Observability and DevTool platform for AI Agents

🐦 Twitter   •   📢 Discord   •   🖇️ Dashboard   •   📙 Documentation

AgentOps helps developers build, evaluate, and monitor AI agents. From prototype to production.

📊 Replay Analytics and Debugging	Step-by-step agent execution graphs
💸 LLM Cost Management	Track spend with LLM foundation model providers
🧪 Agent Benchmarking	Test your agents against 1,000+ evals
🔐 Compliance and Security	Detect common prompt injection and data exfiltration exploits
🤝 Framework Integrations	Native Integrations with CrewAI, AutoGen, & LangChain

Quick Start ⌨️

pip install agentops

Session replays in 2 lines of code

Initialize the AgentOps client and automatically get analytics on all your LLM calls.

Get an API key

import agentops

# Beginning of your program (i.e. main.py, __init__.py)
agentops.init( < INSERT YOUR API KEY HERE >)

...

# End of program
agentops.end_session('Success')

All your sessions can be viewed on the AgentOps dashboard

Agent Debugging

Session Replays

Summary Analytics

First class Developer Experience

Add powerful observability to your agents, tools, and functions with as little code as possible: one line at a time.
Refer to our documentation

# Automatically associate all Events with the agent that originated them
from agentops import track_agent

@track_agent(name='SomeCustomName')
class MyAgent:
  ...

# Automatically create ToolEvents for tools that agents will use
from agentops import record_tool

@record_tool('SampleToolName')
def sample_tool(...):
  ...

# Automatically create ActionEvents for other functions.
from agentops import record_action

@agentops.record_action('sample function being record')
def sample_function(...):
  ...

# Manually record any other Events
from agentops import record, ActionEvent

record(ActionEvent("received_user_input"))

Integrations 🦾

CrewAI 🛶

Build Crew agents with observability with only 2 lines of code. Simply set an AGENTOPS_API_KEY in your environment, and your crews will get automatic monitoring on the AgentOps dashboard.

pip install 'crewai[agentops]'

• AgentOps integration example
• Official CrewAI documentation

AutoGen 🤖

With only two lines of code, add full observability and monitoring to Autogen agents. Set an AGENTOPS_API_KEY in your environment and call agentops.init()

• Autogen Observability Example
• Autogen - AgentOps Documentation

Langchain 🦜🔗

AgentOps works seamlessly with applications built using Langchain. To use the handler, install Langchain as an optional dependency:

Installation

pip install agentops[langchain]

To use the handler, import and set

import os
from langchain.chat_models import ChatOpenAI
from langchain.agents import initialize_agent, AgentType
from agentops.partners.langchain_callback_handler import LangchainCallbackHandler

AGENTOPS_API_KEY = os.environ['AGENTOPS_API_KEY']
handler = LangchainCallbackHandler(api_key=AGENTOPS_API_KEY, tags=['Langchain Example'])

llm = ChatOpenAI(openai_api_key=OPENAI_API_KEY,
                 callbacks=[handler],
                 model='gpt-3.5-turbo')

agent = initialize_agent(tools,
                         llm,
                         agent=AgentType.CHAT_ZERO_SHOT_REACT_DESCRIPTION,
                         verbose=True,
                         callbacks=[handler], # You must pass in a callback handler to record your agent
                         handle_parsing_errors=True)

Check out the Langchain Examples Notebook for more details including Async handlers.

Cohere ⌨️

First class support for Cohere(>=5.4.0). This is a living integration, should you need any added functionality please message us on Discord!

• AgentOps integration example
• Official Cohere documentation

Installation

pip install cohere

import cohere
import agentops

# Beginning of program's code (i.e. main.py, __init__.py)
agentops.init(<INSERT YOUR API KEY HERE>)
co = cohere.Client()

chat = co.chat(
    message="Is it pronounced ceaux-hear or co-hehray?"
)

print(chat)

agentops.end_session('Success')

import cohere
import agentops

# Beginning of program's code (i.e. main.py, __init__.py)
agentops.init(<INSERT YOUR API KEY HERE>)

co = cohere.Client()

stream = co.chat_stream(
    message="Write me a haiku about the synergies between Cohere and AgentOps"
)

for event in stream:
    if event.event_type == "text-generation":
        print(event.text, end='')

agentops.end_session('Success')

Anthropic ﹨

Track agents built with the Anthropic Python SDK (>=0.32.0).

• AgentOps integration example
• Official Anthropic documentation

Installation

pip install anthropic

import anthropic
import agentops

# Beginning of program's code (i.e. main.py, __init__.py)
agentops.init(<INSERT YOUR API KEY HERE>)

client = anthropic.Anthropic(
    # This is the default and can be omitted
    api_key=os.environ.get("ANTHROPIC_API_KEY"),
)

message = client.messages.create(
        max_tokens=1024,
        messages=[
            {
                "role": "user",
                "content": "Tell me a cool fact about AgentOps",
            }
        ],
        model="claude-3-opus-20240229",
    )
print(message.content)

agentops.end_session('Success')

Streaming

import anthropic
import agentops

# Beginning of program's code (i.e. main.py, __init__.py)
agentops.init(<INSERT YOUR API KEY HERE>)

client = anthropic.Anthropic(
    # This is the default and can be omitted
    api_key=os.environ.get("ANTHROPIC_API_KEY"),
)

stream = client.messages.create(
    max_tokens=1024,
    model="claude-3-opus-20240229",
    messages=[
        {
            "role": "user",
            "content": "Tell me something cool about streaming agents",
        }
    ],
    stream=True,
)

response = ""
for event in stream:
    if event.type == "content_block_delta":
        response += event.delta.text
    elif event.type == "message_stop":
        print("\n")
        print(response)
        print("\n")

Async

import asyncio
from anthropic import AsyncAnthropic

client = AsyncAnthropic(
    # This is the default and can be omitted
    api_key=os.environ.get("ANTHROPIC_API_KEY"),
)


async def main() -> None:
    message = await client.messages.create(
        max_tokens=1024,
        messages=[
            {
                "role": "user",
                "content": "Tell me something interesting about async agents",
            }
        ],
        model="claude-3-opus-20240229",
    )
    print(message.content)


await main()

LiteLLM 🚅

AgentOps provides support for LiteLLM(>=1.3.1), allowing you to call 100+ LLMs using the same Input/Output Format.

• AgentOps integration example
• Official LiteLLM documentation

Installation

pip install litellm

# Do not use LiteLLM like this
# from litellm import completion
# ...
# response = completion(model="claude-3", messages=messages)

# Use LiteLLM like this
import litellm
...
response = litellm.completion(model="claude-3", messages=messages)
# or
response = await litellm.acompletion(model="claude-3", messages=messages)

LlamaIndex 🦙

AgentOps works seamlessly with applications built using LlamaIndex, a framework for building context-augmented generative AI applications with LLMs.

Installation

pip install llama-index-instrumentation-agentops

To use the handler, import and set

from llama_index.core import set_global_handler

# NOTE: Feel free to set your AgentOps environment variables (e.g., 'AGENTOPS_API_KEY')
# as outlined in the AgentOps documentation, or pass the equivalent keyword arguments
# anticipated by AgentOps' AOClient as **eval_params in set_global_handler.

set_global_handler("agentops")

Check out the LlamaIndex docs for more details.

Time travel debugging 🔮

Try it out!

Agent Arena 🥊

(coming soon!)

Evaluations Roadmap 🧭

Platform	Dashboard	Evals
✅ Python SDK	✅ Multi-session and Cross-session metrics	✅ Custom eval metrics
🚧 Evaluation builder API	✅ Custom event tag tracking	🔜 Agent scorecards
✅ Javascript/Typescript SDK	✅ Session replays	🔜 Evaluation playground + leaderboard

Debugging Roadmap 🧭

Performance testing	Environments	LLM Testing	Reasoning and execution testing
✅ Event latency analysis	🔜 Non-stationary environment testing	🔜 LLM non-deterministic function detection	🚧 Infinite loops and recursive thought detection
✅ Agent workflow execution pricing	🔜 Multi-modal environments	🚧 Token limit overflow flags	🔜 Faulty reasoning detection
🚧 Success validators (external)	🔜 Execution containers	🔜 Context limit overflow flags	🔜 Generative code validators
🔜 Agent controllers/skill tests	✅ Honeypot and prompt injection detection (PromptArmor)	🔜 API bill tracking	🔜 Error breakpoint analysis
🔜 Information context constraint testing	🔜 Anti-agent roadblocks (i.e. Captchas)	🔜 CI/CD integration checks
🔜 Regression testing	🔜 Multi-agent framework visualization

Why AgentOps? 🤔

Without the right tools, AI agents are slow, expensive, and unreliable. Our mission is to bring your agent from prototype to production. Here's why AgentOps stands out:

• Comprehensive Observability: Track your AI agents' performance, user interactions, and API usage.
• Real-Time Monitoring: Get instant insights with session replays, metrics, and live monitoring tools.
• Cost Control: Monitor and manage your spend on LLM and API calls.
• Failure Detection: Quickly identify and respond to agent failures and multi-agent interaction issues.
• Tool Usage Statistics: Understand how your agents utilize external tools with detailed analytics.
• Session-Wide Metrics: Gain a holistic view of your agents' sessions with comprehensive statistics.

AgentOps is designed to make agent observability, testing, and monitoring easy.

Star History

Check out our growth in the community:

Popular projects using AgentOps

Repository	Stars
geekan / MetaGPT	42787
run-llama / llama_index	34446
crewAIInc / crewAI	18287
camel-ai / camel	5166
superagent-ai / superagent	5050
iyaja / llama-fs	4713
BasedHardware / Omi	2723
MervinPraison / PraisonAI	2007
AgentOps-AI / Jaiqu	272
strnad / CrewAI-Studio	134
alejandro-ao / exa-crewai	55
tonykipkemboi / youtube_yapper_trapper	47
sethcoast / cover-letter-builder	27
bhancockio / chatgpt4o-analysis	19
breakstring / Agentic_Story_Book_Workflow	14
MULTI-ON / multion-python	13

Generated using github-dependents-info, by Nicolas Vuillamy