feat: support Responses to Chat (#5787)

* fix(openai): harden Chat-to-Responses compatibility Add a shared Responses-to-Chat stream state machine and use it from the OpenAI relay path. Preserve assistant text alongside tool calls, bind tool argument deltas by output_index, map incomplete finish reasons, support reasoning/custom tool events, and buffer upstream SSE for non-stream Chat clients. Add deterministic service tests and relay SSE tests for the conversion path. Related to #5745. * refactor: rename openaicompat to relayconvert for improved clarity * feat(gemini): support responses request conversion * feat: add responses to chat conversion support * fix: harden responses chat conversion edge cases
2026-09-07 01:56:53 +00:00 · 2026-06-28 14:25:47 +08:00
parent 3a506f50f0
commit 2d5a041639
35 changed files with 2731 additions and 40 deletions
@@ -0,0 +1,162 @@
+package gemini
+
+import (
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+
+	"github.com/QuantumNous/new-api/common"
+	"github.com/QuantumNous/new-api/constant"
+	"github.com/QuantumNous/new-api/dto"
+	"github.com/QuantumNous/new-api/logger"
+	relaycommon "github.com/QuantumNous/new-api/relay/common"
+	"github.com/QuantumNous/new-api/relay/helper"
+	"github.com/QuantumNous/new-api/service"
+	"github.com/QuantumNous/new-api/service/relayconvert"
+	"github.com/QuantumNous/new-api/types"
+	"github.com/gin-gonic/gin"
+)
+
+func GeminiResponsesHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http.Response) (*dto.Usage, *types.NewAPIError) {
+	defer service.CloseResponseBodyGracefully(resp)
+
+	responseBody, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, types.NewOpenAIError(err, types.ErrorCodeBadResponseBody, http.StatusInternalServerError)
+	}
+	logger.LogDebug(c, "Gemini responses response body: %s", responseBody)
+
+	var geminiResponse dto.GeminiChatResponse
+	if err := common.Unmarshal(responseBody, &geminiResponse); err != nil {
+		return nil, types.NewOpenAIError(err, types.ErrorCodeBadResponseBody, http.StatusInternalServerError)
+	}
+	if len(geminiResponse.Candidates) == 0 {
+		usage := buildUsageFromGeminiMetadata(geminiResponse.UsageMetadata, info.GetEstimatePromptTokens())
+		if geminiResponse.PromptFeedback != nil && geminiResponse.PromptFeedback.BlockReason != nil {
+			common.SetContextKey(c, constant.ContextKeyAdminRejectReason, fmt.Sprintf("gemini_block_reason=%s", *geminiResponse.PromptFeedback.BlockReason))
+			return &usage, types.NewOpenAIError(
+				errors.New("request blocked by Gemini API: "+*geminiResponse.PromptFeedback.BlockReason),
+				types.ErrorCodePromptBlocked,
+				http.StatusBadRequest,
+			)
+		}
+		common.SetContextKey(c, constant.ContextKeyAdminRejectReason, "gemini_empty_candidates")
+		return &usage, types.NewOpenAIError(
+			errors.New("empty response from Gemini API"),
+			types.ErrorCodeEmptyResponse,
+			http.StatusInternalServerError,
+		)
+	}
+
+	chatResp := responseGeminiChat2OpenAI(c, &geminiResponse)
+	chatResp.Model = info.UpstreamModelName
+	usage := buildUsageFromGeminiMetadata(geminiResponse.UsageMetadata, info.GetEstimatePromptTokens())
+	chatResp.Usage = usage
+
+	responsesResp, responsesUsage, err := service.ChatCompletionsResponseToResponsesResponse(chatResp, helper.GetResponseID(c))
+	if err != nil {
+		return nil, types.NewOpenAIError(err, types.ErrorCodeBadResponseBody, http.StatusInternalServerError)
+	}
+	if responsesUsage == nil || responsesUsage.TotalTokens == 0 {
+		responsesResp.Usage = relayconvert.UsageFromChatUsage(&usage)
+	}
+
+	responseBody, err = common.Marshal(responsesResp)
+	if err != nil {
+		return nil, types.NewOpenAIError(err, types.ErrorCodeJsonMarshalFailed, http.StatusInternalServerError)
+	}
+	service.IOCopyBytesGracefully(c, resp, responseBody)
+	return &usage, nil
+}
+
+func GeminiResponsesStreamHandler(c *gin.Context, info *relaycommon.RelayInfo, resp *http.Response) (*dto.Usage, *types.NewAPIError) {
+	responseID := helper.GetResponseID(c)
+	created := common.GetTimestamp()
+	state := relayconvert.NewChatToResponsesStreamState(responseID, info.UpstreamModelName)
+	state.Created = created
+	finishReason := constant.FinishReasonStop
+	toolCallIndexByChoice := make(map[int]map[string]int)
+	nextToolCallIndexByChoice := make(map[int]int)
+	var streamErr *types.NewAPIError
+
+	sendEvent := func(event relayconvert.ChatToResponsesStreamEvent) bool {
+		data, err := common.Marshal(event.Payload)
+		if err != nil {
+			streamErr = types.NewOpenAIError(err, types.ErrorCodeJsonMarshalFailed, http.StatusInternalServerError)
+			return false
+		}
+		helper.ResponseChunkData(c, dto.ResponsesStreamResponse{Type: event.Type}, string(data))
+		return true
+	}
+	sendChunk := func(chunk *dto.ChatCompletionsStreamResponse) bool {
+		events, err := relayconvert.ChatCompletionsStreamChunkToResponsesEvents(chunk, state)
+		if err != nil {
+			streamErr = types.NewOpenAIError(err, types.ErrorCodeBadResponse, http.StatusInternalServerError)
+			return false
+		}
+		for _, event := range events {
+			if !sendEvent(event) {
+				return false
+			}
+		}
+		return true
+	}
+
+	usage, err := geminiStreamHandler(c, info, resp, func(data string, geminiResponse *dto.GeminiChatResponse) bool {
+		response, isStop := streamResponseGeminiChat2OpenAI(geminiResponse)
+		response.Id = responseID
+		response.Created = created
+		response.Model = info.UpstreamModelName
+
+		if response.IsToolCall() {
+			finishReason = constant.FinishReasonToolCalls
+		}
+		for choiceIdx := range response.Choices {
+			choiceKey := response.Choices[choiceIdx].Index
+			for toolIdx := range response.Choices[choiceIdx].Delta.ToolCalls {
+				tool := &response.Choices[choiceIdx].Delta.ToolCalls[toolIdx]
+				if tool.ID == "" {
+					continue
+				}
+				indexByID := toolCallIndexByChoice[choiceKey]
+				if indexByID == nil {
+					indexByID = make(map[string]int)
+					toolCallIndexByChoice[choiceKey] = indexByID
+				}
+				if idx, ok := indexByID[tool.ID]; ok {
+					tool.SetIndex(idx)
+					continue
+				}
+				idx := nextToolCallIndexByChoice[choiceKey]
+				nextToolCallIndexByChoice[choiceKey] = idx + 1
+				indexByID[tool.ID] = idx
+				tool.SetIndex(idx)
+			}
+		}
+
+		if !sendChunk(response) {
+			return false
+		}
+		if isStop {
+			return sendChunk(helper.GenerateStopResponse(responseID, created, info.UpstreamModelName, finishReason))
+		}
+		return true
+	})
+	if err != nil {
+		return usage, err
+	}
+	if streamErr != nil {
+		return nil, streamErr
+	}
+
+	if usage != nil {
+		state.Usage = relayconvert.UsageFromChatUsage(usage)
+	}
+	for _, event := range relayconvert.FinalizeChatCompletionsStreamToResponses(state) {
+		if !sendEvent(event) {
+			return nil, streamErr
+		}
+	}
+	return usage, nil
+}