speedAgent

Minimal pydantic_ai chat agent with a Python REPL tool.

Security warning (USE AT YOUR OWN RISK)

This project intentionally exposes powerful tools to the model, including:

• an unrestricted Python REPL (arbitrary Python execution)
• an unrestricted bash command runner (arbitrary shell command execution)

That means prompts (and model/tool output) can:

• read/modify/delete files accessible to your user account
• execute arbitrary programs and install packages
• access environment variables and other secrets available on the machine
• make network requests (directly or via installed tools)

Only run this on a machine/environment you trust and are willing to risk. Do not deploy it to the public internet. Prefer a sandboxed, disposable environment (e.g., a container/VM) with minimal permissions and no sensitive credentials.

Setup

Sync dependencies:

uv sync

Run

Set your model API key (OpenAI backend by default):

export OPENAI_API_KEY=... 

Or create a .env file in the project root:

OPENAI_API_KEY=...

Start the CLI:

uv run uvicorn agent:app --reload

If OPENAI_API_KEY is not set, the app will show the UI, but not interact with the model.

Example prompt

Ask it to run a small PySCF calculation via the python_repl tool, e.g.:

"Compute RHF energy for H2 at 0.74 Å in STO-3G and report the total energy."