agent-lifecycle-toolkit

Delivering plug-and-play, framework-agnostic technology to boost agents' performance

What is ALTK?

The Agent Lifecycle Toolkit helps agent builders create better performing agents by easily integrating our components into agent pipelines. The components help improve the performance of agents by addressing key gaps in various stages of the agent lifecycle, such as in reasoning, or tool calling errors, or output guardrails.

👉 Read the launch blog to learn more about the vision and capabilities.

Installation

To use ALTK, simply install agent-lifecycle-toolkit from your package manager, e.g. pip:

pip install agent-lifecycle-toolkit

More detailed installation instructions are available in the docs.

Getting Started

Below is an end-to-end example that you can quickly get your hands dirty with. The example has a langgraph agent, a weather tool, and a component that checks for silent errors. Refer to the examples folder for this example and others. The below example will additionally require the langgraph and langchain-anthropic packages along with setting two environment variables.

import random

from langgraph.prebuilt import create_react_agent
from langchain_core.tools import tool
from typing_extensions import Annotated
from langgraph.prebuilt import InjectedState

from altk.post_tool.silent_review.silent_review import SilentReviewForJSONDataComponent
from altk.post_tool.core.toolkit import SilentReviewRunInput, Outcome
from altk.core.toolkit import AgentPhase

# Ensure that the following environment variables are set:
# ANTHROPIC_API_KEY = *** anthropic api key ***
# ALTK_MODEL_NAME = anthropic/claude-sonnet-4-20250514

@tool
def get_weather(city: str, state: Annotated[dict, InjectedState]) -> str:
    """Get weather for a given city."""
    if random.random() >= 0.500:
        # Simulates a silent error from an external service
        result = {"weather": "Weather service is under maintenance."}
    else:
        result = {"weather": f"It's sunny and 70F in {city}!"}

    # Use SilentReview component to check if it's a silent error
    review_input = SilentReviewRunInput(messages=state["messages"], tool_response=result)
    reviewer = SilentReviewForJSONDataComponent()
    review_result = reviewer.process(data=review_input, phase=AgentPhase.RUNTIME)

    if review_result.outcome == Outcome.NOT_ACCOMPLISHED:
        # Agent should retry tool call if silent error was detected
        return "Silent error detected, retry the get_weather tool!"
    else:
        return result

agent = create_react_agent(
    model="anthropic:claude-sonnet-4-20250514",
    tools=[get_weather],
    prompt="You are a helpful assistant"
)

# Runs the agent
result = agent.invoke(
    {"messages": [{"role": "user", "content": "what is the weather in sf"}]}
)
# Show the final result which should not be that the service is in maintenance.
print(result["messages"][-1].content)

Features

Documentation

Check out ALTK's documentation, for details on installation, usage, concepts, and more.

The ALTK supports multiple LLM providers and two methods of configuring the providers. For more information, see the LLMClient documentation.

Examples

Go hands-on with our examples.

Integrations

To further accelerate your AI application development, check out ALTK's native integrations with popular frameworks and tools.

Get Help and Support

Please feel free to connect with us using the discussion section.

Contributing Guidelines

ALTK is open-source and we ❤️ contributions.

To help build ALTK, take a look at our: Contribution guidelines

Bugs

We use GitHub Issues to manage bugs. Before filing a new issue, please check to make sure it hasn't already been logged.

Code of Conduct

This project and everyone participating in it are governed by the Code of Conduct. By participating, you are expected to uphold this code. Please read the full text so that you know which actions may or may not be tolerated.

License

The ALTK codebase is under Apache 2.0 license. For individual model usage, please refer to the model licenses in the original packages.

Contributors

Thanks to all of our contributors who make this project possible. Special thanks to the Global Agentic Middleware team in IBM Research for all the contributions from the many different teams and people.

Paper and Citation

For more information, please see our accepted paper from the Demonstration Track in CAIS2026.

If you want to cite that paper or this work, please use this reference:

@inproceedings{10.1145/3786335.3813206,
    author = {Wright, Zidane and Tsay, Jason and Murthi, Anupama and Elhadad, Osher and Del Rio, Diego and Goyal, Saurabh and Kate, Kiran and Laredo, Jim A. and Lazar, Koren and Muthusamy, Vinod and Rizk, Yara},
    title = {Agent Lifecycle Toolkit (ALTK): Reusable Middleware Components for Robust AI Agents},
    year = {2026},
    isbn = {9798400724152},
    publisher = {Association for Computing Machinery},
    address = {New York, NY, USA},
    url = {https://doi.org/10.1145/3786335.3813206},
    doi = {10.1145/3786335.3813206},
    booktitle = {Proceedings of the ACM Conference on AI and Agentic Systems},
    pages = {1340–1344},
    numpages = {5},
    keywords = {AI Systems, AI Agents, Agentic Middleware},
    series = {CAIS '26}
}