Initial commit: VoIdeaAI - voice-first AI idea assistant

app/integrations/ai/__init__.py

@@ -0,0 +1,14 @@
+"""VoIdea - AI integration module."""
+
+from app.integrations.ai.base import AIProvider, AIResult
+from app.integrations.ai.yandex_gpt import YandexGPTProvider
+from app.integrations.ai.gigachat import GigaChatProvider
+from app.integrations.ai.fallback import FallbackChain
+
+__all__ = [
+    "AIProvider",
+    "AIResult",
+    "YandexGPTProvider",
+    "GigaChatProvider",
+    "FallbackChain",
+]

app/integrations/ai/base.py

@@ -0,0 +1,64 @@
+"""Base classes for AI provider integration."""
+
+from __future__ import annotations
+
+from abc import ABC, abstractmethod
+from dataclasses import dataclass, field
+from datetime import datetime, timezone
+from typing import Any
+
+
+@dataclass
+class AIResult:
+    """Result from an AI provider call."""
+
+    success: bool
+    content: str = ""
+    model: str = ""
+    provider: str = ""
+    tokens_used: int = 0
+    duration_ms: int = 0
+    error: str = ""
+    metadata: dict[str, Any] = field(default_factory=dict)
+
+
+class AIProvider(ABC):
+    """Abstract base class for AI providers.
+
+    All AI providers (Yandex GPT, GigaChat, etc.) must implement this.
+    """
+
+    name: str = ""
+    model: str = ""
+    timeout: int = 10
+    max_retries: int = 3
+
+    def __init__(self, api_key: str = "", **kwargs: Any):
+        self.api_key = api_key
+        for key, value in kwargs.items():
+            setattr(self, key, value)
+        self._initialize()
+
+    def _initialize(self) -> None:
+        """Post-initialization hook for provider-specific setup."""
+
+    @abstractmethod
+    async def analyze(self, prompt: str, **kwargs: Any) -> AIResult:
+        """Send a prompt to the AI provider and return the result.
+
+        Args:
+            prompt: The formatted prompt to send
+            **kwargs: Additional provider-specific parameters
+
+        Returns:
+            AIResult with the response content
+        """
+
+    def format_prompt(
+        self, system_prompt: str, user_prompt: str, **kwargs: Any
+    ) -> str:
+        """Format a prompt with system instructions and user message.
+
+        Override in provider if the API uses separate system/user messages.
+        """
+        return f"{system_prompt}\n\n{user_prompt}"

app/integrations/ai/fallback.py

@@ -0,0 +1,99 @@
+"""Fallback chain for AI providers.
+
+Tries providers in order with retries per provider.
+YandexGPT -> GigaChat -> Error
+"""
+
+import asyncio
+import time
+from typing import Any
+
+from app.core.config import get_settings
+from app.integrations.ai.base import AIProvider, AIResult
+from app.integrations.ai.yandex_gpt import YandexGPTProvider
+from app.integrations.ai.gigachat import GigaChatProvider
+
+settings = get_settings()
+
+
+class FallbackChain:
+    """Chain multiple AI providers with retry logic.
+
+    Tries providers in order (YandexGPT -> GigaChat).
+    Each provider gets up to `max_retries` attempts before falling through.
+    """
+
+    def __init__(
+        self,
+        providers: list[AIProvider] | None = None,
+        max_retries: int = 2,
+        timeout: int = 30,
+    ):
+        self.providers = providers or [
+            YandexGPTProvider(),
+            GigaChatProvider(),
+        ]
+        self.max_retries = max_retries
+        self.timeout = timeout
+
+    async def analyze(self, prompt: str, **kwargs: Any) -> AIResult:
+        """Try each provider in order with retries.
+
+        Args:
+            prompt: The formatted prompt to send
+            **kwargs: Additional parameters passed to each provider
+
+        Returns:
+            AIResult from the first successful provider,
+            or the last error result if all fail.
+        """
+        last_error: AIResult | None = None
+
+        for provider in self.providers:
+            for attempt in range(self.max_retries + 1):
+                try:
+                    result = await asyncio.wait_for(
+                        provider.analyze(prompt, **kwargs),
+                        timeout=self.timeout,
+                    )
+
+                    if result.success:
+                        await self._cleanup_providers(provider)
+                        return result
+
+                    last_error = result
+
+                except asyncio.TimeoutError:
+                    last_error = AIResult(
+                        success=False,
+                        error=f"{provider.name} timeout after {self.timeout}s",
+                        provider=provider.name,
+                        duration_ms=int(self.timeout * 1000),
+                    )
+                except Exception as e:
+                    last_error = AIResult(
+                        success=False,
+                        error=f"{provider.name} error: {str(e)}",
+                        provider=provider.name,
+                    )
+
+                if attempt < self.max_retries:
+                    await asyncio.sleep(1 * (attempt + 1))
+
+        await self._cleanup_providers()
+        return last_error or AIResult(
+            success=False,
+            error="All providers failed",
+            provider="fallback_chain",
+        )
+
+    async def _cleanup_providers(self, skip: AIProvider | None = None) -> None:
+        """Close all provider clients except the one to keep."""
+        for provider in self.providers:
+            if provider is skip:
+                continue
+            if hasattr(provider, "close"):
+                try:
+                    await provider.close()
+                except Exception:
+                    pass

app/integrations/ai/gigachat.py

@@ -0,0 +1,160 @@
+"""GigaChat provider for VoIdea."""
+
+import base64
+import time
+from datetime import datetime, timezone
+from typing import Any, Optional
+
+import httpx
+
+from app.core.config import get_settings
+from app.integrations.ai.base import AIProvider, AIResult
+
+settings = get_settings()
+
+
+class GigaChatProvider(AIProvider):
+    """GigaChat AI provider (Sber).
+
+    Uses OAuth client credentials for authentication.
+    Token is refreshed automatically on 401.
+    """
+
+    name = "gigachat"
+    model = "GigaChat"
+    timeout = settings.ai_timeout
+    max_retries = settings.ai_max_retries
+
+    def __init__(self, api_key: str = ""):
+        super().__init__(api_key)
+        self.base_url = settings.ai_gigachat_url.rstrip("/")
+        self.client_id = settings.ai_gigachat_client_id
+        self.client_secret = settings.ai_gigachat_secret
+        self._client = httpx.AsyncClient(timeout=self.timeout)
+        self._access_token: Optional[str] = None
+        self._token_expires_at: float = 0
+
+    async def _get_access_token(self) -> Optional[str]:
+        """Get or refresh GigaChat access token."""
+        if self._access_token and time.time() < self._token_expires_at:
+            return self._access_token
+
+        if not self.client_id or not self.client_secret:
+            return None
+
+        auth_str = base64.b64encode(
+            f"{self.client_id}:{self.client_secret}".encode()
+        ).decode()
+
+        try:
+            response = await self._client.post(
+                "https://ngw.devices.sber.ru/token",
+                headers={
+                    "Authorization": f"Basic {auth_str}",
+                    "Content-Type": "application/x-www-form-urlencoded",
+                    "RqUID": str(time.time()),
+                },
+                data={"scope": "GIGACHAT_API_PERS"},
+            )
+
+            if response.status_code == 200:
+                data = response.json()
+                self._access_token = data.get("access_token")
+                expires_in = data.get("expires_in", 1800)
+                self._token_expires_at = time.time() + expires_in - 60
+                return self._access_token
+
+        except Exception:
+            return None
+
+        return None
+
+    async def analyze(self, prompt: str, **kwargs: Any) -> AIResult:
+        """Send prompt to GigaChat and return result."""
+        start = time.time()
+        temperature = kwargs.get("temperature", 0.7)
+        max_tokens = kwargs.get("max_tokens", 2000)
+
+        token = await self._get_access_token()
+        if not token:
+            return AIResult(
+                success=False,
+                error="GigaChat authentication failed",
+                provider=self.name,
+            )
+
+        parts = prompt.split("\n\n", 1)
+        system_text = parts[0]
+        user_text = parts[1] if len(parts) > 1 else ""
+
+        body = {
+            "model": self.model,
+            "temperature": temperature,
+            "max_tokens": max_tokens,
+            "messages": [
+                {"role": "system", "content": system_text},
+                {"role": "user", "content": user_text},
+            ],
+        }
+
+        try:
+            response = await self._client.post(
+                f"{self.base_url}/api/v1/chat/completion",
+                json=body,
+                headers={"Authorization": f"Bearer {token}"},
+            )
+
+            duration = int((time.time() - start) * 1000)
+
+            if response.status_code == 401:
+                self._access_token = None
+                token = await self._get_access_token()
+                if token:
+                    response = await self._client.post(
+                        f"{self.base_url}/api/v1/chat/completion",
+                        json=body,
+                        headers={"Authorization": f"Bearer {token}"},
+                    )
+                    duration = int((time.time() - start) * 1000)
+
+            if response.status_code == 200:
+                data = response.json()
+                result_text = (
+                    data.get("choices", [{}])[0]
+                    .get("message", {})
+                    .get("content", "")
+                )
+
+                return AIResult(
+                    success=True,
+                    content=result_text,
+                    model=self.model,
+                    provider=self.name,
+                    duration_ms=duration,
+                )
+            else:
+                error_body = response.text[:500]
+                return AIResult(
+                    success=False,
+                    error=f"GigaChat {response.status_code}: {error_body}",
+                    provider=self.name,
+                    duration_ms=duration,
+                )
+
+        except httpx.TimeoutException:
+            return AIResult(
+                success=False,
+                error="GigaChat timeout",
+                provider=self.name,
+                duration_ms=int((time.time() - start) * 1000),
+            )
+        except Exception as e:
+            return AIResult(
+                success=False,
+                error=f"GigaChat error: {str(e)}",
+                provider=self.name,
+                duration_ms=int((time.time() - start) * 1000),
+            )
+
+    async def close(self) -> None:
+        await self._client.aclose()

app/integrations/ai/prompt_loader.py

@@ -0,0 +1,90 @@
+"""Prompt loader for AI agents.
+
+Loads prompts from agent spec files or YAML configuration.
+"""
+
+import re
+from pathlib import Path
+from typing import Any, Optional
+
+import yaml
+
+AGENT_SPECS_DIR = Path("docs/specs/agents")
+AGENT_PROMPTS_YAML = Path("docs/agent_prompts.yaml")
+
+
+def get_agent_roles() -> list[str]:
+    """Get all available agent roles."""
+    roles = []
+    if AGENT_PROMPTS_YAML.exists():
+        data = yaml.safe_load(AGENT_PROMPTS_YAML.read_text(encoding="utf-8"))
+        return list(data.keys()) if data else []
+
+    if AGENT_SPECS_DIR.exists():
+        for f in sorted(AGENT_SPECS_DIR.glob("*.md")):
+            roles.append(f.stem)
+
+    return roles
+
+
+def load_prompt_from_yaml(role: str) -> Optional[dict[str, Any]]:
+    """Load prompt config from YAML file."""
+    if not AGENT_PROMPTS_YAML.exists():
+        return None
+
+    data = yaml.safe_load(AGENT_PROMPTS_YAML.read_text(encoding="utf-8"))
+    return data.get(role) if data else None
+
+
+def load_prompt_from_spec(role: str) -> Optional[dict[str, Any]]:
+    """Load prompt from agent spec markdown file."""
+    spec_path = AGENT_SPECS_DIR / f"{role}.md"
+    if not spec_path.exists():
+        return None
+
+    content = spec_path.read_text(encoding="utf-8")
+
+    def extract_field(label: str) -> Optional[str]:
+        match = re.search(rf"\*\*{label}:\*\*\s*(.+)", content)
+        return match.group(1).strip() if match else None
+
+    def extract_prompt() -> Optional[str]:
+        match = re.search(
+            r"## Prompt Template\n+```\n(.+?)\n```",
+            content,
+            re.DOTALL,
+        )
+        return match.group(1).strip() if match else None
+
+    provider_str = extract_field("Провайдер") or extract_field("Provider") or "yandex_gpt"
+    provider = "gigachat" if "gigachat" in (provider_str or "").lower() else "yandex_gpt"
+
+    prompt_text = extract_prompt()
+    if not prompt_text:
+        return None
+
+    return {
+        "system_prompt": prompt_text,
+        "provider": provider,
+        "temperature": 0.7,
+        "max_tokens": 2000,
+    }
+
+
+def get_prompt_config(role: str) -> Optional[dict[str, Any]]:
+    """Get prompt configuration for a role.
+
+    Tries YAML first, falls back to spec markdown.
+
+    Args:
+        role: Agent role name (e.g. "coordinator", "business_analyst")
+
+    Returns:
+        Dict with keys: system_prompt, provider, temperature, max_tokens
+        or None if not found.
+    """
+    config = load_prompt_from_yaml(role)
+    if config:
+        return config
+
+    return load_prompt_from_spec(role)

app/integrations/ai/yandex_gpt.py

@@ -0,0 +1,122 @@
+"""Yandex GPT provider for VoIdea."""
+
+import time
+from datetime import datetime, timezone
+from typing import Any, Optional
+
+import httpx
+
+from app.core.config import get_settings
+from app.integrations.ai.base import AIProvider, AIResult
+
+settings = get_settings()
+
+
+class YandexGPTProvider(AIProvider):
+    """Yandex GPT AI provider.
+
+    Supports both API key (Api-Key header) and IAM token (Bearer header).
+    Automatically falls back between auth methods on 401.
+    """
+
+    name = "yandex_gpt"
+    model = "yandexgpt/latest"
+    timeout = settings.ai_timeout
+    max_retries = settings.ai_max_retries
+
+    def __init__(self, api_key: str = ""):
+        super().__init__(api_key or settings.ai_yandex_key)
+        self.base_url = settings.ai_yandex_url.rstrip("/")
+        self.folder_id = settings.ai_yandex_folder_id
+        self._client = httpx.AsyncClient(timeout=self.timeout)
+
+    async def analyze(self, prompt: str, **kwargs: Any) -> AIResult:
+        """Send prompt to Yandex GPT and return result."""
+        start = time.time()
+        temperature = kwargs.get("temperature", 0.7)
+        max_tokens = kwargs.get("max_tokens", 2000)
+
+        model_uri = f"gpt://{self.folder_id}/{self.model}" if self.folder_id else self.model
+
+        # Split prompt into system and user parts
+        parts = prompt.split("\n\n", 1)
+        system_text = parts[0]
+        user_text = parts[1] if len(parts) > 1 else ""
+
+        body = {
+            "modelUri": model_uri,
+            "completionOptions": {
+                "temperature": temperature,
+                "maxTokens": max_tokens,
+            },
+            "messages": [
+                {"role": "system", "text": system_text},
+                {"role": "user", "text": user_text},
+            ],
+        }
+
+        try:
+            response = await self._client.post(
+                f"{self.base_url}/foundationModels/v1/completion",
+                json=body,
+            )
+
+            duration = int((time.time() - start) * 1000)
+
+            if response.status_code == 401:
+                # Try with Api-Key header instead
+                response = await self._client.post(
+                    f"{self.base_url}/foundationModels/v1/completion",
+                    json=body,
+                    headers={"Authorization": f"Api-Key {self.api_key}"},
+                )
+                duration = int((time.time() - start) * 1000)
+
+            if response.status_code == 200:
+                data = response.json()
+                result_text = (
+                    data.get("result", {})
+                    .get("alternatives", [{}])[0]
+                    .get("message", {})
+                    .get("text", "")
+                )
+                tokens = (
+                    data.get("result", {})
+                    .get("usage", {})
+                    .get("inputTextTokens", 0)
+                )
+
+                return AIResult(
+                    success=True,
+                    content=result_text,
+                    model=self.model,
+                    provider=self.name,
+                    tokens_used=tokens,
+                    duration_ms=duration,
+                )
+            else:
+                error_body = response.text[:500]
+                return AIResult(
+                    success=False,
+                    error=f"Yandex GPT {response.status_code}: {error_body}",
+                    provider=self.name,
+                    duration_ms=duration,
+                )
+
+        except httpx.TimeoutException:
+            return AIResult(
+                success=False,
+                error="Yandex GPT timeout",
+                provider=self.name,
+                duration_ms=int((time.time() - start) * 1000),
+            )
+        except Exception as e:
+            return AIResult(
+                success=False,
+                error=f"Yandex GPT error: {str(e)}",
+                provider=self.name,
+                duration_ms=int((time.time() - start) * 1000),
+            )
+
+    async def close(self) -> None:
+        await self._client.aclose()