ABZ AgentThe Enterprise Agent
Build & Runtime
ABZ Agent is a powerful, lightweight framework for building, running, and scaling AI agents in production.
LightweightMinimal. Fast. Powerful.from abz_agent import Agent, Runner
agent = Agent(
name="ResearchAgent",
model="gemini-1.5-pro",
tools=[search_web, read_file, write_file],
instructions="You are a helpful research assistant."
)
result = Runner.run(agent, "Summarize the latest AI trends.")Minimal boilerplate
Define an agent in a handful of lines — no config files, no scaffolding to maintain.
Model-agnostic
Switch between Gemini and Groq with a single parameter — your agent code stays the same.
Open source
Built in the open and shaped by the developers who use it every day.
Everything you need. Nothing you don't.
A focused toolkit for building AI agents — not a general-purpose orchestration framework.
Clean API
Create and run an agent in a handful of readable lines of Python.
Multi-model
Gemini and Groq today, with more providers on the way — swap models without rewriting logic.
Custom tools
Extend an agent with your own Python functions — no special interfaces to learn.
Lightweight core
A small dependency footprint that installs fast and stays out of your way.
Fast to learn
If you know Python, you already know most of ABZ Agent.
Community-driven
Open source and shaped by the people building with it.
See it run.
Pick a prompt below or hit Run and ask your own — a taste of how ABZ Helper responds.
from abagentsdk import Agent agent = Agent( name="ABZ Helper", instructions="Be concise and use tools efficiently.", model="groq/llama-3.3-70b-versatile", )print(agent.run("What is 2 + 2?").content)
Supported models
Pick a provider per agent, or mix providers across a project. The API doesn't change.
Gemini
Fast, capable models for everyday agent workloads.
model="gemini-2.0-flash"
Groq
Low-latency inference for agents that need to respond instantly.
model="groq/llama-3.3-70b-versatile"Built for developer experience.
Readable errors that point straight at the problem.
Works with your existing Python tooling — no special runtime.
Sensible defaults, with every part overridable when you need it.
Documentation that matches the code, not a roadmap of it.
$ pip install abagentsdk
Successfully installed abagentsdkHow it works
Four pieces, one flow. An agent takes instructions, reasons with a model, calls tools when it needs to, and returns a response.
Instructions
You describe what the agent should do, in plain language.
Model
Gemini or Groq reasons over the request and decides what to do next.
Tools
Custom Python functions the agent can call to act or fetch data.
Response
A finished result, returned straight to your application code.
Join the community
ABZ Agent is built in the open. Follow along, ask questions, and contribute.
Frequently asked questions
An open-source Python framework for building AI agents — a small, focused API for defining an agent's instructions, model, and tools.
Start building your first agent.
Open source. Multi-model. Ready in minutes.