Observability Overview
Capture every LLM call, tool invocation, and agent decision with automatic instrumentation.
What is Observability?
Observability in AI systems means having complete visibility into what your agents are doing, why they're making decisions, and how they're performing. TuringPulse provides:
- Traces - End-to-end execution paths through your agent
- Spans - Individual operations within a trace (LLM calls, tools, etc.)
- Metrics - Quantitative measurements (latency, tokens, cost)
- Logs - Detailed event data for debugging
Quick Start
Add observability to your agent in three steps:
1. Install the SDK
pip install turingpulse_sdk2. Initialize
config.py
from turingpulse_sdk import init
init(api_key="sk_live_your_api_key", workflow_name="my-agent")3. Instrument Your Code
agent.py
from turingpulse_sdk import instrument
@instrument(name="my-agent")
def run_agent(query: str):
response = llm.chat(query)
return response💡
Zero-Config Instrumentation
If you're using LangGraph, LangChain, or CrewAI, the SDK automatically captures traces without any code changes. Just initialize and go!
What Gets Captured
TuringPulse automatically captures rich telemetry from your AI workflows:
LLM Calls
- Model name and provider
- Input messages/prompts
- Output completions
- Token counts (input/output)
- Latency and cost
- Temperature and other parameters
Tool Invocations
- Tool name and description
- Input arguments
- Output results
- Execution time
- Success/failure status
Agent Decisions
- Reasoning steps (for ReAct agents)
- Plan steps (for Plan-and-Execute)
- State transitions
- Iteration counts
Viewing Traces
Once traces are captured, you can view them in the TuringPulse dashboard:
- Trace List - Browse all traces with filtering and search
- Trace Detail - Drill into individual traces with span waterfall
- Span Inspector - View inputs, outputs, and metadata for each span
- Timeline View - Visualize execution timing and parallelism
Next Steps
- Log Traces - Detailed tracing guide
- Log Conversations - Multi-turn tracking
- Log User Feedback - Capture ratings
- Cost Tracking - Monitor spend