lucidicai 1.1.20__tar.gz → 3.0.0__tar.gz

lucidicai-3.0.0/PKG-INFO

@@ -0,0 +1,10 @@
+Metadata-Version: 2.1
+Name: lucidicai
+Version: 3.0.0
+Summary: Lucidic AI Python SDK
+Author: Andy Liang
+Author-email: andy@lucidic.ai
+Classifier: Programming Language :: Python :: 3
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Operating System :: OS Independent
+Requires-Python: >=3.6

lucidicai-3.0.0/README.md

@@ -0,0 +1,666 @@
+# Lucidic AI Python SDK
+
+The official Python SDK for [Lucidic AI](https://lucidic.ai), providing comprehensive observability and analytics for LLM-powered applications.
+
+## Features
+
+- **Session & Event Tracking** - Track complex AI workflows with typed, immutable events and automatic nesting
+- **Multi-Provider Support** - Automatic instrumentation for OpenAI, Anthropic, LangChain, Google Generative AI (Gemini), Vertex AI, AWS Bedrock, Cohere, Groq, and more
+- **Real-time Analytics** - Monitor costs, performance, and behavior of your AI applications
+- **Data Privacy** - Built-in masking functions to protect sensitive information
+- **Screenshot Support** - Capture and analyze visual context in your AI workflows
+- **Production Ready** - OpenTelemetry-based instrumentation for enterprise-scale applications
+- **Decorators** - Pythonic decorators for effortless function tracking with automatic nesting
+- **Async Support** - Full support for async/await patterns and concurrent execution
+
+## Installation
+
+```bash
+pip install lucidicai
+```
+
+## Quick Start
+
+```python
+import lucidicai as lai
+from openai import OpenAI
+
+# Initialize the SDK
+session_id = lai.init(
+    session_name="My AI Assistant",
+    providers=["openai"]
+)
+
+# Use your LLM as normal - Lucidic automatically tracks the interaction
+client = OpenAI()
+response = client.chat.completions.create(
+    model="gpt-4",
+    messages=[{"role": "user", "content": "Hello, how are you?"}]
+)
+
+# Events are automatically created and queued for delivery
+# Session automatically ends on process exit (auto_end=True by default)
+```
+
+### Quick Start with Context Manager
+
+```python
+import lucidicai as lai
+from openai import OpenAI
+
+# All-in-one lifecycle: init → bind → run → auto-end at context exit
+with lai.session(session_name="My AI Assistant", providers=["openai"]):
+    client = OpenAI()
+    response = client.chat.completions.create(
+        model="gpt-4",
+        messages=[{"role": "user", "content": "Hello, how are you?"}]
+    )
+# Session automatically ends when exiting the context
+```
+
+## Configuration
+
+### Environment Variables
+
+Create a `.env` file or set these environment variables:
+
+```bash
+LUCIDIC_API_KEY=your_api_key       # Required: Your Lucidic API key
+LUCIDIC_AGENT_ID=your_agent_id     # Required: Your agent identifier
+LUCIDIC_DEBUG=False                 # Optional: Enable debug logging
+LUCIDIC_VERBOSE=False               # Optional: Enable verbose event logging
+```
+
+### Initialization Options
+
+```python
+lai.init(
+    session_name="My Session",              # Optional: Name for this session
+    api_key="...",                          # Optional: Override env var
+    agent_id="...",                         # Optional: Override env var
+    providers=["openai", "anthropic"],      # Optional: LLM providers to track
+    task="Process customer request",        # Optional: High-level task description
+    auto_end=True,                          # Optional: Auto-end session on exit (default: True)
+    masking_function=my_mask_func,          # Optional: Custom PII masking
+    tags=["customer-support", "v1.2"],      # Optional: Session tags
+    evaluators=[...],                          # Optional: Evaluation criteria
+    experiment_id="...",                    # Optional: Link to experiment
+    capture_uncaught=True                   # Optional: Capture crash events (default: True)
+)
+```
+
+## Core Concepts
+
+### Sessions
+A session represents a complete interaction or workflow. Sessions are automatically tracked and can be nested across threads and async tasks.
+
+```python
+# Start a new session
+session_id = lai.init(session_name="Customer Support Chat")
+
+# Update session metadata
+lai.update_session(
+    task="Resolved billing issue",
+    session_eval=0.95,
+    is_successful=True
+)
+
+# End session (or let auto_end handle it)
+lai.end_session(is_successful=True, session_eval=0.9)
+```
+
+### Session Context Management (Thread & Async Safe)
+
+Lucidic uses Python's `contextvars` to bind sessions to the current execution context, ensuring correct attribution in concurrent environments.
+
+#### Pattern 1: Full Lifecycle Management
+
+```python
+import lucidicai as lai
+from openai import OpenAI
+
+# Synchronous version
+with lai.session(session_name="order-flow", providers=["openai"]):
+    OpenAI().chat.completions.create(
+        model="gpt-4",
+        messages=[{"role": "user", "content": "Place order"}]
+    )
+# Session automatically ends at context exit
+```
+
+```python
+# Async version
+import asyncio
+from openai import AsyncOpenAI
+
+async def main():
+    async with lai.session_async(session_name="async-flow", providers=["openai"]):
+        await AsyncOpenAI().chat.completions.create(
+            model="gpt-4",
+            messages=[{"role": "user", "content": "Hello"}]
+        )
+
+asyncio.run(main())
+```
+
+#### Pattern 2: Session Binding (Without Auto-End)
+
+```python
+# Create session without auto-end
+sid = lai.init(session_name="long-running", providers=["openai"], auto_end=False)
+
+# Bind for specific operations
+with lai.bind_session(sid):
+    # Operations here are attributed to this session
+    OpenAI().chat.completions.create(...)
+
+# Session remains open - end manually when ready
+lai.end_session()
+```
+
+#### Pattern 3: Function Wrappers
+
+```python
+def process_request():
+    from openai import OpenAI
+    return OpenAI().chat.completions.create(
+        model="gpt-4",
+        messages=[{"role": "user", "content": "Process this"}]
+    )
+
+# Full lifecycle wrapper
+result = lai.run_session(
+    process_request,
+    init_params={"session_name": "wrapped", "providers": ["openai"]}
+)
+
+# Bind-only wrapper
+sid = lai.init(session_name="manual", providers=["openai"], auto_end=False)
+result = lai.run_in_session(sid, process_request)
+lai.end_session()
+```
+
+### Automatic Session Management
+
+By default, Lucidic automatically ends your session when your process exits:
+
+```python
+# Default behavior - session auto-ends on exit
+lai.init(session_name="My Session")  # auto_end=True by default
+
+# Disable auto-end for manual control
+lai.init(session_name="My Session", auto_end=False)
+# Must call lai.end_session() explicitly
+```
+
+### Events
+
+Events are automatically created when using instrumented providers. All events are typed and immutable once created.
+
+```python
+# Manual event creation with typed payloads
+event_id = lai.create_event(
+    type="function_call",  # or "llm_generation", "error_traceback", "generic"
+    function_name="process_data",
+    arguments={"input": "data"},
+    return_value={"result": "success"},
+    duration=1.5
+)
+```
+
+Event types and their payloads:
+- **llm_generation**: LLM API calls with request/response/usage data
+- **function_call**: Function executions with arguments and return values
+- **error_traceback**: Errors with full traceback information
+- **generic**: General events with custom details
+
+## Provider Integration
+
+### OpenAI
+```python
+from openai import OpenAI
+
+lai.init(session_name="OpenAI Example", providers=["openai"])
+client = OpenAI()
+
+# All OpenAI API calls are automatically tracked
+response = client.chat.completions.create(
+    model="gpt-4",
+    messages=[{"role": "user", "content": "Write a haiku about coding"}]
+)
+```
+
+### Anthropic
+```python
+from anthropic import Anthropic
+
+lai.init(session_name="Claude Example", providers=["anthropic"])
+client = Anthropic()
+
+# Anthropic API calls are automatically tracked
+response = client.messages.create(
+    model="claude-3-opus-20240229",
+    messages=[{"role": "user", "content": "Explain quantum computing"}]
+)
+```
+
+### LangChain
+```python
+from langchain_openai import ChatOpenAI
+from langchain_core.messages import HumanMessage
+
+lai.init(session_name="LangChain Example", providers=["langchain"])
+
+# LangChain calls are automatically tracked
+llm = ChatOpenAI(model="gpt-4")
+response = llm.invoke([HumanMessage(content="Hello!")])
+```
+
+### Google Generative AI (Gemini)
+```python
+import google.generativeai as genai
+
+lai.init(session_name="Gemini Example", providers=["google"])
+genai.configure(api_key=os.getenv("GOOGLE_API_KEY"))
+
+model = genai.GenerativeModel("gemini-1.5-flash")
+response = model.generate_content("Write a haiku about clouds")
+```
+
+### Vertex AI
+```python
+from google.cloud import aiplatform
+from vertexai.generative_models import GenerativeModel
+
+lai.init(session_name="Vertex Example", providers=["vertexai"])
+aiplatform.init(project=os.getenv("GCP_PROJECT"), location="us-central1")
+
+model = GenerativeModel("gemini-1.5-flash")
+response = model.generate_content("Say hello")
+```
+
+### AWS Bedrock
+```python
+import boto3
+
+lai.init(session_name="Bedrock Example", providers=["bedrock"])
+client = boto3.client("bedrock-runtime", region_name="us-east-1")
+
+response = client.invoke_model(
+    modelId="amazon.nova-lite-v1:0",
+    body=b'{"inputText": "Hello from Bedrock"}',
+    contentType="application/json",
+    accept="application/json"
+)
+```
+
+### Cohere
+```python
+import cohere
+
+lai.init(session_name="Cohere Example", providers=["cohere"])
+co = cohere.ClientV2(api_key=os.getenv("COHERE_API_KEY"))
+response = co.chat(
+    model="command-r",
+    messages=[{"role": "user", "content": "Hello"}]
+)
+```
+
+### Groq
+```python
+from groq import Groq
+
+lai.init(session_name="Groq Example", providers=["groq"])
+client = Groq(api_key=os.getenv("GROQ_API_KEY"))
+response = client.chat.completions.create(
+    model="llama-3.1-8b-instant",
+    messages=[{"role": "user", "content": "Hello from Groq"}]
+)
+```
+
+## Advanced Features
+
+### Function Tracking with Decorators
+
+The `@lai.event` decorator automatically tracks function calls as nested events:
+
+```python
+@lai.event()
+def process_data(input_data: dict) -> dict:
+    # Function automatically tracked with arguments and return value
+    result = transform(input_data)
+    return result
+
+# Creates a FUNCTION_CALL event with full tracking
+output = process_data({"key": "value"})
+```
+
+#### Nested Event Tracking
+
+Events automatically nest when functions call other tracked functions:
+
+```python
+@lai.event()
+def outer_function(data: str) -> dict:
+    # This creates a parent event
+    result = inner_function(data)
+    return {"processed": result}
+
+@lai.event()
+def inner_function(data: str) -> str:
+    # This creates a child event nested under outer_function
+    return data.upper()
+
+# Creates nested events with parent-child relationship
+output = outer_function("hello")
+```
+
+#### Error Tracking
+
+The decorator automatically captures exceptions:
+
+```python
+@lai.event()
+def risky_operation(value: int) -> int:
+    if value < 0:
+        raise ValueError("Value must be positive")
+    return value * 2
+
+try:
+    risky_operation(-1)
+except ValueError:
+    pass  # Error is still tracked in the event
+```
+
+#### Async Function Support
+
+```python
+@lai.event()
+async def async_process(url: str) -> dict:
+    async with aiohttp.ClientSession() as session:
+        async with session.get(url) as response:
+            return await response.json()
+
+# Async functions are fully supported
+result = await async_process("https://api.example.com/data")
+```
+
+### Data Masking
+
+Protect sensitive information with custom masking functions:
+
+```python
+def mask_pii(text):
+    # Your PII masking logic here
+    import re
+    # Example: mask email addresses
+    text = re.sub(r'\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Z|a-z]{2,}\b', '[EMAIL]', text)
+    # Example: mask phone numbers
+    text = re.sub(r'\b\d{3}[-.]?\d{3}[-.]?\d{4}\b', '[PHONE]', text)
+    return text
+
+lai.init(
+    session_name="Secure Session",
+    masking_function=mask_pii
+)
+```
+
+### Experiments
+
+Create experiments to group and analyze multiple sessions:
+
+```python
+# Create an experiment
+experiment_id = lai.create_experiment(
+    experiment_name="Prompt Optimization Test",
+    LLM_boolean_evaluators=["response_quality", "latency"],
+    LLM_numeric_evaluators=["coherence", "relevance"],
+    description="Testing different prompt strategies",
+    tags=["A/B-test", "prompts"]
+)
+
+# Link sessions to the experiment
+lai.init(
+    session_name="Test Variant A",
+    experiment_id=experiment_id,
+    providers=["openai"]
+)
+```
+
+### Prompt Management
+
+Fetch and cache prompts from the Lucidic platform:
+
+```python
+prompt = lai.get_prompt(
+    prompt_name="customer_support",
+    variables={"issue_type": "billing", "customer_name": "John"},
+    cache_ttl=3600,  # Cache for 1 hour
+    label="v1.2"
+)
+
+# Variables are replaced in the prompt template
+# {{issue_type}} → "billing"
+# {{customer_name}} → "John"
+```
+
+### Manual Flush
+
+Force flush all pending telemetry data:
+
+```python
+# Ensure all events are sent immediately
+lai.flush(timeout_seconds=2.0)
+```
+
+## Error Handling
+
+The SDK provides specific exceptions for different error scenarios:
+
+```python
+from lucidicai.errors import (
+    APIKeyVerificationError,
+    InvalidOperationError,
+    LucidicNotInitializedError,
+    PromptError
+)
+
+try:
+    lai.init(session_name="My Session")
+except APIKeyVerificationError:
+    print("Invalid API key - check your credentials")
+except LucidicNotInitializedError:
+    print("SDK not initialized - call lai.init() first")
+```
+
+## Crash Event Capture
+
+The SDK automatically captures uncaught exceptions and creates error events:
+
+```python
+lai.init(
+    session_name="my-session",
+    capture_uncaught=True  # Default: True
+)
+
+# If an uncaught exception occurs:
+# 1. An error_traceback event is created with the full traceback
+# 2. The session is ended as unsuccessful
+# 3. Telemetry is flushed before exit
+```
+
+This feature also handles:
+- **SIGINT/SIGTERM signals**: Graceful shutdown with event creation
+- **Thread exceptions**: Main thread exceptions trigger full shutdown
+- **Masking**: Error messages are masked if a masking_function is provided
+
+## Performance & Architecture
+
+### Non-Blocking Event Delivery
+- Events are queued and delivered asynchronously
+- Returns immediately with client-side UUID
+- Background worker handles batching and retries
+
+### Efficient Batching
+- Events batched every 100ms or 100 events
+- Large payloads (>64KB) automatically compressed with gzip
+- Automatic blob storage for oversized events
+
+### Thread & Async Safety
+- Context-aware session binding using contextvars
+- Thread-safe singleton pattern
+- Full async/await support
+
+### OpenTelemetry Integration
+- Industry-standard observability
+- Automatic span → event conversion
+- Configurable export intervals
+
+## Best Practices
+
+1. **Initialize Once**: Call `lai.init()` at the start of your application
+2. **Use Context Managers**: Prefer `with lai.session()` for automatic lifecycle management
+3. **Enable Auto-End**: Let the SDK handle session cleanup (default behavior)
+4. **Handle Errors**: Wrap SDK calls in try-except blocks for production
+5. **Mask Sensitive Data**: Always use masking functions when handling PII
+6. **Leverage Decorators**: Use `@lai.event` for automatic function tracking
+7. **Group Related Work**: Use experiments to analyze A/B tests and variants
+
+## Examples
+
+### Customer Support Bot
+```python
+import lucidicai as lai
+from openai import OpenAI
+
+# Initialize with context manager for automatic cleanup
+with lai.session(
+    session_name="Customer Support",
+    providers=["openai"],
+    task="Handle customer inquiry",
+    tags=["support", "chat"]
+):
+    @lai.event()
+    def analyze_issue(customer_message: str) -> str:
+        """Analyze the customer's issue"""
+        # LLM call is automatically tracked
+        client = OpenAI()
+        response = client.chat.completions.create(
+            model="gpt-4",
+            messages=[
+                {"role": "system", "content": "You are a support analyst"},
+                {"role": "user", "content": f"Categorize this issue: {customer_message}"}
+            ]
+        )
+        return response.choices[0].message.content
+    
+    @lai.event()
+    def generate_response(issue_category: str) -> str:
+        """Generate a helpful response"""
+        # Nested event tracking
+        # ... response generation logic ...
+        return "Response generated"
+    
+    # Process customer request with automatic nesting
+    issue = analyze_issue("I can't login to my account")
+    response = generate_response(issue)
+```
+
+### Data Analysis Pipeline
+```python
+import lucidicai as lai
+import pandas as pd
+from typing import Dict, Any
+
+lai.init(
+    session_name="Quarterly Sales Analysis",
+    providers=["openai"],
+    task="Generate sales insights",
+    auto_end=True  # Session will end when process exits
+)
+
+@lai.event()
+def load_data(file_path: str) -> pd.DataFrame:
+    """Load and validate sales data"""
+    df = pd.read_csv(file_path)
+    # Data loading logic
+    return df
+
+@lai.event()
+def analyze_with_llm(data_summary: Dict[str, Any]) -> str:
+    """Generate insights using GPT-4"""
+    from openai import OpenAI
+    client = OpenAI()
+    
+    response = client.chat.completions.create(
+        model="gpt-4",
+        messages=[
+            {"role": "system", "content": "You are a data analyst"},
+            {"role": "user", "content": f"Analyze this sales data: {data_summary}"}
+        ]
+    )
+    return response.choices[0].message.content
+
+# Execute pipeline with automatic tracking
+df = load_data("sales_q4.csv")
+summary = df.describe().to_dict()
+insights = analyze_with_llm(summary)
+
+# Session automatically ends on process exit
+```
+
+### Concurrent Processing
+```python
+import lucidicai as lai
+import asyncio
+from typing import List
+
+async def process_item(item_id: str, session_id: str) -> dict:
+    """Process a single item with session binding"""
+    # Bind this coroutine to the session
+    with lai.bind_session(session_id):
+        @lai.event()
+        async def fetch_data(id: str) -> dict:
+            # Async operation tracked as nested event
+            await asyncio.sleep(0.1)
+            return {"id": id, "data": "processed"}
+        
+        return await fetch_data(item_id)
+
+async def main():
+    # Create session
+    session_id = lai.init(
+        session_name="Batch Processing",
+        providers=["openai"],
+        auto_end=False  # Manual control for async
+    )
+    
+    # Process items concurrently
+    items = ["item1", "item2", "item3"]
+    tasks = [process_item(item, session_id) for item in items]
+    results = await asyncio.gather(*tasks)
+    
+    # End session manually
+    lai.end_session(is_successful=True)
+    return results
+
+# Run async pipeline
+asyncio.run(main())
+```
+
+## Environment Variables
+
+| Variable | Default | Description |
+|----------|---------|-------------|
+| `LUCIDIC_API_KEY` | Required | API authentication key |
+| `LUCIDIC_AGENT_ID` | Required | Agent identifier |
+| `LUCIDIC_AUTO_END` | True | Auto-end sessions on exit |
+
+## Support
+
+- **Documentation**: [https://docs.lucidic.ai](https://docs.lucidic.ai)
+- **Issues**: [GitHub Issues](https://github.com/Lucidic-AI/Lucidic-Python/issues)
+
+## License
+
+This SDK is distributed under the MIT License.