kinjal1234/sandbox-runtime repository overview

⁠A.D.A.M. (Agentic Data Analysis Manager)

⁠Code Execution Sandbox Service Specifications

Overview: The sandbox image provides a strictly guarded environment for executing python data cleaning and statistical code generated by the Gemini agent, preventing raw model outputs from interacting with the host machine.

Key Technical Aspects:

• Secure Isolation: Limits system privileges and restricts network access within the container to prevent malicious or unintended shell commands from compromising the infrastructure.
• Pre-baked Scientific Stack: Includes specialized data processing libraries ($Pandas$, $NumPy$, $SciPy$, $Statsmodels$) to instantly execute complex statistical tests without download overhead.
• State Return Pipeline: Safely catches script execution outputs, syntax errors, and structured dataframes, piping the execution logs cleanly back to the FastAPI backend.

Main Project Link:

View on GitHub⁠