Using the CLI

The core-agent binary supports two interaction modes. Pick the one that matches what you’re trying to do:

Interactive

You’re driving the agent yourself from a terminal. The Bubble Tea TUI takes over, conversation history is preserved across turns, slash commands surface session controls. Best for code review, exploration, ad-hoc tasks, anything where you want to see the model’s response and react.

Start with the interactive quickstart for the first-15-minutes path.

Autonomous

The agent runs unattended against a goal you defined up front — typically a long-lived monitor, scheduled job, or remediation worker. Budgets cap turns/tokens/cost/wallclock; the session log is the audit trail. Best for headless deployments where no operator is at the keyboard.

Start with the autonomous quickstart for the first-15-minutes path.

Library users

If you’re embedding core-agent inside your own Go binary — building a coding assistant, web service, or custom autonomous worker — see Using the library instead. The CLI guides above describe how the bundled binary behaves; the library guide covers the Go API you’d use to construct your own.