aline-ai 0.6.2__py3-none-any.whl → 0.6.3__py3-none-any.whl

realign/llm_client.py

@@ -6,14 +6,13 @@ This module provides a centralized interface for calling LLM providers (Claude,
 with configurable models and parameters.
 """
 
-import os
 import sys
 import time
 import json
 import logging
 import tempfile
 from datetime import datetime
-from typing import Optional, Tuple, Dict, Any, Callable, List
+from typing import Optional, Tuple, Dict, Any
 from pathlib import Path
 
 # Setup dedicated LLM logger
@@ -67,563 +66,245 @@ def _setup_llm_call_logger():
     return _llm_call_logger
 
 
-def call_llm(
-    system_prompt: str,
-    user_prompt: str,
-    provider: Optional[str] = None,
-    model: Optional[str] = None,
-    max_tokens: Optional[int] = None,
-    temperature: Optional[float] = None,
-    json_mode: bool = False,
-    debug_callback: Optional[Callable[[Dict[str, Any]], None]] = None,
-    purpose: str = "generic",
-    silent: bool = False,
-) -> Tuple[Optional[str], Optional[str]]:
+def extract_json(response_text: str) -> Dict[str, Any]:
     """
-    Unified LLM calling function.
+    Extract JSON object from a raw LLM response, handling Markdown fences.
+    Uses strict=False to tolerate control characters in JSON strings.
 
     Args:
-        system_prompt: System prompt
-        user_prompt: User prompt
-        provider: LLM provider ("auto", "claude", "openai"), None = read from config
-        model: Model name, None = use default from config
-        max_tokens: Maximum tokens to generate, None = use default
-        temperature: Temperature parameter, None = use default
-        json_mode: Enable JSON mode (OpenAI only)
-        debug_callback: Debug callback function
-        purpose: Purpose string for logging
-        silent: If True, suppress progress messages to stderr
+        response_text: Raw LLM response
 
     Returns:
-        (model_name, response_text) or (None, None) on failure
+        Parsed JSON dict
 
     Raises:
-        No exceptions raised - returns (None, None) on failure
+        json.JSONDecodeError: If JSON parsing fails
     """
-    # Load configuration
-    from .config import ReAlignConfig
+    if not response_text:
+        raise json.JSONDecodeError("Empty response", "", 0)
 
-    config = ReAlignConfig.load()
+    json_str = response_text.strip()
 
-    # Resolve provider from config if not specified
-    if provider is None:
-        provider = config.llm_provider
-
-    # Resolve default parameters from config if not specified
-    if max_tokens is None:
-        max_tokens = config.llm_max_tokens
-    if temperature is None:
-        temperature = config.llm_temperature
-
-    def _should_use_openai_responses(model_name: str) -> bool:
-        """
-        Decide if the OpenAI responses/reasoning API should be used for this model.
-        """
-        # Check explicit override
-        if config.llm_openai_use_responses:
-            return True
-
-        # Also check environment variable for backwards compatibility
-        override = os.getenv("REALIGN_OPENAI_USE_RESPONSES", "").strip().lower()
-        if override in ("1", "true", "yes"):
-            return True
-        if override in ("0", "false", "no"):
-            return False
-        if not model_name:
-            return False
-        # Auto-detect: use responses API for GPT-5+ models
-        lowered = model_name.lower()
-        return lowered.startswith("gpt-5")
-
-    def _collect_responses_output_text(response: Any) -> str:
-        """
-        Aggregate textual content from OpenAI responses API objects.
-        """
-        parts: List[str] = []
-        output_items = getattr(response, "output", None) or []
-        for item in output_items:
-            content = getattr(item, "content", None) or []
-            for block in content:
-                text = getattr(block, "text", None)
-                if text:
-                    parts.append(text)
-        text = "".join(parts).strip()
-        if text:
-            return text
-        fallback = getattr(response, "output_text", "") or ""
-        return fallback.strip()
-
-    def _emit_debug(payload: Dict[str, Any]) -> None:
-        """Emit debug event if callback is provided."""
-        if not debug_callback:
-            return
-        try:
-            debug_callback(payload)
-        except Exception:
-            logger.debug("LLM debug callback failed for payload=%s", payload, exc_info=True)
+    # Remove markdown code fences if present
+    if "```json" in response_text:
+        json_start = response_text.find("```json") + 7
+        json_end = response_text.find("```", json_start)
+        if json_end != -1:
+            json_str = response_text[json_start:json_end].strip()
+    elif "```" in response_text:
+        json_start = response_text.find("```") + 3
+        json_end = response_text.find("```", json_start)
+        if json_end != -1:
+            json_str = response_text[json_start:json_end].strip()
+
+    if not json_str:
+        raise json.JSONDecodeError("No JSON content found", response_text, 0)
+
+    # Use strict=False to allow control characters in JSON strings
+    return json.loads(json_str, strict=False)
+
+
+def call_llm_cloud(
+    task: str,
+    payload: Dict[str, Any],
+    custom_prompt: Optional[str] = None,
+    preset_id: Optional[str] = None,
+    timeout: float = 60.0,
+    silent: bool = False,
+) -> Tuple[Optional[str], Optional[Dict[str, Any]]]:
+    """
+    Call LLM via Aline server (cloud proxy).
 
-    # Setup detailed logging
+    This function sends structured task/payload to the server which handles
+    the LLM call, protecting API keys and prompts from being exposed to clients.
+
+    Args:
+        task: Task type ("summary" | "metadata" | "session_summary" | "event_summary" | "ui_metadata")
+        payload: Task-specific data dict
+        custom_prompt: Optional user custom prompt override (from ~/.aline/prompts/)
+        preset_id: Optional preset ID for ui_metadata task
+        timeout: Request timeout in seconds
+        silent: If True, suppress progress messages to stderr
+
+    Returns:
+        (model_name, result_dict) or (None, None) on failure
+    """
+    # Setup logging
     call_logger = _setup_llm_call_logger()
     call_start_time = time.time()
     call_timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
 
-    # Log call initiation with all parameters
+    # Log call initiation
     call_logger.info("=" * 80)
-    call_logger.info(f"LLM CALL INITIATED")
+    call_logger.info("LLM CLOUD CALL INITIATED")
     call_logger.info(f"Timestamp: {call_timestamp}")
-    call_logger.info(f"Purpose: {purpose}")
-    call_logger.info(f"Provider: {provider}")
-    call_logger.info(f"Model: {model or 'default from config'}")
-    call_logger.info(f"Max Tokens: {max_tokens}")
-    call_logger.info(f"Temperature: {temperature}")
-    call_logger.info(f"JSON Mode: {json_mode}")
+    call_logger.info(f"Task: {task}")
+    call_logger.info(f"Payload keys: {list(payload.keys())}")
+    call_logger.info(f"Custom prompt: {'yes' if custom_prompt else 'no'}")
     call_logger.info("-" * 80)
-    call_logger.info(f"SYSTEM PROMPT:\n{system_prompt}")
-    call_logger.info("-" * 80)
-    call_logger.info(f"USER PROMPT:\n{user_prompt}")
-    call_logger.info("-" * 80)
-
-    try_claude = provider in ("auto", "claude")
-    try_openai = provider in ("auto", "openai")
-
-    _emit_debug(
-        {
-            "event": "llm_prompt",
-            "target_provider": provider,
-            "system_prompt": system_prompt,
-            "user_prompt": user_prompt,
-            "provider_options": {
-                "try_claude": try_claude,
-                "try_openai": try_openai,
-            },
-            "purpose": purpose,
-        }
-    )
 
-    # Try Claude
-    anthropic_key = config.anthropic_api_key
-    if try_claude and anthropic_key:
-        logger.debug("ANTHROPIC_API_KEY found, attempting Claude")
+    # Check if httpx is available
+    try:
+        import httpx
+    except ImportError:
+        logger.error("httpx not available for cloud LLM calls")
         if not silent:
-            print("   → Trying Anthropic (Claude)...", file=sys.stderr)
-        try:
-            import anthropic
-
-            start_time = time.time()
-            client = anthropic.Anthropic(api_key=anthropic_key)
-
-            # Use model parameter if specified, otherwise read from config/env
-            claude_model = model or config.llm_anthropic_model
-
-            response = client.messages.create(
-                model=claude_model,
-                max_tokens=max_tokens,
-                temperature=temperature,
-                system=system_prompt,
-                messages=[{"role": "user", "content": user_prompt}],
-            )
-
-            elapsed = time.time() - start_time
-            response_text = response.content[0].text.strip()
-            logger.info(f"Claude API success: {len(response_text)} chars in {elapsed:.2f}s")
-            logger.debug(f"Claude response: {response_text[:200]}...")
-            _emit_debug(
-                {
-                    "event": "llm_response",
-                    "provider": "anthropic",
-                    "model": claude_model,
-                    "elapsed_seconds": elapsed,
-                    "raw_response": response_text,
-                    "purpose": purpose,
-                }
-            )
-
-            # Log successful response
-            total_elapsed = time.time() - call_start_time
-            call_logger.info(f"LLM CALL SUCCEEDED")
-            call_logger.info(f"Provider: Anthropic (Claude)")
-            call_logger.info(f"Model: {claude_model}")
-            call_logger.info(f"Elapsed Time: {elapsed:.2f}s")
-            call_logger.info(f"Total Time: {total_elapsed:.2f}s")
-            call_logger.info(f"Response Length: {len(response_text)} chars")
-            call_logger.info("-" * 80)
-            call_logger.info(f"RESPONSE:\n{response_text}")
-            call_logger.info("=" * 80 + "\n")
-
-            return claude_model, response_text
-
-        except ImportError:
-            logger.warning("Anthropic package not installed")
-            if provider == "claude":
-                if not silent:
-                    print("   ❌ Anthropic package not installed", file=sys.stderr)
-                total_elapsed = time.time() - call_start_time
-                call_logger.error(f"LLM CALL FAILED")
-                call_logger.error(f"Provider: Anthropic (Claude)")
-                call_logger.error(f"Reason: Anthropic package not installed")
-                call_logger.error(f"Total Time: {total_elapsed:.2f}s")
-                call_logger.error("=" * 80 + "\n")
-                return None, None
-            if not silent:
-                print(
-                    "   ❌ Anthropic package not installed, trying OpenAI...",
-                    file=sys.stderr,
-                )
-        except Exception as e:
-            error_msg = str(e)
-            logger.error(f"Claude API error: {error_msg}", exc_info=True)
-            if provider == "claude":
-                if not silent:
-                    if "authentication" in error_msg.lower() or "invalid" in error_msg.lower():
-                        print(
-                            f"   ❌ Anthropic authentication failed (check API key)",
-                            file=sys.stderr,
-                        )
-                    elif "quota" in error_msg.lower() or "credit" in error_msg.lower():
-                        print(f"   ❌ Anthropic quota/credit issue", file=sys.stderr)
-                    else:
-                        print(f"   ❌ Anthropic API error: {e}", file=sys.stderr)
-                total_elapsed = time.time() - call_start_time
-                call_logger.error(f"LLM CALL FAILED")
-                call_logger.error(f"Provider: Anthropic (Claude)")
-                call_logger.error(f"Reason: {error_msg}")
-                call_logger.error(f"Total Time: {total_elapsed:.2f}s")
-                call_logger.error("=" * 80 + "\n")
-                return None, None
-            if not silent:
-                print(f"   ❌ Anthropic API error: {e}, trying OpenAI...", file=sys.stderr)
+            print("   ❌ httpx package not installed", file=sys.stderr)
+        call_logger.error("LLM CLOUD CALL FAILED: httpx not installed")
+        call_logger.error("=" * 80 + "\n")
+        return None, None
 
-    elif try_claude:
-        logger.debug("Anthropic API key not configured in config file")
-        if provider == "claude":
-            if not silent:
-                print(
-                    "   ❌ Anthropic API key not configured in config file",
-                    file=sys.stderr,
-                )
-            return None, None
+    # Get auth token
+    try:
+        from .auth import get_access_token, is_logged_in
+    except ImportError:
+        logger.error("Auth module not available")
         if not silent:
-            print("Anthropic API key not configured, trying OpenAI...", file=sys.stderr)
+            print("   ❌ Auth module not available", file=sys.stderr)
+        call_logger.error("LLM CLOUD CALL FAILED: auth module not available")
+        call_logger.error("=" * 80 + "\n")
+        return None, None
 
-    # Try OpenAI
-    openai_key = config.openai_api_key
-    if try_openai and openai_key:
-        logger.debug("OPENAI_API_KEY found, attempting OpenAI")
+    if not is_logged_in():
+        logger.debug("Not logged in, cannot use cloud LLM")
         if not silent:
-            print("Trying OpenAI (GPT)...", file=sys.stderr)
-        try:
-            import openai
-
-            start_time = time.time()
-            client = openai.OpenAI(api_key=openai_key)
-
-            # Use model parameter if specified, otherwise read from config/env
-            openai_model = model or config.llm_openai_model
-            use_responses_api = _should_use_openai_responses(openai_model)
-
-            def _call_openai_chat_completion() -> Tuple[Any, str]:
-                use_completion_tokens = False
-                temperature_value = temperature
-                bad_request_error = getattr(openai, "BadRequestError", Exception)
-                last_error: Optional[Exception] = None
-                for _ in range(3):
-                    completion_kwargs = {
-                        "model": openai_model,
-                        "messages": [
-                            {"role": "system", "content": system_prompt},
-                            {"role": "user", "content": user_prompt},
-                        ],
-                        "temperature": temperature_value,
-                    }
-
-                    # Add JSON mode if requested
-                    if json_mode:
-                        completion_kwargs["response_format"] = {"type": "json_object"}
-
-                    token_key = "max_completion_tokens" if use_completion_tokens else "max_tokens"
-                    completion_kwargs[token_key] = max_tokens
-
-                    try:
-                        completion = client.chat.completions.create(**completion_kwargs)
-                        text = (completion.choices[0].message.content or "").strip()
-                        return completion, text
-                    except bad_request_error as bad_request:
-                        error_msg = str(bad_request)
-                        last_error = bad_request
-                        needs_completion_tokens = (
-                            "max_tokens" in error_msg and "max_completion_tokens" in error_msg
-                        )
-                        needs_default_temp = (
-                            "temperature" in error_msg
-                            and "default (1)" in error_msg
-                            and abs(temperature_value - 1) > 1e-6
-                        )
-                        if needs_completion_tokens and not use_completion_tokens:
-                            use_completion_tokens = True
-                            logger.info(
-                                "OpenAI model %s requires max_completion_tokens; retrying request",
-                                openai_model,
-                            )
-                            print(
-                                "   ⓘ Retrying OpenAI call with max_completion_tokens...",
-                                file=sys.stderr,
-                            )
-                            continue
-                        if needs_default_temp:
-                            temperature_value = 1.0
-                            logger.info(
-                                "OpenAI model %s requires default temperature; retrying request",
-                                openai_model,
-                            )
-                            print(
-                                "   ⓘ Retrying OpenAI call with temperature=1...",
-                                file=sys.stderr,
-                            )
-                            continue
-                        raise
-                raise last_error or RuntimeError(
-                    "Failed to obtain OpenAI response after multiple attempts"
-                )
-
-            def _call_openai_responses_api() -> Tuple[Any, str]:
-                def _int_env(name: str, default: int) -> int:
-                    value = os.getenv(name)
-                    if not value:
-                        return default
-                    try:
-                        return max(1, int(value))
-                    except ValueError:
-                        return default
-
-                def _float_env(name: str, default: float) -> float:
-                    value = os.getenv(name)
-                    if not value:
-                        return default
-                    try:
-                        return float(value)
-                    except ValueError:
-                        return default
-
-                max_output_tokens = _int_env("REALIGN_OPENAI_MAX_OUTPUT_TOKENS", max_tokens)
-                reasoning_effort = os.getenv("REALIGN_OPENAI_REASONING_EFFORT", "medium").strip()
-                responses_temperature = _float_env(
-                    "REALIGN_OPENAI_RESPONSES_TEMPERATURE", temperature
-                )
-
-                inputs: List[Dict[str, str]] = []
-                if system_prompt:
-                    inputs.append({"role": "developer", "content": system_prompt})
-                inputs.append({"role": "user", "content": user_prompt})
-
-                request_kwargs: Dict[str, Any] = {
-                    "model": openai_model,
-                    "input": inputs,
-                    "max_output_tokens": max_output_tokens,
-                    "temperature": responses_temperature,
-                }
-                if reasoning_effort:
-                    request_kwargs["reasoning"] = {"effort": reasoning_effort}
-
-                response = client.responses.create(**request_kwargs)
-                text = _collect_responses_output_text(response)
-                return response, text
-
-            endpoint_type = "responses" if use_responses_api else "chat.completions"
-            if use_responses_api:
-                response, response_text = _call_openai_responses_api()
-            else:
-                response, response_text = _call_openai_chat_completion()
-
-            elapsed = time.time() - start_time
-            response_text = (response_text or "").strip()
-            response_model = getattr(response, "model", openai_model)
-            logger.info(
-                f"OpenAI {endpoint_type} success: {len(response_text)} chars in {elapsed:.2f}s"
-            )
-            logger.debug(f"OpenAI response: {response_text[:200]}...")
-            _emit_debug(
-                {
-                    "event": "llm_response",
-                    "provider": "openai",
-                    "model": response_model,
-                    "elapsed_seconds": elapsed,
-                    "raw_response": response_text,
-                    "purpose": purpose,
-                    "endpoint": endpoint_type,
-                    "response_status": getattr(response, "status", None),
-                }
-            )
-
-            # Log successful response
-            total_elapsed = time.time() - call_start_time
-            call_logger.info(f"LLM CALL SUCCEEDED")
-            call_logger.info(f"Provider: OpenAI (GPT)")
-            call_logger.info(f"Model: {response_model}")
-            call_logger.info(f"Endpoint: {endpoint_type}")
-            call_logger.info(f"Elapsed Time: {elapsed:.2f}s")
-            call_logger.info(f"Total Time: {total_elapsed:.2f}s")
-            call_logger.info(f"Response Length: {len(response_text)} chars")
-            call_logger.info("-" * 80)
-            call_logger.info(f"RESPONSE:\n{response_text}")
-            call_logger.info("=" * 80 + "\n")
-
-            return response_model, response_text
-
-        except ImportError:
-            logger.warning("OpenAI package not installed")
-            if not silent:
-                print("   ❌ OpenAI package not installed", file=sys.stderr)
-            total_elapsed = time.time() - call_start_time
-            call_logger.error(f"LLM CALL FAILED")
-            call_logger.error(f"Provider: OpenAI (GPT)")
-            call_logger.error(f"Reason: OpenAI package not installed")
-            call_logger.error(f"Total Time: {total_elapsed:.2f}s")
-            call_logger.error("=" * 80 + "\n")
-            return None, None
-        except Exception as e:
-            error_msg = str(e)
-            logger.error(f"OpenAI API error: {error_msg}", exc_info=True)
-            if not silent:
-                if "authentication" in error_msg.lower():
-                    print(
-                        "   ❌ OpenAI authentication failed (check API key)",
-                        file=sys.stderr,
-                    )
-                elif "quota" in error_msg.lower() or "billing" in error_msg.lower():
-                    print("   ❌ OpenAI quota/billing issue", file=sys.stderr)
-                else:
-                    print(f"   ❌ OpenAI API error: {e}", file=sys.stderr)
-            total_elapsed = time.time() - call_start_time
-            call_logger.error(f"LLM CALL FAILED")
-            call_logger.error(f"Provider: OpenAI (GPT)")
-            call_logger.error(f"Reason: {error_msg}")
-            call_logger.error(f"Total Time: {total_elapsed:.2f}s")
-            call_logger.error("=" * 80 + "\n")
-            return None, None
+            print("   ❌ Not logged in to Aline cloud", file=sys.stderr)
+        call_logger.error("LLM CLOUD CALL FAILED: not logged in")
+        call_logger.error("=" * 80 + "\n")
+        return None, None
 
-    elif try_openai:
-        logger.debug("OpenAI API key not configured in config file")
+    access_token = get_access_token()
+    if not access_token:
+        logger.warning("Failed to get access token")
         if not silent:
-            print("   ❌ OpenAI API key not configured in config file", file=sys.stderr)
+            print("   ❌ Failed to get access token", file=sys.stderr)
+        call_logger.error("LLM CLOUD CALL FAILED: no access token")
+        call_logger.error("=" * 80 + "\n")
         return None, None
 
-    logger.warning(f"No LLM API keys available (provider: {provider})")
-    if provider == "auto" and not silent:
-        print("   ❌ No LLM API keys configured", file=sys.stderr)
-
-    # Log failure
-    total_elapsed = time.time() - call_start_time
-    call_logger.error(f"LLM CALL FAILED")
-    call_logger.error(f"Reason: No LLM API keys available")
-    call_logger.error(f"Provider: {provider}")
-    call_logger.error(f"Total Time: {total_elapsed:.2f}s")
-    call_logger.error("=" * 80 + "\n")
-
-    return None, None
-
-
-def call_llm_json(
-    system_prompt: str,
-    user_prompt: str,
-    provider: Optional[str] = None,
-    model: Optional[str] = None,
-    max_tokens: Optional[int] = None,
-    temperature: Optional[float] = None,
-    debug_callback: Optional[Callable[[Dict[str, Any]], None]] = None,
-    purpose: str = "generic",
-    silent: bool = False,
-) -> Tuple[Optional[str], Optional[Dict[str, Any]]]:
-    """
-    Call LLM and parse JSON response.
+    # Get backend URL from config
+    from .config import ReAlignConfig
 
-    Args:
-        Same as call_llm()
+    config = ReAlignConfig.load()
+    backend_url = config.share_backend_url or "https://realign-server.vercel.app"
+    endpoint = f"{backend_url}/api/llm/invoke"
+
+    # Build request body
+    request_body: Dict[str, Any] = {
+        "task": task,
+        "payload": payload,
+    }
+    if custom_prompt:
+        request_body["custom_prompt"] = custom_prompt
+    if preset_id:
+        request_body["preset_id"] = preset_id
+
+    if not silent:
+        print(f"   → Calling Aline cloud LLM ({task})...", file=sys.stderr)
+
+    call_logger.info(f"Endpoint: {endpoint}")
+    call_logger.info(f"Request body: {json.dumps(request_body, ensure_ascii=False)[:2000]}")
+    call_logger.info("-" * 80)
 
-    Returns:
-        (model_name, json_dict) where json_dict is None on failure
-    """
-    model_name, response_text = call_llm(
-        system_prompt=system_prompt,
-        user_prompt=user_prompt,
-        provider=provider,
-        model=model,
-        max_tokens=max_tokens,
-        temperature=temperature,
-        json_mode=True,  # Always enable JSON mode for this function
-        debug_callback=debug_callback,
-        purpose=purpose,
-        silent=silent,
-    )
+    try:
+        start_time = time.time()
+        response = httpx.post(
+            endpoint,
+            json=request_body,
+            headers={
+                "Authorization": f"Bearer {access_token}",
+                "Content-Type": "application/json",
+            },
+            timeout=timeout,
+        )
 
-    if not response_text:
-        if not silent:
-            print(
-                f"   ⚠️  LLM returned empty response (purpose={purpose})",
-                file=sys.stderr,
-            )
-        return model_name, None
+        elapsed = time.time() - start_time
+        call_logger.info(f"Response status: {response.status_code}")
+        call_logger.info(f"Response time: {elapsed:.2f}s")
 
-    try:
-        parsed = extract_json(response_text)
-    except Exception as e:
-        logger.warning("Failed to parse LLM JSON (purpose=%s): %s", purpose, e, exc_info=True)
-        if not silent:
-            print(f"   ⚠️  Failed to parse JSON (purpose={purpose}): {e}", file=sys.stderr)
-            print(
-                f"   ⚠️  Response text (first 500 chars): {response_text[:500]}",
-                file=sys.stderr,
-            )
-            print(
-                f"   ⚠️  Response text (last 500 chars): ...{response_text[-500:]}",
-                file=sys.stderr,
-            )
-        return model_name, None
-
-    if not isinstance(parsed, dict):
-        logger.warning("LLM JSON was not an object (purpose=%s): %r", purpose, type(parsed))
-        if not silent:
-            print(
-                f"   ⚠️  LLM returned {type(parsed)} instead of dict (purpose={purpose})",
-                file=sys.stderr,
-            )
-        return model_name, None
+        # Handle HTTP errors
+        if response.status_code == 401:
+            logger.warning("Cloud LLM authentication failed")
+            if not silent:
+                print("   ❌ Cloud LLM authentication failed", file=sys.stderr)
+            call_logger.error("LLM CLOUD CALL FAILED: authentication error (401)")
+            call_logger.error("=" * 80 + "\n")
+            return None, None
 
-    return model_name, parsed
+        if response.status_code == 429:
+            logger.warning("Cloud LLM rate limited")
+            if not silent:
+                print("   ❌ Cloud LLM rate limited", file=sys.stderr)
+            call_logger.error("LLM CLOUD CALL FAILED: rate limited (429)")
+            call_logger.error("=" * 80 + "\n")
+            return None, None
 
+        if response.status_code >= 500:
+            logger.warning(f"Cloud LLM server error: {response.status_code}")
+            if not silent:
+                print(f"   ❌ Cloud LLM server error ({response.status_code})", file=sys.stderr)
+            call_logger.error(f"LLM CLOUD CALL FAILED: server error ({response.status_code})")
+            call_logger.error("=" * 80 + "\n")
+            return None, None
 
-def extract_json(response_text: str) -> Dict[str, Any]:
-    """
-    Extract JSON object from a raw LLM response, handling Markdown fences.
-    Uses strict=False to tolerate control characters in JSON strings.
+        # Parse response
+        data = response.json()
+        call_logger.info(f"Response body: {json.dumps(data, ensure_ascii=False)[:2000]}")
 
-    Args:
-        response_text: Raw LLM response
+        if not data.get("success"):
+            error_msg = data.get("error", "Unknown error")
+            logger.warning(f"Cloud LLM call failed: {error_msg}")
+            if not silent:
+                print(f"   ❌ Cloud LLM error: {error_msg}", file=sys.stderr)
+            call_logger.error(f"LLM CLOUD CALL FAILED: {error_msg}")
+            call_logger.error("=" * 80 + "\n")
+            return None, None
 
-    Returns:
-        Parsed JSON dict
+        model_name = data.get("model", "cloud")
+        result = data.get("result", {})
 
-    Raises:
-        json.JSONDecodeError: If JSON parsing fails
-    """
-    if not response_text:
-        raise json.JSONDecodeError("Empty response", "", 0)
+        # Log success
+        total_elapsed = time.time() - call_start_time
+        call_logger.info("LLM CLOUD CALL SUCCEEDED")
+        call_logger.info(f"Provider: Cloud ({model_name})")
+        call_logger.info(f"Task: {task}")
+        call_logger.info(f"Elapsed Time: {elapsed:.2f}s")
+        call_logger.info(f"Total Time: {total_elapsed:.2f}s")
+        call_logger.info("-" * 80)
+        call_logger.info(f"RESULT: {json.dumps(result, ensure_ascii=False)}")
+        call_logger.info("=" * 80 + "\n")
 
-    json_str = response_text.strip()
+        if not silent:
+            print(f"   ✅ Cloud LLM success ({model_name})", file=sys.stderr)
 
-    # Remove markdown code fences if present
-    if "```json" in response_text:
-        json_start = response_text.find("```json") + 7
-        json_end = response_text.find("```", json_start)
-        if json_end != -1:
-            json_str = response_text[json_start:json_end].strip()
-    elif "```" in response_text:
-        json_start = response_text.find("```") + 3
-        json_end = response_text.find("```", json_start)
-        if json_end != -1:
-            json_str = response_text[json_start:json_end].strip()
+        return model_name, result
 
-    if not json_str:
-        raise json.JSONDecodeError("No JSON content found", response_text, 0)
+    except httpx.TimeoutException:
+        logger.warning(f"Cloud LLM request timed out after {timeout}s")
+        if not silent:
+            print(f"   ❌ Cloud LLM request timed out", file=sys.stderr)
+        total_elapsed = time.time() - call_start_time
+        call_logger.error(f"LLM CLOUD CALL FAILED: timeout after {timeout}s")
+        call_logger.error(f"Total Time: {total_elapsed:.2f}s")
+        call_logger.error("=" * 80 + "\n")
+        return None, None
 
-    # Use strict=False to allow control characters in JSON strings
-    return json.loads(json_str, strict=False)
+    except httpx.RequestError as e:
+        logger.warning(f"Cloud LLM request error: {e}")
+        if not silent:
+            print(f"   ❌ Cloud LLM connection error", file=sys.stderr)
+        total_elapsed = time.time() - call_start_time
+        call_logger.error(f"LLM CLOUD CALL FAILED: request error - {e}")
+        call_logger.error(f"Total Time: {total_elapsed:.2f}s")
+        call_logger.error("=" * 80 + "\n")
+        return None, None
+
+    except Exception as e:
+        logger.error(f"Cloud LLM unexpected error: {e}", exc_info=True)
+        if not silent:
+            print(f"   ❌ Cloud LLM error: {e}", file=sys.stderr)
+        total_elapsed = time.time() - call_start_time
+        call_logger.error(f"LLM CLOUD CALL FAILED: unexpected error - {e}")
+        call_logger.error(f"Total Time: {total_elapsed:.2f}s")
+        call_logger.error("=" * 80 + "\n")
+        return None, None