criogen (Python)

pip install criogen

Wheels for Linux (x64, ARM64), macOS (Apple Silicon, Intel) and Windows (x64); Python 3.9 or newer.

The fast AI agent framework for Python, a LangChain and LangGraph alternative: agents and workflows with tool calling, human approval, durable threads, long-term memory, an audit trail and encryption at rest, on Criogen's Rust engine. In our benchmark an agent turn costs 0.29 ms of framework overhead, against 1.23 ms for LangGraph and 34 ms for CrewAI.

import asyncio
from criogen import Agent, tool, ToolError

@tool
async def get_order(order_id: str) -> dict:
    """Look up an order by id."""
    if order_id != "A-1":
        raise ToolError(f"No order {order_id}")
    return {"id": order_id, "status": "shipped"}

agent = Agent("openai:gpt-5-mini", tools=[get_order], storage="sqlite:///support.db")

async def main():
    chat = agent.thread("user-42")
    reply = await chat.send("Where is order A-1?")
    print(reply.text)

asyncio.run(main())

Keep your keys in a .env file

from criogen import load_env

load_env()   # reads ./.env: OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY, OPENROUTER_API_KEY

It returns the names it set (never the values) and never replaces a variable that is already set. A model with no key fails at once and says which variable to set.

Status: first slice: agents, tools, approvals, typed output, threads (send, approve, reject, decide, retry, history, rewind, delete, stream), storage (SQLite, Postgres, encryption, audit, retention) and memory. Workflows too (Workflow, design 0015), and from design 0017: agents as tools, retries and fallback models, memory by meaning, tracing and hooks, Gateway for model calls without an agent (design 0018), and ready-made patterns, extendable specialist agents and web search (criogen.patterns, specialists, web; design 0019). Tests use FakeModel; no real model is called.

Free to use, also commercially (see LICENSE).