abstractcore 2.5.2__py3-none-any.whl → 2.6.0__py3-none-any.whl

abstractcore/utils/trace_export.py

@@ -0,0 +1,287 @@
+"""
+Trace export utilities for interaction observability.
+
+Provides functions to export LLM interaction traces to various formats
+for debugging, analysis, and compliance purposes.
+"""
+
+import json
+from typing import List, Dict, Any, Optional, Union
+from pathlib import Path
+from datetime import datetime
+
+
+def export_traces(
+    traces: Union[Dict[str, Any], List[Dict[str, Any]]],
+    format: str = 'jsonl',
+    file_path: Optional[Union[str, Path]] = None
+) -> str:
+    """
+    Export interaction traces to file or return formatted string.
+
+    Args:
+        traces: Single trace dict or list of trace dicts
+        format: Output format - 'jsonl', 'json', or 'markdown'
+        file_path: Optional file path to write to. If None, returns string.
+
+    Returns:
+        Formatted trace data as string
+
+    Raises:
+        ValueError: If format is not supported
+
+    Examples:
+        >>> # Export single trace to JSONL
+        >>> trace = llm.get_traces(trace_id="...")
+        >>> export_traces(trace, format='jsonl', file_path='trace.jsonl')
+
+        >>> # Export multiple traces to JSON
+        >>> traces = llm.get_traces(last_n=10)
+        >>> export_traces(traces, format='json', file_path='traces.json')
+
+        >>> # Get markdown report as string
+        >>> report = export_traces(traces, format='markdown')
+        >>> print(report)
+    """
+    # Normalize to list
+    if isinstance(traces, dict):
+        traces = [traces]
+
+    # Validate format
+    if format not in ['jsonl', 'json', 'markdown']:
+        raise ValueError(f"Unsupported format: {format}. Use 'jsonl', 'json', or 'markdown'")
+
+    # Format based on type
+    if format == 'jsonl':
+        content = _format_as_jsonl(traces)
+    elif format == 'json':
+        content = _format_as_json(traces)
+    elif format == 'markdown':
+        content = _format_as_markdown(traces)
+
+    # Write to file if path provided
+    if file_path:
+        path = Path(file_path)
+        path.parent.mkdir(parents=True, exist_ok=True)
+        with open(path, 'w', encoding='utf-8') as f:
+            f.write(content)
+
+    return content
+
+
+def _format_as_jsonl(traces: List[Dict[str, Any]]) -> str:
+    """Format traces as JSON Lines (one JSON object per line)."""
+    lines = [json.dumps(trace, ensure_ascii=False) for trace in traces]
+    return '\n'.join(lines)
+
+
+def _format_as_json(traces: List[Dict[str, Any]]) -> str:
+    """Format traces as pretty-printed JSON array."""
+    return json.dumps(traces, indent=2, ensure_ascii=False)
+
+
+def _format_as_markdown(traces: List[Dict[str, Any]]) -> str:
+    """Format traces as human-readable markdown report."""
+    lines = ["# LLM Interaction Trace Report", ""]
+    lines.append(f"**Generated:** {datetime.now().isoformat()}")
+    lines.append(f"**Total Interactions:** {len(traces)}")
+    lines.append("")
+
+    for i, trace in enumerate(traces, 1):
+        lines.append(f"## Interaction {i}: {trace.get('trace_id', 'unknown')}")
+        lines.append("")
+
+        # Metadata section
+        lines.append("### Metadata")
+        lines.append(f"- **Timestamp:** {trace.get('timestamp', 'N/A')}")
+        lines.append(f"- **Provider:** {trace.get('provider', 'N/A')}")
+        lines.append(f"- **Model:** {trace.get('model', 'N/A')}")
+
+        # Custom metadata
+        metadata = trace.get('metadata', {})
+        if metadata:
+            for key, value in metadata.items():
+                lines.append(f"- **{key.replace('_', ' ').title()}:** {value}")
+        lines.append("")
+
+        # Input section
+        lines.append("### Input")
+
+        system_prompt = trace.get('system_prompt')
+        if system_prompt:
+            lines.append(f"**System Prompt:**")
+            lines.append(f"```")
+            lines.append(system_prompt)
+            lines.append(f"```")
+            lines.append("")
+
+        prompt = trace.get('prompt', '')
+        lines.append(f"**User Prompt:**")
+        lines.append(f"```")
+        lines.append(prompt)
+        lines.append(f"```")
+        lines.append("")
+
+        # Parameters
+        parameters = trace.get('parameters', {})
+        if parameters:
+            lines.append("**Parameters:**")
+            for key, value in parameters.items():
+                if value is not None:
+                    lines.append(f"- `{key}`: {value}")
+            lines.append("")
+
+        # Tools
+        tools = trace.get('tools')
+        if tools:
+            lines.append(f"**Tools Available:** {len(tools)} tools")
+            lines.append("")
+
+        # Response section
+        response = trace.get('response', {})
+        lines.append("### Response")
+
+        content = response.get('content', '')
+        if content:
+            lines.append(f"**Content:**")
+            lines.append(f"```")
+            lines.append(content)
+            lines.append(f"```")
+            lines.append("")
+
+        # Tool calls
+        tool_calls = response.get('tool_calls')
+        if tool_calls:
+            lines.append(f"**Tool Calls:** {len(tool_calls)}")
+            for call in tool_calls:
+                lines.append(f"- `{call.get('name', 'unknown')}({call.get('arguments', {})})`")
+            lines.append("")
+
+        # Usage metrics
+        usage = response.get('usage', {})
+        if usage:
+            lines.append("**Usage:**")
+            input_tokens = usage.get('input_tokens') or usage.get('prompt_tokens', 0)
+            output_tokens = usage.get('output_tokens') or usage.get('completion_tokens', 0)
+            total_tokens = usage.get('total_tokens', input_tokens + output_tokens)
+
+            lines.append(f"- Input tokens: {input_tokens}")
+            lines.append(f"- Output tokens: {output_tokens}")
+            lines.append(f"- Total tokens: {total_tokens}")
+
+            visual_tokens = usage.get('visual_tokens')
+            if visual_tokens:
+                lines.append(f"- Visual tokens: {visual_tokens}")
+            lines.append("")
+
+        # Performance
+        gen_time = response.get('generation_time_ms')
+        finish_reason = response.get('finish_reason')
+        if gen_time or finish_reason:
+            lines.append("**Performance:**")
+            if gen_time:
+                lines.append(f"- Generation time: {gen_time:.2f}ms")
+            if finish_reason:
+                lines.append(f"- Finish reason: {finish_reason}")
+            lines.append("")
+
+        lines.append("---")
+        lines.append("")
+
+    return '\n'.join(lines)
+
+
+def summarize_traces(traces: Union[Dict[str, Any], List[Dict[str, Any]]]) -> Dict[str, Any]:
+    """
+    Generate summary statistics for traces.
+
+    Args:
+        traces: Single trace dict or list of trace dicts
+
+    Returns:
+        Dictionary with summary statistics:
+            - total_interactions: Number of traces
+            - total_tokens: Sum of all tokens used
+            - total_time_ms: Sum of generation times
+            - avg_tokens_per_interaction: Average tokens
+            - avg_time_ms: Average generation time
+            - providers: Set of providers used
+            - models: Set of models used
+            - date_range: First and last timestamps
+
+    Example:
+        >>> traces = session.get_interaction_history()
+        >>> summary = summarize_traces(traces)
+        >>> print(f"Total interactions: {summary['total_interactions']}")
+        >>> print(f"Total tokens: {summary['total_tokens']}")
+        >>> print(f"Average time: {summary['avg_time_ms']:.2f}ms")
+    """
+    # Normalize to list
+    if isinstance(traces, dict):
+        traces = [traces]
+
+    if not traces:
+        return {
+            'total_interactions': 0,
+            'total_tokens': 0,
+            'total_time_ms': 0,
+            'avg_tokens_per_interaction': 0,
+            'avg_time_ms': 0,
+            'providers': set(),
+            'models': set(),
+            'date_range': None
+        }
+
+    total_tokens = 0
+    total_time_ms = 0
+    providers = set()
+    models = set()
+    timestamps = []
+
+    for trace in traces:
+        # Extract usage
+        response = trace.get('response', {})
+        usage = response.get('usage', {})
+        if usage:
+            total_tokens += usage.get('total_tokens', 0)
+
+        # Extract timing
+        gen_time = response.get('generation_time_ms')
+        if gen_time:
+            total_time_ms += gen_time
+
+        # Extract metadata
+        provider = trace.get('provider')
+        if provider:
+            providers.add(provider)
+
+        model = trace.get('model')
+        if model:
+            models.add(model)
+
+        timestamp = trace.get('timestamp')
+        if timestamp:
+            timestamps.append(timestamp)
+
+    num_traces = len(traces)
+    avg_tokens = total_tokens / num_traces if num_traces > 0 else 0
+    avg_time = total_time_ms / num_traces if num_traces > 0 else 0
+
+    date_range = None
+    if timestamps:
+        timestamps_sorted = sorted(timestamps)
+        date_range = {
+            'first': timestamps_sorted[0],
+            'last': timestamps_sorted[-1]
+        }
+
+    return {
+        'total_interactions': num_traces,
+        'total_tokens': total_tokens,
+        'total_time_ms': total_time_ms,
+        'avg_tokens_per_interaction': avg_tokens,
+        'avg_time_ms': avg_time,
+        'providers': list(providers),
+        'models': list(models),
+        'date_range': date_range
+    }

abstractcore/utils/version.py

@@ -11,4 +11,4 @@ including when the package is installed from PyPI where pyproject.toml is not av
 
 # Package version - update this when releasing new versions
 # This must be manually synchronized with the version in pyproject.toml
-__version__ = "2.5.2"
+__version__ = "2.6.0"