feat: add support for custom OpenAI endpoint path

- Introduced a new endpoint option for OpenAI integrations, allowing users to specify a custom path.
- Updated relevant components and validation logic to accommodate the new endpoint.
- Enhanced documentation and error messages to reflect the addition of the custom endpoint option.

internal/backend/agent/model/openai.go

@@ -304,15 +304,44 @@ func OpenAIEndpointURL(baseURL string, endpoint string) string {
 	if !strings.HasPrefix(normalizedEndpoint, "/") {
 		normalizedEndpoint = "/" + normalizedEndpoint
 	}
+	// 规则0：自定义路径模式 → 直接用 baseURL 作为完整请求地址
+	if normalizedEndpoint == modelchannel.OpenAIEndpointCustom {
+		return base
+	}
+	// 规则1：baseURL 已含 endpoint 后缀 → 直接用 base
 	if OpenAIEndpointFromBaseURL(base) != "" {
 		return base
 	}
-	if strings.HasSuffix(base, "/v1") && strings.HasPrefix(normalizedEndpoint, "/v1/") {
-		return base + strings.TrimPrefix(normalizedEndpoint, "/v1")
+	// 规则2：通用版本段去重（/v1 /v2 /v3 /v4 ... 任意版本号）
+	if version, ok := trailingVersionSegment(base); ok {
+		prefix := "/" + version + "/"
+		if strings.HasPrefix(normalizedEndpoint, prefix) {
+			return base + strings.TrimPrefix(normalizedEndpoint, "/"+version)
+		}
 	}
+	// 规则3：兜底原样拼接
 	return base + normalizedEndpoint
 }
 
+// trailingVersionSegment 检测 URL 末尾是否以 /vN 形式结尾（N 为数字），
+// 返回版本段（如 "v4"）和是否匹配。用于通用版本段去重。
+func trailingVersionSegment(base string) (string, bool) {
+	idx := strings.LastIndex(base, "/")
+	if idx < 0 {
+		return "", false
+	}
+	seg := base[idx+1:]
+	if len(seg) < 2 || seg[0] != 'v' {
+		return "", false
+	}
+	for i := 1; i < len(seg); i++ {
+		if seg[i] < '0' || seg[i] > '9' {
+			return "", false
+		}
+	}
+	return seg, true
+}
+
 func ResolveOpenAIEndpoint(baseURL string, endpoint string) string {
 	if endpointFromURL := OpenAIEndpointFromBaseURL(baseURL); endpointFromURL != "" {
 		return endpointFromURL
@@ -377,11 +406,11 @@ func (adapter *OpenAIAdapter) Stream(ctx context.Context, req StreamRequest, sin
 	req.OpenAIEndpoint = endpoint
 	if req.RequestKnobs != nil {
 		req.RequestKnobs["openai_endpoint"] = endpoint
-		if endpoint == modelchannel.OpenAIEndpointResponses {
+		if modelchannel.OpenAIEndpointShape(endpoint) == "responses" {
 			req.RequestKnobs["max_output_tokens"] = req.MaxTokens
 		}
 	}
-	if endpoint == modelchannel.OpenAIEndpointResponses {
+	if modelchannel.OpenAIEndpointShape(endpoint) == "responses" {
 		return adapter.streamResponses(ctx, req, baseURL, apiKey, modelID, sink)
 	}
 	return adapter.streamChatCompletions(ctx, req, baseURL, apiKey, modelID, sink)
@@ -425,14 +454,14 @@ func (adapter *OpenAIAdapter) streamChatCompletions(ctx context.Context, req Str
 		recordLLMSummaryArtifact(req, buildLLMSummaryPayload(req, "openai", modelID, startedAt, time.Time{}, finishedAt, "", 0, 0, 0, 0, err))
 		return err
 	}
-	applyOpenAIThinkingDisable(bodyMap, req, baseURL, modelID, modelchannel.OpenAIEndpointChatCompletions)
+	applyOpenAIThinkingDisable(bodyMap, req, baseURL, modelID, req.OpenAIEndpoint)
 	if err := ApplyOpenAIExtraParams(bodyMap, req.OpenAIExtraParamsEnabled, req.OpenAIExtraParamsJSON); err != nil {
 		finishedAt = time.Now().UTC()
 		recordLLMSummaryArtifact(req, buildLLMSummaryPayload(req, "openai", modelID, startedAt, time.Time{}, finishedAt, "", 0, 0, 0, 0, err))
 		return err
 	}
 	body = bodyMap
-	requestURL := OpenAIEndpointURL(baseURL, modelchannel.OpenAIEndpointChatCompletions)
+	requestURL := OpenAIEndpointURL(baseURL, req.OpenAIEndpoint)
 	recordLLMRequestArtifact(req, "openai", modelID, "POST", requestURL, body)
 
 	payload, err := json.Marshal(body)
@@ -901,7 +930,7 @@ func (adapter *OpenAIAdapter) streamResponses(ctx context.Context, req StreamReq
 		recordLLMSummaryArtifact(req, buildLLMSummaryPayload(req, "openai", modelID, startedAt, time.Time{}, finishedAt, "", 0, 0, 0, 0, err))
 		return err
 	}
-	applyOpenAIThinkingDisable(bodyMap, req, baseURL, modelID, modelchannel.OpenAIEndpointResponses)
+	applyOpenAIThinkingDisable(bodyMap, req, baseURL, modelID, req.OpenAIEndpoint)
 	if err := ApplyOpenAIExtraParams(bodyMap, req.OpenAIExtraParamsEnabled, req.OpenAIExtraParamsJSON); err != nil {
 		finishedAt = time.Now().UTC()
 		recordLLMSummaryArtifact(req, buildLLMSummaryPayload(req, "openai", modelID, startedAt, time.Time{}, finishedAt, "", 0, 0, 0, 0, err))
@@ -909,7 +938,7 @@ func (adapter *OpenAIAdapter) streamResponses(ctx context.Context, req StreamReq
 	}
 	body = bodyMap
 
-	requestURL := OpenAIEndpointURL(baseURL, modelchannel.OpenAIEndpointResponses)
+	requestURL := OpenAIEndpointURL(baseURL, req.OpenAIEndpoint)
 	recordLLMRequestArtifact(req, "openai", modelID, "POST", requestURL, body)
 
 	payload, err := json.Marshal(body)
@@ -1819,7 +1848,7 @@ func applyOpenAIThinkingDisable(body map[string]any, req StreamRequest, baseURL
 		delete(body, "reasoning_effort")
 		setRequestKnob(req, "thinking_disabled_provider_param", "enable_thinking")
 	case "reasoning_none":
-		if endpoint == modelchannel.OpenAIEndpointResponses {
+		if modelchannel.OpenAIEndpointShape(endpoint) == "responses" {
 			body["reasoning"] = map[string]any{"effort": "none"}
 		} else {
 			body["reasoning_effort"] = "none"