Modes
Codebuff has four modes. Switch during a session with Shift+Tab or /mode: commands.
| Mode | Model | Editor Agent | Code Review |
|---|---|---|---|
| Default | Claude Opus 4.5 | editor | Yes |
| Lite | Grok 4.1 Fast | editor-gpt-5 | No |
| Max | Claude Opus 4.5 | editor-multi-prompt | Yes |
| Plan | Claude Opus 4.5 | None | No |
Default
Standard mode with Claude Opus 4.5:
- Spawns file pickers and code searchers to gather context
- Uses the
editoragent for code changes - Spawns
thinkerfor complex problems - Runs code review after changes
- Validates with typechecks and tests
bash
codebuff # no flag needed
Lite
Grok 4.1 Fast, cheaper and faster:
- Less file context gathering
- Uses GPT-5.1 for code changes (
editor-gpt-5agent) - Skips code review
- No todo tracking
Good for quick fixes and simple questions.
bash
codebuff --lite
Max
Claude Opus 4.5 with best-of-N selection:
- Reads 12-20+ files per task
- Spawns multiple editor agents in parallel, each with a different strategy
- A selector compares outputs and picks the best one
- Spawns
thinker-best-of-n-opusfor hard problems - Full code review after changes
- Runs typechecks and tests on both the changed files and the full project
Use for complex features and refactors.
bash
codebuff --max
Plan
Planning mode. No file writes.
- Gathers context about your codebase
- Asks clarifying questions using
ask_user - Outputs a spec/plan wrapped in
<PLAN>tags - Includes requirements, constraints, and relevant files
Use this to scope out work before implementing, or to discuss approaches without making changes.
bash
codebuff --plan