xhillman/axis-core ? reverse-engineered prompt

Build me a Python AI agent framework that I can use in real projects, not just demos.

I want one clean package with an agent that can think through a task step by step, use tools, keep track of memory, and let me swap out different model providers without rewriting everything. It should have a simple command line too, so I can ask a one off question or start a chat session and come back to it later. Make sure it can handle streaming output, basic budgets like cost or token limits, retries and timeouts, and safe handling for tool calls so it does not break when a run is resumed.

Please include a small setup flow for configuring defaults, plus examples showing how to make a custom tool and create an agent. If you need to check current provider docs or best practices, look them up online. Keep the code polished, type checked, and ready to ship as a proper open source library.