Run

One-shot autonomous task (mazu run)

Task (leave empty to use Resume below)

Max steps

Checkpoint every N steps

Max cost (USD, optional)

Model (optional)

Keep last N checkpoints (optional)

Shell allowlist (optional, comma-separated)

Allow shell (skip confirmation)

Dry run (plan only, no writes/checkpoints)

Resume an earlier run

Run id

Fork from a checkpoint (new branch)

Checkpoint id

New branch name

Explore

Parallel branches, ranked comparison (mazu explore)

Task

Models (comma-separated -- ignored if Auto-models is checked)

Auto-models (pick from this project's router history)

Test command (optional)

Max cost (USD, shared)

Max steps / branch

Fork from checkpoint (optional)

Council

Ask multiple models independently, one lead synthesizes (mazu council)

Question

Members (comma-separated)

Lead (synthesizes)

Max cost (USD, optional)

Checkpoints

Timeline & rollback

Keep last N

Memory

Project-scoped agent memory

Why would QUERY retrieve? (optional)

Skills

Saved reusable solutions

Runs

History of mazu run / explore invocations

Compare: run A id

run B id

Router

Model win-rate from explore history

Usage

Estimated API spend

Action Log

Every tool call, every session

Select a session

Models

Streaming, tool use, context window, pricing (best-effort)

Doctor

Setup diagnostics

Also verify keys with a real API call (tiny cost)

Config

~/.mazu/config.toml

Quick setup

Provider

API key

Verify with a real API call

Set as default model

Set a value

Key

Value