ant-ai 1.7.0

A lightweight Python framework for building tool-driven AI agents and multi-agent systems.

ant-ai is a lightweight Python framework for building multi-agent systems: graph-based workflow orchestration, first-class agent-to-agent (A2A) communication via the A2A protocol, MCP tool integration, lifecycle hooks for guardrails, and built-in observability — all on top of an LLM-agnostic core.

Why ANT AI

Multi-agent by design. Agents communicate and delegate via the A2A protocol — no custom glue code required.

No lock-in. Swap LLMs, tools, or observability backends without touching your agent logic.

Structured, not scripted. Model complex behavior as graphs — know exactly what runs, when, and why.

Observable from day one. Built-in tracing via Langfuse and lifecycle hooks for guardrails.

Installation

Requires Python 3.14+. Install with uv:

uv add ant-ai

Or install with optional extras:

uv add "ant-ai[all]"
uv add "ant-ai[openai]"
uv add "ant-ai[langfuse]"
uv add "ant-ai[mem0]"
uv add "ant-ai[viz]"
uv add "ant-ai[guardrails-ai]"
uv add "ant-ai[datafog]"

The guardrails-ai extra installs the core library. Validators (e.g. ToxicLanguage, DetectPII) are distributed via the Guardrails Hub and must be installed separately after running guardrails configure:

# Run from outside the project directory so uv doesn't pick up exclude-newer
(cd /tmp && guardrails hub install hub://guardrails/toxic_language hub://guardrails/detect_pii)
# Copy the registry into the project root so imports resolve from here
cp /tmp/.guardrails/hub_registry.json .guardrails/

Note: hub validators are installed outside of uv's lockfile and must be reinstalled after uv sync. The cd /tmp wrapper is required because these packages ship without upload dates, which conflicts with the project's exclude-newer = "P2D" supply-chain setting. Running from /tmp (where no pyproject.toml exists) lets uv install them without that constraint while still targeting the active virtualenv. The hub_registry.json copy is needed because guardrails resolves hub imports from a .guardrails/hub_registry.json file relative to the current working directory.

The datafog extra powers PIIGuardrailHook, a native, dependency-light guardrail hook that scans LLM output for PII and retries/blocks on detection — an alternative to GuardrailsAIHook + DetectPII that needs no separate hub install. For guardrailing on arbitrary criteria (e.g. "no medical advice"), subclass LLMGuardrailHook — an LLM-as-judge base class that needs no extra dependency.

Or clone and sync for local development:

git clone git@github.com:idea-idsia/ant-ai.git
cd ant-ai
uv sync --all-packages --all-groups --all-extras

Quickstart

Single agent

from ant_ai import Agent, Message, State, tool
from ant_ai.llm.integrations import LiteLLMChat


@tool
def get_weather(city: str) -> str:
    """Return the current weather for a city."""
    return f"Sunny, 22°C in {city}"


llm = LiteLLMChat(model="gpt-4o-mini")

agent = Agent(
    name="WeatherAgent",
    system_prompt="You are a helpful weather assistant.",
    llm=llm,
    tools=[get_weather],
)

state = State(messages=[Message(role="user", content="What's the weather in Lugano?")])
answer = agent.invoke(state)
print(answer)

Streaming events

from ant_ai.core import FinalAnswerEvent

async for event in agent.stream(state):
    if isinstance(event, FinalAnswerEvent):
        print(event.content)

Structured output

from pydantic import BaseModel


class WeatherReport(BaseModel):
    city: str
    temperature: int
    condition: str


answer = agent.invoke(state, response_schema=WeatherReport)
# answer is a JSON string matching WeatherReport

Development

# Install dev dependencies and pre-commit hooks
uv sync --all-extras
uv run pre-commit install

# Run tests
uv run pytest

# Serve docs locally
uv run mkdocs serve

See CONTRIBUTING.md for the full contributing guide, branching model, and review process.

License

This software is licensed under the MIT license. See the LICENSE file for details.

Citation

If you use ant-ai in your research, please cite it. See CITATION.cff for the machine-readable citation metadata, or use the BibTeX entry below.

@software{Sas_ant-ai_A_lightweight_2026,
author = {Sas, Cezar and Giuffrida, Vincenzo and Mitrović, Sandra and Salani, Matteo},
doi = {10.5281/zenodo.21276625},
license = {MIT},
month = jun,
title = {{ant-ai: A lightweight Python framework for building multi-agent AI systems}},
url = {https://github.com/idea-idsia/ant-ai},
year = {2026}
}

Funding

This project is supported by the following grants.

Acknowledgement
Funded by the Swiss State Secretariat for Education, Research and Innovation (SERI), Project number 24.00596.
Funded by the European Union under Grant Agreement No. 101189745 (HIVEMIND).