DeepSeekTUI.wiki

Plan, Agent, and YOLO modes

Modes define how much autonomy DeepSeek TUI has while working inside your repository. Choosing the right mode is the fastest way to avoid destructive surprises.

At a glance

Mode Reads files Writes files Shell commands Typical use
Plan Yes No No Audits, onboarding, design discussions
Agent Yes Yes (with approvals) Yes (with approvals) Daily coding with guardrails
YOLO Yes Yes (auto) Yes (auto) Trusted sandboxes only

Plan mode

Plan mode keeps the assistant in an investigative posture: it can traverse files, search, and explain architecture, but it should not mutate your workspace or launch shells. Use it when you are learning a repo, reviewing risk, or pairing with teammates who want explanations—not edits. 

deepseek --mode plan

Agent mode

Agent mode is the default sweet spot for implementing changes. DeepSeek TUI can propose patches and commands, while you confirm anything sensitive. Pair Agent mode with sane approval settings in Configuration.

YOLO mode

Risk: YOLO removes friction by auto-approving actions. Reserve it for disposable clones, generated scratch branches, or tightly scoped chores where rollback is trivial.

Switching during a session

/mode plan
/mode agent
/mode yolo

A practical loop is Plan → Agent → Plan: investigate, implement with approvals, then re-scan.

FAQ

Is Plan mode “safe”?

It is the safest built-in mode because it avoids writes and shell—but you should still treat repo contents as confidential.

Does Agent mode slow me down?

Approvals add clicks, yet they prevent accidental mass edits—usually a worthwhile trade.

What should I read next?

Continue with Tools or RLM.