Skip to main content

Overview

Agent mode is Sphinx’s default and most autonomous interaction mode. In this mode, Sphinx works step by step—explaining its thoughts and actions—and continues iterating until it determines the task is complete.
Agent mode provides intelligent safeguards through guardrails that flag potentially risky operations for your approval, while allowing routine code to execute automatically.

How It Works

When you submit a prompt in Agent mode, Sphinx:
  1. Analyzes your request and the current notebook state
  2. Generates code to accomplish the task
  3. Executes cells automatically (unless flagged by guardrails)
  4. Inspects results and iterates as needed
  5. Completes when the task objective is achieved

Capabilities

Sphinx can perform the following actions autonomously in Agent mode:
  • Add cells — Insert new code or markdown cells
  • Edit cells — Modify existing cell content
  • Execute cells — Run code and analyze outputs
  • Delete cells — Remove unnecessary cells
  • Ask questions — Request clarification when stuck

When to Use Agent Mode

Agent mode is ideal for:
  • Routine data science tasks — Loading data, cleaning, basic analysis
  • Exploratory work — Quick investigations where speed matters
  • Trusted workflows — Tasks where you’re confident in the approach
  • Iterative development — Rapid prototyping and experimentation
Agent mode excels when you have a clear goal and trust Sphinx to find the path. For complex strategic decisions, consider Plan Mode instead.

Guardrails and Safety

Even in Agent mode, Sphinx doesn’t have unchecked autonomy. The guardrail system automatically:
  • Detects risky operations like file writes or system modifications
  • Pauses for approval when potentially dangerous code is generated
  • Respects your custom rules configured in Sphinx settings
This means you get the speed of autonomous execution with safety nets for operations that could have unintended consequences.