GoClaw/server/ollama.ts

/**
 * Ollama API Client — серверный прокси для безопасного доступа к Ollama Cloud API.
 * Хранит API-ключ на сервере, обходит CORS, предоставляет типизированные функции.
 */
import { ENV } from "./_core/env";

const TIMEOUT_MS = 30_000;

interface OllamaModel {
  id: string;
  object: string;
  created: number;
  owned_by: string;
}

interface OllamaModelsResponse {
  object: string;
  data: OllamaModel[];
}

interface ChatMessage {
  role: "system" | "user" | "assistant" | "tool" | "function";
  content: string | null;
  tool_call_id?: string;
  name?: string;
  tool_calls?: ToolCall[];
}

interface ToolFunction {
  name: string;
  description?: string;
  parameters?: Record<string, unknown>;
}

interface Tool {
  type: "function";
  function: ToolFunction;
}

interface ToolCallFunction {
  name: string;
  arguments: string;
}

interface ToolCall {
  id: string;
  type: "function";
  function: ToolCallFunction;
}

interface ChatChoice {
  index: number;
  message: {
    role: string;
    content: string | null;
    tool_calls?: ToolCall[];
  };
  finish_reason: string;
}

interface ChatCompletionResponse {
  id: string;
  object: string;
  created: number;
  model: string;
  choices: ChatChoice[];
  usage?: {
    prompt_tokens: number;
    completion_tokens: number;
    total_tokens: number;
  };
}

function getHeaders(): Record<string, string> {
  const headers: Record<string, string> = {
    "Content-Type": "application/json",
  };
  if (ENV.ollamaApiKey) {
    headers["Authorization"] = `Bearer ${ENV.ollamaApiKey}`;
  }
  return headers;
}

function getBaseUrl(): string {
  return ENV.ollamaBaseUrl.replace(/\/$/, "");
}

/**
 * Проверка доступности Ollama API
 */
export async function checkOllamaHealth(): Promise<{
  connected: boolean;
  latencyMs: number;
  error?: string;
}> {
  const start = Date.now();
  try {
    const res = await fetch(`${getBaseUrl()}/models`, {
      method: "GET",
      headers: getHeaders(),
      signal: AbortSignal.timeout(10_000),
    });
    const latencyMs = Date.now() - start;
    if (res.ok) {
      return { connected: true, latencyMs };
    }
    const text = await res.text();
    return { connected: false, latencyMs, error: `HTTP ${res.status}: ${text}` };
  } catch (err: any) {
    return { connected: false, latencyMs: Date.now() - start, error: err.message };
  }
}

/**
 * Получение списка доступных моделей (OpenAI-совместимый формат)
 */
export async function listModels(): Promise<OllamaModelsResponse> {
  const res = await fetch(`${getBaseUrl()}/models`, {
    method: "GET",
    headers: getHeaders(),
    signal: AbortSignal.timeout(TIMEOUT_MS),
  });
  if (!res.ok) {
    const text = await res.text();
    throw new Error(`Ollama API error (${res.status}): ${text}`);
  }
  return res.json();
}

/**
 * Отправка сообщения в чат (OpenAI-совместимый формат, без стриминга)
 * Поддерживает tool calling для оркестратора.
 */
export async function chatCompletion(
  model: string,
  messages: ChatMessage[],
  options?: {
    temperature?: number;
    max_tokens?: number;
    tools?: Tool[];
    tool_choice?: "auto" | "none" | "required";
  }
): Promise<ChatCompletionResponse> {
  const body: Record<string, unknown> = {
    model,
    messages,
    stream: false,
  };
  if (options?.temperature !== undefined) body.temperature = options.temperature;
  if (options?.max_tokens !== undefined) body.max_tokens = options.max_tokens;
  if (options?.tools?.length) {
    body.tools = options.tools;
    body.tool_choice = options.tool_choice ?? "auto";
  }

  const res = await fetch(`${getBaseUrl()}/chat/completions`, {
    method: "POST",
    headers: getHeaders(),
    body: JSON.stringify(body),
    signal: AbortSignal.timeout(120_000), // LLM может думать долго
  });
  if (!res.ok) {
    const text = await res.text();
    throw new Error(`Ollama Chat API error (${res.status}): ${text}`);
  }
  return res.json();
}

export type { OllamaModel, OllamaModelsResponse, ChatMessage, ChatCompletionResponse };