АДРЕСНЫЙ РЕЖИМ -Step-5 - feat(assistant): стабилизация свободного LLM-роутинга, прическа маршрутов chat/address, прототип прогноза НДС

2026-04-08 23:40:12 +03:00
parent 71762af575
commit df29798fa2
47 changed files with 11439 additions and 651 deletions
@@ -203,6 +203,103 @@ function buildBaseUrlCandidates(config: OpenAIRequestConfig): string[] {
 }

 export class OpenAIResponsesClient {
+  public async chat(
+    config: OpenAIRequestConfig,
+    prompt: {
+      systemPrompt?: string;
+      developerPrompt?: string;
+      userMessage: string;
+      maxOutputTokens?: number;
+      temperature?: number;
+    }
+  ): Promise<OpenAIResponseEnvelope> {
+    const responsesPayload = {
+      model: config.model,
+      temperature: prompt.temperature ?? config.temperature ?? 0.2,
+      max_output_tokens: prompt.maxOutputTokens ?? config.maxOutputTokens ?? 400,
+      input: [
+        ...(String(prompt.systemPrompt ?? "").trim().length > 0
+          ? [
+              {
+                role: "system",
+                content: [{ type: "input_text", text: String(prompt.systemPrompt ?? "").trim() }]
+              }
+            ]
+          : []),
+        ...(String(prompt.developerPrompt ?? "").trim().length > 0
+          ? [
+              {
+                role: "developer",
+                content: [{ type: "input_text", text: String(prompt.developerPrompt ?? "").trim() }]
+              }
+            ]
+          : []),
+        {
+          role: "user",
+          content: [{ type: "input_text", text: String(prompt.userMessage ?? "") }]
+        }
+      ]
+    };
+
+    const provider = resolveProvider(config);
+    if (provider === "openai") {
+      const raw = await this.postResponses(config, responsesPayload);
+      return {
+        raw,
+        outputText: extractOutputTextFromResponses(raw),
+        usage: extractUsage(raw)
+      };
+    }
+
+    try {
+      const raw = await this.postResponses(config, responsesPayload);
+      return {
+        raw,
+        outputText: extractOutputTextFromResponses(raw),
+        usage: extractUsage(raw)
+      };
+    } catch (error) {
+      if (!shouldFallbackToChatCompletions(error)) {
+        throw error;
+      }
+    }
+
+    const chatPayload = {
+      model: config.model,
+      temperature: prompt.temperature ?? config.temperature ?? 0.2,
+      max_tokens: prompt.maxOutputTokens ?? config.maxOutputTokens ?? 400,
+      messages: [
+        ...(String(prompt.systemPrompt ?? "").trim().length > 0
+          ? [
+              {
+                role: "system",
+                content: String(prompt.systemPrompt ?? "").trim()
+              }
+            ]
+          : []),
+        ...(String(prompt.developerPrompt ?? "").trim().length > 0
+          ? [
+              {
+                role: "developer",
+                content: String(prompt.developerPrompt ?? "").trim()
+              }
+            ]
+          : []),
+        {
+          role: "user",
+          content: String(prompt.userMessage ?? "")
+        }
+      ]
+    };
+
+    const raw = await this.postChatCompletions(config, chatPayload);
+    return {
+      raw,
+      outputText: extractOutputTextFromChatCompletions(raw),
+      usage: extractUsage(raw)
+    };
+  }
+
  public async listModels(config: OpenAIRequestConfig): Promise<string[]> {
    const payload = await this.getModels(config);
    const data = Array.isArray(payload.data) ? payload.data : [];