agentops/docs/v1/concepts/decorators.mdx

151 lines
4.0 KiB
Plaintext

---
title: "Decorators"
description: "Seemingly magic tools that can be added to functions and classes for easier instrumenting."
---
> Decorators work by wrapping functions or classes that they are placed above. You've probably seen this before.
> Using decorators allows us to add a lot of functionality to your code with minimal work on your part.
> ```python python
> @example_decorator()
> def hello_world():
> ...
> ```
AgentOps provides a set of decorators that allow you to easily instrument your code for tracing and monitoring AI agent workflows. These decorators create spans (units of work) that are organized hierarchically to track different types of operations.
## `@session`
The `@session` decorator creates a session span, which serves as the root for all other spans. No spans can exist without a session at the top.
```python
from agentops.sdk.decorators import session
@session
def my_workflow():
# Your session code here
return result
```
You can also use the decorator with parameters:
```python
@session(name="custom-session-name", version=1)
def my_workflow():
# Your session code here
return result
```
## `@agent`
The `@agent` decorator creates an agent span for tracking agent operations. Agent spans are typically children of session spans and parents of operation spans.
```python
from agentops.sdk.decorators import agent
@agent
class MyAgent:
def __init__(self, name):
self.name = name
# Agent methods here
```
You can also specify a custom name for the agent:
```python
@agent(name="research-assistant")
class MyAgent:
# Agent implementation
```
## `@operation` / `@task`
The `@operation` and `@task` decorators are aliases that create operation/task spans for tracking specific operations. These spans are typically children of agent spans.
```python
from agentops.sdk.decorators import agent, operation
@agent
class MyAgent:
@operation
def perform_task(self, task):
# Operation implementation
return result
```
Operations can also be used outside of agent classes:
```python
from agentops.sdk.decorators import operation
@operation
def process_data(data):
# Process the data
return result
```
## `@workflow`
The `@workflow` decorator creates workflow spans for tracking workflows, which can contain multiple operations.
```python
from agentops.sdk.decorators import workflow
@workflow
def my_workflow(data):
# Workflow implementation
return result
```
## Nesting and Hierarchy
The decorators automatically manage the context propagation, ensuring that spans are properly nested within their parent spans. The typical hierarchy is:
1. Session (root)
2. Agent
3. Operation/Task
4. Nested Operations
Example of proper nesting:
```python
from agentops.sdk.decorators import session, agent, operation
@agent
class MyAgent:
@operation
def nested_operation(self, message):
return f"Processed: {message}"
@operation
def main_operation(self):
result = self.nested_operation("test message")
return result
@session
def my_session():
agent = MyAgent()
return agent.main_operation()
# Run the session
result = my_session()
```
## Additional Features
The decorators provide several additional features:
1. **Input/Output Recording**: The decorators automatically record the input arguments and output results of the decorated functions.
2. **Exception Handling**: If an exception occurs within a decorated function, it's recorded in the span.
3. **Support for Different Function Types**: The decorators handle different types of functions:
- Regular synchronous functions
- Asynchronous functions (using `async/await`)
- Generator functions (using `yield`)
- Asynchronous generator functions (using `async` and `yield`)
4. **Custom Attributes**: You can add custom attributes to spans using the decorator parameters.
<script type="module" src="/scripts/github_stars.js"></script>
<script type="module" src="/scripts/adjust_api_dynamically.js"></script>