Skip to content

Quick Start

Get your first trace in 5 minutes with Prela's auto-instrumentation.

Step 1: Get Your API Key

Sign in at dashboard.prela.dev, go to API Keys, and create a new key. Then set it as an environment variable:

export PRELA_API_KEY="prela_sk_..."

Step 2: Install Prela

pip install prela

Step 3: Initialize Prela

Add one line to your application:

import prela

# Initialize with your service name
# Traces are sent to Prela Cloud automatically when PRELA_API_KEY is set
prela.init(service_name="my-agent")

That's it! Prela will automatically instrument any LLM SDKs you use and send traces to your Prela dashboard.

Step 4: Run Your Agent

Use your LLM SDK as normal:

import prela
from anthropic import Anthropic

# Initialize Prela
prela.init(service_name="my-agent")

# Use Anthropic normally - automatic tracing!
client = Anthropic(api_key="sk-...")
response = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Explain quantum computing"}
    ]
)

print(response.content[0].text)

import prela
from openai import OpenAI

# Initialize Prela
prela.init(service_name="my-agent")

# Use OpenAI normally - automatic tracing!
client = OpenAI(api_key="sk-...")
response = client.chat.completions.create(
    model="gpt-4",
    messages=[
        {"role": "user", "content": "Explain quantum computing"}
    ]
)

print(response.choices[0].message.content)

import prela
from langchain.llms import OpenAI
from langchain.agents import initialize_agent, Tool

# Initialize Prela
prela.init(service_name="my-agent")

# Create agent - automatic tracing!
llm = OpenAI(temperature=0)
tools = [
    Tool(
        name="Calculator",
        func=lambda x: eval(x),
        description="Useful for math"
    )
]

agent = initialize_agent(
    tools,
    llm,
    agent="zero-shot-react-description"
)

result = agent.run("What is 25 * 4?")
print(result)

Step 5: View Your Traces

Traces appear in your Prela dashboard in real time. If PRELA_API_KEY is not set, traces fall back to printing to the console:

{
  "trace_id": "abc123...",
  "span_id": "def456...",
  "name": "anthropic.messages.create",
  "span_type": "llm",
  "status": "success",
  "started_at": "2025-01-26T10:30:00.123456Z",
  "ended_at": "2025-01-26T10:30:01.234567Z",
  "duration_ms": 1111.1,
  "attributes": {
    "service.name": "my-agent",
    "llm.vendor": "anthropic",
    "llm.model": "claude-sonnet-4-20250514",
    "llm.input_tokens": 50,
    "llm.output_tokens": 200,
    "llm.total_tokens": 250,
    "llm.latency_ms": 1111.1
  },
  "events": [
    {
      "name": "llm.request",
      "timestamp": "2025-01-26T10:30:00.123456Z",
      "attributes": {
        "messages": [{"role": "user", "content": "Explain quantum computing"}]
      }
    },
    {
      "name": "llm.response",
      "timestamp": "2025-01-26T10:30:01.234567Z",
      "attributes": {
        "content": "Quantum computing is..."
      }
    }
  ]
}

What's Captured?

Every trace automatically includes:

Request Data

  • Model name (llm.model)
  • Temperature, max tokens, etc.
  • Input messages or prompt
  • System prompt (if provided)

Response Data

  • Token usage (input, output, total)
  • Completion reason (stop, length, tool_use)
  • Response content
  • Model used (actual model from API)

Performance

  • Total latency in milliseconds
  • Time-to-first-token (for streaming)
  • Start and end timestamps

Tool Usage

  • Tool/function calls detected
  • Tool names and arguments
  • Tool call IDs

Errors

  • Exception type and message
  • Stack traces
  • Error status codes

Export to File

Save traces to a file instead of console:

import prela

prela.init(
    service_name="my-agent",
    exporter="file",
    directory="./traces"  # Traces saved here
)

Traces are saved in JSONL format (one JSON object per line):

cat traces/2025-01-26/trace_abc123.json

Custom Spans

Create custom spans for your own functions:

import prela

tracer = prela.get_tracer()

def my_function():
    with tracer.span("my_operation", span_type="custom"):
        # Your code here
        result = expensive_computation()
        return result

CLI Commands

Use the CLI to explore traces:

# List recent traces
prela list

# Show trace details
prela show trace_abc123

# Search for traces
prela search --status error --service my-agent

What's Next?

Now that you have basic tracing working:

Troubleshooting

No traces appearing?

Check that:

  1. PRELA_API_KEY is set in your environment (echo $PRELA_API_KEY)
  2. You called prela.init() before using your LLM SDK
  3. Your LLM SDK is installed (pip list | grep anthropic)
  4. Auto-instrumentation is enabled (default)

Want to disable auto-instrumentation?

prela.init(service_name="my-agent", auto_instrument=False)

Then manually instrument:

from prela.instrumentation import AnthropicInstrumentor

instrumentor = AnthropicInstrumentor()
instrumentor.instrument(tracer=prela.get_tracer())

Need help?