)
body = helps.DropGeminiTrailingThoughtOnlyContent(body, "contents")
body = helps.EnsureGeminiBoundaryUserContent(body, "contents")
// For API key auth, use simpler URL format without project/location
url := vertexAPIKeyModelsURL(baseURL, baseModel, action)
// Imagen models don't support streaming, skip SSE params
if !isImagenModel(baseModel) {
if opts.Alt == "" {
url = url + "?alt=sse"
} else {
url = url + fmt.Sprintf("?$alt=%s", opts.Alt)
}
}
body, _ = sjson.DeleteBytes(body, "session_id")
reporter.SetTranslatedReasoningEffort(body, to.String())
body = helps.ApplyPayloadConfigWithRequest(e.cfg, baseModel, to.String(), from.String(), "", body, originalTranslated, requestedModel, requestPath, opts.Headers)
httpReq, errNewReq := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
if errNewReq != nil {
return nil, errNewReq
}
httpReq.Header.Set("Content-Type", "application/json")
if apiKey != "" {
httpReq.Header.Set("x-goog-api-key", apiKey)
}
applyGeminiHeaders(httpReq, auth, opts.Headers)
var attrs map[string]string
if auth != nil {
attrs = auth.Attributes
}
util.ApplyCustomHeadersFromAttrs(httpReq, attrs, opts.Headers)
var authID, authLabel, authType, authValue string
if auth != nil {
authID = auth.ID
authLabel = auth.Label
authType, authValue = auth.AccountInfo()
}
helps.RecordAPIRequest(ctx, e.cfg, helps.UpstreamRequestLog{
URL: url,
Method: http.MethodPost,
Headers: httpReq.Header.Clone(),
Body: body,
Provider: e.Identifier(),
AuthID: authID,
AuthLabel: authLabel,
AuthType: authType,
AuthValue: authValue,
})
httpClient := helps.NewProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
httpClient = reporter.TrackHTTPClient(httpClient)
httpResp, errDo := httpClient.Do(httpReq)
if errDo != nil {
helps.RecordAPIResponseError(ctx, e.cfg, errDo)
return nil, errDo
}
helps.RecordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
b, _ := io.ReadAll(httpResp.Body)
helps.AppendAPIResponseChunk(ctx, e.cfg, b)
helps.LogWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, helps.SummarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
if errClose := httpResp.Body.Close(); errClose != nil {
log.Errorf("vertex executor: close response body error: %v", errClose)
}
return nil, statusErr{code: httpResp.StatusCode, msg: string(b)}
}
out := make(chan cliproxyexecutor.StreamChunk, 1)
go func() {
defer close(out)
defer reporter.EnsurePublished(ctx)
defer func() {
if errClose := httpResp.Body.Close(); errClose != nil {
log.Errorf("vertex executor: close response body error: %v", errClose)
}
}()
scanner := bufio.NewScanner(httpResp.Body)
scanner.Buffer(nil, streamScannerBuffer)
claudeInputTokens := helps.NewClaudeInputTokenState(from, to, responseFormat, originalPayload)
var streamUsage helps.StreamUsageBuffer
defer streamUsage.Publish(ctx, reporter)
var streamErr error
defer func() {
_ = helps.StopTranslatedClaudeStream(ctx, &streamErr, reporter, &streamUsage, out)
}()
var terminalDelivered bool
var sawFinishReason bool
var param any
helps.InitializeApplyPatchStream(ctx, to, responseFormat, req.Model, helps.ApplyPatchOriginalRequest(req, opts), body, &param)
for scanner.Scan() {
line := scanner.Bytes()
helps.AppendAPIResponseChunk(ctx, e.cfg, line)
reporter.ObserveResponseModel(line)
if !sawFinishReason && geminiStreamHasFinishSignal(helps.JSONPayload(line)) {
sawFinishReason = true
}
if detail, ok := helps.ParseGeminiStreamUsage(line); ok {
streamUsage.Observe(detail, true)
}
lines := helps.TranslateStreamWithClaudeInputTokens(ctx, to, responseFormat, req.Model, helps.ApplyPatchOriginalRequest(req, opts), body, bytes.Clone(line), &param, claudeInputTokens)
helps.RecordApplyPatchStreamFailure(ctx, param, reporter, statusErr{code: http.StatusBadGateway, msg: helps.ApplyPatchUpstreamErrorMessage})
helps.CaptureTranslatedClaudeStreamError(responseFormat, lines, &streamErr, func(msg string) error {
return statusErr{code: http.StatusBadGateway, msg: msg}
})
for i := range lines {
select {
case out <- cliproxyexecutor.StreamChunk{Payload: lines[i]}:
if isGeminiVertexTerminalStreamChunk(lines[i]) {
terminalDelivered = true
}
case <-ctx.Done():
return
}
}
if helps.StopApplyPatchStream(ctx, param, reporter, out, statusErr{code: http.StatusBadGateway, msg: helps.ApplyPatchUpstreamErrorMessage}) {
return
}
}
if helps.EndApplyPatchStream(ctx, param, reporter, out, statusErr{code: http.StatusBadGateway, msg: helps.ApplyPatchUpstreamErrorMessage}) {
return
}
if helps.StopTranslatedClaudeStream(ctx, &streamErr, reporter, &streamUsage, out) {
return
}
if errScan := scanner.Err(); errScan != nil {
if terminalDelivered && errors.Is(errScan, context.Canceled) {
return
}
helps.RecordAPIResponseError(ctx, e.cfg, errScan)
reporter.PublishFailure(ctx, errScan)
select {
case out <- cliproxyexecutor.StreamChunk{Err: errScan}:
case <-ctx.Done():
}
return
}
if ctx.Err() != nil {
return
}
if responseFormat == sdktranslator.FormatClaude && !sawFinishReason {
truncErr := statusErr{code: http.StatusBadGateway, msg: "upstream stream closed before finishReason"}
helps.RecordAPIResponseError(ctx, e.cfg, truncErr)
if !streamUsage.PublishFailure(ctx, reporter, truncErr) {
reporter.PublishFailure(ctx, truncErr)
}
select {
case out <- cliproxyexecutor.StreamChunk{Err: truncErr}:
case <-ctx.Done():
}
return
}
lines := helps.TranslateStreamWithClaudeInputTokens(ctx, to, responseFormat, req.Model, helps.ApplyPatchOriginalRequest(req, opts), body, []byte("[DONE]"), &param, claudeInputTokens)
helps.RecordApplyPatchStreamFailure(ctx, param, reporter, statusErr{code: http.StatusBadGateway, msg: helps.ApplyPatchUpstreamErrorMessage})
helps.CaptureTranslatedClaudeStreamError(responseFormat, lines, &streamErr, func(msg string) error {
return statusErr{code: http.StatusBadGateway, msg: msg}
})
for i := range lines {
select {
case out <- cliproxyexecutor.StreamChunk{Payload: lines[i]}:
case <-ctx.Done():
return
}
}
if helps.StopApplyPatchStream(ctx, param, reporter, out, statusErr{code: http.StatusBadGateway, msg: helps.ApplyPatchUpstreamErrorMessage}) {
return
}
if helps.StopTranslatedClaudeStream(ctx, &streamErr, reporter, &streamUsage, out) {
return
}
}()
return &cliproxyexecutor.StreamResult{Headers: httpResp.Header.Clone(), Chunks: out}, nil
}
// countTokensWithServiceAccount counts tokens using service account credentials.
func (e *GeminiVertexExecutor) countTokensWithServiceAccount(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options, projectID, location string, saJSON []byte) (cliproxyexecutor.Response, error) {
baseModel := thinking.ParseSuffix(req.Model).ModelName
from := opts.SourceFormat
responseFormat := cliproxyexecutor.ResponseFormatOrSource(opts)
to := sdktranslator.FromString("gemini")
translatedReq, err := helps.TranslateRequestReturningError(ctx, opts.Headers, e.cfg, from, to, baseModel, req.Payload, false, false)
if err != nil {
return cliproxyexecutor.Response{}, err
}
originalTranslatedForPayload := append([]byte(nil), translatedReq...)
if len(opts.OriginalRequest) > 0 {
originalTranslatedForPayload = helps.TranslateRequestWithCodexMultiAgentV2(ctx, opts.Headers, e.cfg, from, to, baseModel, opts.OriginalRequest, false)
}
translatedReq, err = helps.ApplyRequestThinking(translatedReq, req, opts, from.String(), to.String(), e.Identifier())
if err != nil {
return cliproxyexecutor.Response{}, err
}
translatedReq = fixGeminiImageAspectRatio(baseModel, translatedReq)
translatedReq, _ = sjson.SetBytes(translatedReq, "model", baseModel)
translatedReq = helps.StripVertexOpenAIResponsesToolCallIDs(translatedReq, from.String())
respCtx := context.WithValue(ctx, "alt", opts.Alt)
translatedReq, _ = sjson.DeleteBytes(translatedReq, "tools")
translatedReq, _ = sjson.DeleteBytes(translatedReq, "generationConfig")
translatedReq, _ = sjson.DeleteBytes(translatedReq, "safetySettings")
translatedReq = internalsignature.SanitizeGeminiRequestThoughtSignatures(translatedReq, "contents")
translatedReq = helps.EnsureGeminiLeadingUserContent(translatedReq, "contents")
baseURL := vertexBaseURL(location)
url := fmt.Sprintf("%s/%s/projects/%s/locations/%s/publishers/google/models/%s:%s", baseURL, vertexAPIVersion, projectID, location, baseModel, "countTokens")
translatedReq = helps.NewPayloadFinalizer(e.cfg, e.Identifier(), baseModel, to.String(), "", originalTranslatedForPayload, req, opts)(translatedReq)
httpReq, errNewReq := http.NewRequestWithContext(respCtx, http.MethodPost, url, bytes.NewReader(translatedReq))
if errNewReq != nil {
return cliproxyexecutor.Response{}, errNewReq
}
httpReq.Header.Set("Content-Type", "application/json")
if token, errTok := vertexAccessToken(ctx, e.cfg, auth, saJSON); errTok == nil && token != "" {
httpReq.Header.Set("Authorization", "Bearer "+token)
} else if errTok != nil {
log.Errorf("vertex executor: access token error: %v", errTok)
return cliproxyexecutor.Response{}, statusErr{code: 500, msg: "internal server error"}
}
applyGeminiHeaders(httpReq, auth, opts.Headers)
var attrs map[string]string
if auth != nil {
attrs = auth.Attributes
}
util.ApplyCustomHeadersFromAttrs(httpReq, attrs, opts.Headers)
var authID, authLabel, authType, authValue string
if auth != nil {
authID = auth.ID
authLabel = auth.Label
authType, authValue = auth.AccountInfo()
}
helps.RecordAPIRequest(ctx, e.cfg, helps.UpstreamRequestLog{
URL: url,
Method: http.MethodPost,
Headers: httpReq.Header.Clone(),
Body: translatedReq,
Provider: e.Identifier(),
AuthID: authID,
AuthLabel: authLabel,
AuthType: authType,
AuthValue: authValue,
})
httpClient := helps.NewProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
cliproxyexecutor.MarkUpstreamAttempt(ctx)
httpResp, errDo := httpClient.Do(httpReq)
if errDo != nil {
helps.RecordAPIResponseError(ctx, e.cfg, errDo)
return cliproxyexecutor.Response{}, errDo
}
defer func() {
if errClose := httpResp.Body.Close(); errClose != nil {
log.Errorf("vertex executor: close response body error: %v", errClose)
}
}()
helps.RecordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
b, _ := io.ReadAll(httpResp.Body)
helps.AppendAPIResponseChunk(ctx, e.cfg, b)
helps.LogWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, helps.SummarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
return cliproxyexecutor.Response{}, statusErr{code: httpResp.StatusCode, msg: string(b)}
}
data, errRead := io.ReadAll(httpResp.Body)
if errRead != nil {
helps.RecordAPIResponseError(ctx, e.cfg, errRead)
return cliproxyexecutor.Response{}, errRead
}
helps.AppendAPIResponseChunk(ctx, e.cfg, data)
count := gjson.GetBytes(data, "totalTokens").Int()
out := sdktranslator.TranslateTokenCount(ctx, to, responseFormat, count, data)
return cliproxyexecutor.Response{Payload: out, Headers: httpResp.Header.Clone()}, nil
}
// countTokensWithAPIKey handles token counting using API key credentials.
func (e *GeminiVertexExecutor) countTokensWithAPIKey(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options, apiKey, baseURL string) (cliproxyexecutor.Response, error) {
baseModel := thinking.ParseSuffix(req.Model).ModelName
from := opts.SourceFormat
responseFormat := cliproxyexecutor.ResponseFormatOrSource(opts)
to := sdktranslator.FromString("gemini")
translatedReq, err := helps.TranslateRequestReturningError(ctx, opts.Headers, e.cfg, from, to, baseModel, req.Payload, false, false)
if err != nil {
return cliproxyexecutor.Response{}, err
}
originalTranslatedForPayload := append([]byte(nil), translatedReq...)
if len(opts.OriginalRequest) > 0 {
originalTranslatedForPayload = helps.TranslateRequestWithCodexMultiAgentV2(ctx, opts.Headers, e.cfg, from, to, baseModel, opts.OriginalRequest, false)
}
translatedReq, err = helps.ApplyRequestThinking(translatedReq, req, opts, from.String(), to.String(), e.Identifier())
if err != nil {
return cliproxyexecutor.Response{}, err
}
translatedReq = fixGeminiImageAspectRatio(baseModel, translatedReq)
translatedReq, _ = sjson.SetBytes(translatedReq, "model", baseModel)
translatedReq = helps.StripVertexOpenAIResponsesToolCallIDs(translatedReq, from.String())
respCtx := context.WithValue(ctx, "alt", opts.Alt)
translatedReq, _ = sjson.DeleteBytes(translatedReq, "tools")
translatedReq, _ = sjson.DeleteBytes(translatedReq, "generationConfig")
translatedReq, _ = sjson.DeleteBytes(translatedReq, "safetySettings")
translatedReq = internalsignature.SanitizeGeminiRequestThoughtSignatures(translatedReq, "contents")
translatedReq = helps.EnsureGeminiLeadingUserContent(translatedReq, "contents")
// For API key auth, use simpler URL format without project/location
url := vertexAPIKeyModelsURL(baseURL, baseModel, "countTokens")
translatedReq = helps.NewPayloadFinalizer(e.cfg, e.Identifier(), baseModel, to.String(), "", originalTranslatedForPayload, req, opts)(translatedReq)
httpReq, errNewReq := http.NewRequestWithContext(respCtx, http.MethodPost, url, bytes.NewReader(translatedReq))
if errNewReq != nil {
return cliproxyexecutor.Response{}, errNewReq
}
httpReq.Header.Set("Content-Type", "application/json")
if apiKey != "" {
httpReq.Header.Set("x-goog-api-key", apiKey)
}
applyGeminiHeaders(httpReq, auth, opts.Headers)
var attrs map[string]string
if auth != nil {
attrs = auth.Attributes
}
util.ApplyCustomHeadersFromAttrs(httpReq, attrs, opts.Headers)
var authID, authLabel, authType, authValue string
if auth != nil {
authID = auth.ID
authLabel = auth.Label
authType, authValue = auth.AccountInfo()
}
helps.RecordAPIRequest(ctx, e.cfg, helps.UpstreamRequestLog{
URL: url,
Method: http.MethodPost,
Headers: httpReq.Header.Clone(),
Body: translatedReq,
Provider: e.Identifier(),
AuthID: authID,
AuthLabel: authLabel,
AuthType: authType,
AuthValue: authValue,
})
httpClient := helps.NewProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
cliproxyexecutor.MarkUpstreamAttempt(ctx)
httpResp, errDo := httpClient.Do(httpReq)
if errDo != nil {
helps.RecordAPIResponseError(ctx, e.cfg, errDo)
return cliproxyexecutor.Response{}, errDo
}
defer func() {
if errClose := httpResp.Body.Close(); errClose != nil {
log.Errorf("vertex executor: close response body error: %v", errClose)
}
}()
helps.RecordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
b, _ := io.ReadAll(httpResp.Body)
helps.AppendAPIResponseChunk(ctx, e.cfg, b)
helps.LogWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, helps.SummarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
return cliproxyexecutor.Response{}, statusErr{code: httpResp.StatusCode, msg: string(b)}
}
data, errRead := io.ReadAll(httpResp.Body)
if errRead != nil {
helps.RecordAPIResponseError(ctx, e.cfg, errRead)
return cliproxyexecutor.Response{}, errRead
}
helps.AppendAPIResponseChunk(ctx, e.cfg, data)
count := gjson.GetBytes(data, "totalTokens").Int()
out := sdktranslator.TranslateTokenCount(ctx, to, responseFormat, count, data)
return cliproxyexecutor.Response{Payload: out, Headers: httpResp.Header.Clone()}, nil
}
// vertexCreds extracts project, location and raw service account JSON from auth metadata.
func vertexCreds(a *cliproxyauth.Auth) (projectID, location string, serviceAccountJSON []byte, err error) {
if a == nil || a.Metadata == nil {
return "", "", nil, fmt.Errorf("vertex executor: missing auth metadata")
}
if v, ok := a.Metadata["project_id"].(string); ok {
projectID = strings.TrimSpace(v)
}
if projectID == "" {
// Some service accounts may use "project"; still prefer standard field
if v, ok := a.Metadata["project"].(string); ok {
projectID = strings.TrimSpace(v)
}
}
if projectID == "" {
return "", "", nil, fmt.Errorf("vertex executor: missing project_id in credentials")
}
if v, ok := a.Metadata["location"].(string); ok && strings.TrimSpace(v) != "" {
location = strings.TrimSpace(v)
} else {
location = "us-central1"
}
var sa map[string]any
if raw, ok := a.Metadata["service_account"].(map[string]any); ok {
sa = raw
}
if sa == nil {
return "", "", nil, fmt.Errorf("vertex executor: missing service_account in credentials")
}
normalized, errNorm := vertexauth.NormalizeServiceAccountMap(sa)
if errNorm != nil {
return "", "", nil, fmt.Errorf("vertex executor: %w", errNorm)
}
saJSON, errMarshal := json.Marshal(normalized)
if errMarshal != nil {
return "", "", nil, fmt.Errorf("vertex executor: marshal service_account failed: %w", errMarshal)
}
return projectID, location, saJSON, nil
}
// vertexAPICreds extracts API key and base URL from auth attributes following the claudeCreds pattern.
func vertexAPICreds(a *cliproxyauth.Auth) (apiKey, baseURL string) {
if a == nil {
return "", ""
}
if a.Attributes != nil {
apiKey = a.Attributes["api_key"]
baseURL = a.Attributes["base_url"]
}
if apiKey == "" && a.Metadata != nil {
if v, ok := a.Metadata["access_token"].(string); ok {
apiKey = v
}
}
return
}
// vertexAPIKeyModelsURL builds the API-key models endpoint URL.
//
// The configured base URL is treated as a full resource prefix when it
// already contains a "/publishers/" segment (e.g. a real Vertex endpoint
// such as "https://aiplatform.googleapis.com/v1/projects/{p}/locations/{l}/publishers/google/models"),
// in which case only "/{model}:{action}" is appended. Otherwise the default
// "/v1/publishers/google/models" segment is inserted so short third-party
// hosts keep working; if the base already ends in a version segment such
// as "/v1" or "/v1beta1", only "/publishers/google/models" is appended to
// avoid producing a duplicated version path.
func vertexAPIKeyModelsURL(baseURL, model, action string) string {
base := strings.TrimRight(strings.TrimSpace(baseURL), "/")
if base == "" {
base = "https://aiplatform.googleapis.com"
}
if strings.Contains(base, "/publishers/") {
return fmt.Sprintf("%s/%s:%s", base, model, action)
}
if !vertexAPIVersionSuffix(base) {
base = fmt.Sprintf("%s/%s", base, vertexAPIVersion)
}
return fmt.Sprintf("%s/publishers/google/models/%s:%s", base, model, action)
}
// vertexAPIVersionSuffix reports whether the trimmed base URL already ends
// with a Google API version segment like "/v1" or "/v1beta1".
func vertexAPIVersionSuffix(base string) bool {
if base == "" {
return false
}
seg := base[strings.LastIndexByte(base, '/')+1:]
if len(seg) < 2 || seg[0] != 'v' {
return false
}
hasDigit := false
for i := 1; i < len(seg); i++ {
c := seg[i]
switch {
case c >= '0' && c <= '9':
hasDigit = true
case c == '_' || c == '-' || (c >= 'a' && c <= 'z'):
default:
return false
}
}
return hasDigit
}
func vertexBaseURL(location string) string {
loc := strings.TrimSpace(location)
if loc == "" {
loc = "us-central1"
} else if loc == "global" {
return "https://aiplatform.googleapis.com"
}
return fmt.Sprintf("https://%s-aiplatform.googleapis.com", loc)
}
func vertexAccessToken(ctx context.Context, cfg *config.Config, auth *cliproxyauth.Auth, saJSON []byte) (string, error) {
// Token exchange stays on the credential or global proxy. The execution proxy
// applies only to the model request that follows.
tokenCtx := cliproxyexecutor.WithoutRequestProxyURL(ctx)
if httpClient := helps.NewProxyAwareHTTPClient(tokenCtx, cfg, auth, 0); httpClient != nil {
ctx = context.WithValue(ctx, oauth2.HTTPClient, httpClient)
}
// Use cloud-platform scope for Vertex AI.
creds, errCreds := google.CredentialsFromJSON(ctx, saJSON, "https://www.googleapis.com/auth/cloud-platform")
if errCreds != nil {
return "", fmt.Errorf("vertex executor: parse service account json failed: %w", errCreds)
}
tok, errTok := creds.TokenSource.Token()
if errTok != nil {
return "", fmt.Errorf("vertex executor: get access token failed: %w", errTok)
}
return tok.AccessToken, nil
}
// resolveVertexConfig finds the matching vertex-api-key configuration entry for the given auth.
func (e *GeminiVertexExecutor) resolveVertexConfig(auth *cliproxyauth.Auth) *config.VertexCompatKey {
if auth == nil || e.cfg == nil {
return nil
}
var attrKey, attrBase string
if auth.Attributes != nil {
attrKey = strings.TrimSpace(auth.Attributes["api_key"])
attrBase = strings.TrimSpace(auth.Attributes["base_url"])
}
for i := range e.cfg.VertexCompatAPIKey {
entry := &e.cfg.VertexCompatAPIKey[i]
cfgKey := strings.TrimSpace(entry.APIKey)
cfgBase := strings.TrimSpace(entry.BaseURL)
if attrKey != "" && attrBase != "" {
if strings.EqualFold(cfgKey, attrKey) && strings.EqualFold(cfgBase, attrBase) {
return entry
}
continue
}
if attrKey != "" && strings.EqualFold(cfgKey, attrKey) {
if cfgBase == "" || strings.EqualFold(cfgBase, attrBase) {
return entry
}
}
if attrKey == "" && attrBase != "" && strings.EqualFold(cfgBase, attrBase) {
return entry
}
}
if attrKey != "" {
for i := range e.cfg.VertexCompatAPIKey {
entry := &e.cfg.VertexCompatAPIKey[i]
if strings.EqualFold(strings.TrimSpace(entry.APIKey), attrKey) {
return entry
}
}
}
return nil
}
// SupportsApplyPatch reports the actual executor contract, independent of its provider name.
func (e *GeminiVertexExecutor) SupportsApplyPatch() bool { return e != nil }
func shouldExecuteVertexInteractions(auth *cliproxyauth.Auth, opts cliproxyexecutor.Options) bool {
return opts.SourceFormat == sdktranslator.FormatInteractions && isNativeVertexInteractionsAuth(auth)
}
func isNativeVertexInteractionsAuth(auth *cliproxyauth.Auth) bool {
if auth == nil {
return false
}
if auth.Attributes != nil {
if v := strings.TrimSpace(auth.Attributes["interactions"]); strings.EqualFold(v, "true") || strings.EqualFold(v, "1") {
return true
}
}
if auth.Metadata != nil {
if v, ok := auth.Metadata["interactions"].(bool); ok && v {
return true
}
if v, ok := auth.Metadata["native_interactions"].(bool); ok && v {
return true
}
}
return false
}
func vertexInteractionsURL(baseURL, projectID string, isStream bool) string {
base := strings.TrimSpace(baseURL)
if base == "" {
base = "https://aiplatform.googleapis.com"
}
base = strings.TrimRight(base, "/")
var url string
if strings.TrimSpace(projectID) != "" {
url = fmt.Sprintf("%s/v1beta1/projects/%s/locations/global/interactions", base, strings.TrimSpace(projectID))
} else {
url = fmt.Sprintf("%s/v1beta1/interactions", base)
}
if isStream {
url = url + "?alt=sse"
}
return url
}
func (e *GeminiVertexExecutor) executeInteractions(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
targetName := thinking.ParseSuffix(req.Model).ModelName
reporter := helps.NewExecutorUsageReporter(ctx, e, targetName, auth)
defer reporter.TrackFailure(ctx, &err)
isCompat := helps.APIKeyModelIsCompat(req)
originalTranslated, body, errTranslate := helps.TranslateGeminiInteractionsRequestPair(ctx, e.cfg, targetName, req.Payload, opts, false, isCompat)
if errTranslate != nil {
return resp, errTranslate
}
if gjson.GetBytes(body, "model").Exists() && targetName != "" {
body = helps.SetStringIfDifferent(body, "model", targetName)
}
var errThinking error
body, errThinking = helps.ApplyGeminiInteractionsThinking(body, req, opts)
if errThinking != nil {
return resp, errThinking
}
requestedModel := helps.PayloadRequestedModel(opts, req.Model)
requestPath := helps.PayloadRequestPath(opts)
fromProtocol := opts.SourceFormat.String()
body = helps.SanitizeGeminiInteractionsUnsupportedInputIDs(body)
apiKey, baseURL := vertexAPICreds(auth)
var projectID string
var saJSON []byte
if apiKey == "" {
var errCreds error
projectID, _, saJSON, errCreds = vertexCreds(auth)
if errCreds != nil {
return resp, errCreds
}
if auth != nil && auth.Attributes != nil && auth.Attributes["base_url"] != "" {
baseURL = strings.TrimSpace(auth.Attributes["base_url"])
}
} else if auth != nil && auth.Metadata != nil {
if pid, ok := auth.Metadata["project_id"].(string); ok {
projectID = pid
}
}
url := vertexInteractionsURL(baseURL, projectID, false)
body = helps.ApplyPayloadConfigWithRequest(e.cfg, targetName, "interactions", fromProtocol, "", body, originalTranslated, requestedModel, requestPath, opts.Headers)
httpReq, errRequest := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
if errRequest != nil {
return resp, errRequest
}
httpReq.Header.Set("Content-Type", "application/json")
if apiKey != "" {
httpReq.Header.Set("x-goog-api-key", apiKey)
} else {
token, errTok := vertexAccessToken(ctx, e.cfg, auth, saJSON)
if errTok != nil {
log.Errorf("vertex executor: interactions access token error: %v", errTok)
return resp, statusErr{code: 500, msg: "internal server error"}
}
if strings.TrimSpace(token) == "" {
return resp, statusErr{code: http.StatusUnauthorized, msg: "missing access token"}
}
httpReq.Header.Set("Authorization", "Bearer "+token)
}
applyGeminiHeaders(httpReq, auth, opts.Headers)
helps.ApplyGeminiInteractionsRequestHeaders(httpReq, opts.Headers)
helps.ApplyGeminiInteractionsRevisionHeader(httpReq)
var attrs map[string]string
if auth != nil {
attrs = auth.Attributes
}
util.ApplyCustomHeadersFromAttrs(httpReq, attrs, opts.Headers)
var authID, authLabel, authType, authValue string
if auth != nil {
authID = auth.ID
authLabel = auth.Label
authType, authValue = auth.AccountInfo()
}
helps.RecordAPIRequest(ctx, e.cfg, helps.UpstreamRequestLog{
URL: url,
Method: http.MethodPost,
Headers: httpReq.Header.Clone(),
Body: body,
Provider: e.Identifier(),
AuthID: authID,
AuthLabel: authLabel,
AuthType: authType,
AuthValue: authValue,
})
httpClient := reporter.TrackHTTPClient(helps.NewProxyAwareHTTPClient(ctx, e.cfg, auth, 0))
httpResp, errDo := httpClient.Do(httpReq)
if errDo != nil {
helps.RecordAPIResponseError(ctx, e.cfg, errDo)
return resp, errDo
}
defer func() {
if errClose := httpResp.Body.Close(); errClose != nil {
log.Errorf("vertex executor: close interactions response body error: %v", errClose)
}
}()
helps.RecordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
data, errRead := io.ReadAll(httpResp.Body)
if errRead != nil {
helps.RecordAPIResponseError(ctx, e.cfg, errRead)
return resp, errRead
}
helps.AppendAPIResponseChunk(ctx, e.cfg, data)
if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
helps.LogWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, helps.SummarizeErrorBody(httpResp.Header.Get("Content-Type"), data))
err = statusErr{code: httpResp.StatusCode, msg: string(data)}
return resp, err
}
reporter.ObserveResponseModel(data)
targetFormat := cliproxyexecutor.ResponseFormatOrSource(opts)
var param any
out := sdktranslator.TranslateNonStream(ctx, sdktranslator.FormatInteractions, targetFormat, req.Model, helps.ApplyPatchOriginalRequest(req, opts), body, data, &param)
if helps.ApplyPatchTranslationError(param) != nil || len(out) == 0 {
return cliproxyexecutor.Response{}, statusErr{code: http.StatusBadGateway, msg: helps.ApplyPatchUpstreamErrorMessage}
}
reporter.Publish(ctx, helps.ParseInteractionsUsage(data))
if targetFormat == sdktranslator.FormatOpenAIResponse {
out = helps.EnsureResponsesUsageDetails(out)
}
return cliproxyexecutor.Response{Payload: out, Headers: httpResp.Header.Clone()}, nil
}
func (e *GeminiVertexExecutor) executeInteractionsStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (_ *cliproxyexecutor.StreamResult, err error) {
targetName := thinking.ParseSuffix(req.Model).ModelName
reporter := helps.NewExecutorUsageReporter(ctx, e, targetName, auth)
defer reporter.TrackFailure(ctx, &err)
isCompat := helps.APIKeyModelIsCompat(req)
originalTranslated, body, errTranslate := helps.TranslateGeminiInteractionsRequestPair(ctx, e.cfg, targetName, req.Payload, opts, true, isCompat)
if errTranslate != nil {
return nil, errTranslate
}
if gjson.GetBytes(body, "model").Exists() && targetName != "" {
body = helps.SetStringIfDifferent(body, "model", targetName)
}
var errThinking error
body, errThinking = helps.ApplyGeminiInteractionsThinking(body, req, opts)
if errThinking != nil {
return nil, errThinking
}
requestedModel := helps.PayloadRequestedModel(opts, req.Model)
requestPath := helps.PayloadRequestPath(opts)
fromProtocol := opts.SourceFormat.String()
body = helps.SanitizeGeminiInteractionsUnsupportedInputIDs(body)
body = helps.SetBoolIfDifferent(body, "stream", true)
apiKey, baseURL := vertexAPICreds(auth)
var projectID string
var saJSON []byte
if apiKey == "" {
var errCreds error
projectID, _, saJSON, errCreds = vertexCreds(auth)
if errCreds != nil {
return nil, errCreds
}
if auth != nil && auth.Attributes != nil && auth.Attributes["base_url"] != "" {
baseURL = strings.TrimSpace(auth.Attributes["base_url"])
}
} else if auth != nil && auth.Metadata != nil {
if pid, ok := auth.Metadata["project_id"].(string); ok {
projectID = pid
}
}
url := vertexInteractionsURL(baseURL, projectID, true)
body = helps.ApplyPayloadConfigWithRequest(e.cfg, targetName, "interactions", fromProtocol, "", body, originalTranslated, requestedModel, requestPath, opts.Headers)
httpReq, errRequest := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
if errRequest != nil {
return nil, errRequest
}
httpReq.Header.Set("Content-Type", "application/json")
if apiKey != "" {
httpReq.Header.Set("x-goog-api-key", apiKey)
} else {
token, errTok := vertexAccessToken(ctx, e.cfg, auth, saJSON)
if errTok != nil {
log.Errorf("vertex executor: interactions access token error: %v", errTok)
return nil, statusErr{code: 500, msg: "internal server error"}
}
if strings.TrimSpace(token) == "" {
return nil, statusErr{code: http.StatusUnauthorized, msg: "missing access token"}
}
httpReq.Header.Set("Authorization", "Bearer "+token)
}
applyGeminiHeaders(httpReq, auth, opts.Headers)
helps.ApplyGeminiInteractionsRequestHeaders(httpReq, opts.Headers)
helps.ApplyGeminiInteractionsRevisionHeader(httpReq)
var attrs map[string]string
if auth != nil {
attrs = auth.Attributes
}
util.ApplyCustomHeadersFromAttrs(httpReq, attrs, opts.Headers)
var authID, authLabel, authType, authValue string
if auth != nil {
authID = auth.ID
authLabel = auth.Label
authType, authValue = auth.AccountInfo()
}
helps.RecordAPIRequest(ctx, e.cfg, helps.UpstreamRequestLog{
URL: url,
Method: http.MethodPost,
Headers: httpReq.Header.Clone(),
Body: body,
Provider: e.Identifier(),
AuthID: authID,
AuthLabel: authLabel,
AuthType: authType,
AuthValue: authValue,
})
httpClient := reporter.TrackHTTPClient(helps.NewProxyAwareHTTPClient(ctx, e.cfg, auth, 0))
httpResp, errDo := httpClient.Do(httpReq)
if errDo != nil {
helps.RecordAPIResponseError(ctx, e.cfg, errDo)
return nil, errDo
}
helps.RecordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
data, _ := io.ReadAll(httpResp.Body)
if errClose := httpResp.Body.Close(); errClose != nil {
log.Errorf("vertex executor: close interactions error response body error: %v", errClose)
}
helps.AppendAPIResponseChunk(ctx, e.cfg, data)
return nil, statusErr{code: httpResp.StatusCode, msg: string(data)}
}
out := make(chan cliproxyexecutor.StreamChunk)
responseFormat := cliproxyexecutor.ResponseFormatOrSource(opts)
go func() {
defer close(out)
defer reporter.EnsurePublished(ctx)
defer func() {
if errClose := httpResp.Body.Close(); errClose != nil {
log.Errorf("vertex executor: close interactions stream body error: %v", errClose)
}
}()
scanner := bufio.NewScanner(httpResp.Body)
scanner.Buffer(nil, helps.StreamScannerBuffer)
originalRequest := opts.OriginalRequest
if len(originalRequest) == 0 {
originalRequest = req.Payload
}
claudeInputTokens := helps.NewClaudeInputTokenState(opts.SourceFormat, sdktranslator.FormatInteractions, responseFormat, originalRequest)
var streamUsage helps.StreamUsageBuffer
defer streamUsage.Publish(ctx, reporter)
var param any
helps.InitializeApplyPatchStream(ctx, sdktranslator.FormatInteractions, responseFormat, req.Model, helps.ApplyPatchOriginalRequest(req, opts), body, &param)
var frame []byte
emitFrame := func() bool {
rawFrame := bytes.Clone(frame)
trimmed := bytes.TrimSpace(rawFrame)
frame = frame[:0]
if len(trimmed) == 0 {
return true
}
payload := helps.GeminiInteractionsSSEPayload(rawFrame)
if len(payload) == 0 && helps.GeminiInteractionsSSEDone(rawFrame) {
payload = []byte("[DONE]")
}
if len(payload) > 0 {
reporter.ObserveResponseModel(payload)
if detail, ok := helps.ParseInteractionsStreamUsage(payload); ok {
streamUsage.Observe(detail, true)
}
}
if responseFormat == sdktranslator.FormatInteractions {
visibleFrame := append(bytes.TrimRight(rawFrame, "\r\n"), '\n', '\n')
select {
case out <- cliproxyexecutor.StreamChunk{Payload: visibleFrame}:
case <-ctx.Done():
return false
}
return true
}
if len(payload) == 0 {
return true
}
var lines [][]byte
lines = helps.TranslateStreamWithClaudeInputTokens(ctx, sdktranslator.FormatInteractions, responseFormat, req.Model, helps.ApplyPatchOriginalRequest(req, opts), body, payload, &param, claudeInputTokens)
helps.RecordApplyPatchStreamFailure(ctx, param, reporter, statusErr{code: http.StatusBadGateway, msg: helps.ApplyPatchUpstreamErrorMessage})
for i := range lines {
select {
case out <- cliproxyexecutor.StreamChunk{Payload: lines[i]}:
case <-ctx.Done():
return false
}
}
if helps.StopApplyPatchStream(ctx, param, reporter, out, statusErr{code: http.StatusBadGateway, msg: helps.ApplyPatchUpstreamErrorMessage}) {
return false
}
return true
}
for scanner.Scan() {
line := scanner.Bytes()
helps.AppendAPIResponseChunk(ctx, e.cfg, line)
trimmed := bytes.TrimSpace(line)
if len(trimmed) == 0 {
if !emitFrame() {
return
}
continue
}
if len(frame) > 0 {
frame = append(frame, '\n')
}
frame = append(frame, line...)
}
if !emitFrame() {
return
}
if helps.EndApplyPatchStream(ctx, param, reporter, out, statusErr{code: http.StatusBadGateway, msg: helps.ApplyPatchUpstreamErrorMessage}) {
return
}
if errScan := scanner.Err(); errScan != nil {
helps.RecordAPIResponseError(ctx, e.cfg, errScan)
reporter.PublishFailure(ctx, errScan)
select {
case out <- cliproxyexecutor.StreamChunk{Err: errScan}:
case <-ctx.Done():
}
}
}()
return &cliproxyexecutor.StreamResult{
Chunks: out,
Headers: httpResp.Header.Clone(),
}, nil
}