Interactive vs goals

FlareCode gives you two modes on the same agent. They share the same sandbox, repo access, and safety rules — they differ in how much you steer.

Goals

You describe an outcome and walk away. The agent plans, works, checks itself, and opens a Pull Request. You review later.

Reach for goals when:

• the task has a clear done condition ("add a test", "fix this bug", "bump the dependency"),
• you want to hand off and keep moving,
• you'll review the result as a PR on your own schedule.

This is the async-first path, and it's what most people use day to day.

Interactive

You drive the agent directly in a hosted workspace — a coding terminal you steer in real time. You see files, run commands, and course-correct as it goes.

Reach for interactive when:

• you're exploring an unfamiliar codebase,
• the task needs tight back-and-forth ("try that — no, use the other helper"),
• you want to watch and learn how the agent approaches the repo.

When you're happy, the same branch-and-PR flow applies: the agent commits to flarecode/task-<id> and opens a PR.

They mix

Start a goal, then jump into interactive to nudge it if it gets stuck — or explore interactively, then hand the rest off as a goal. Same agent, same sandbox, same review gate at the end.