Plan first.
Use /plan when the task touches files, carries risk, or needs user review.
Control the real agent runtime without leaving the TUI or Mac canvas.
Control the real agent runtime without leaving the TUI or Mac canvas.
/plan when the task touches files, carries risk, or needs user review./goal when the work has clear completion criteria and may take multiple loops./memory to inspect what the agent will retrieve, and /replay to verify what the agent actually used./mode safe/mode auto/mode paranoid| Trigger | Description | Example |
|---|---|---|
| /plan | Enter read-only architecture mode and draft a reviewable plan. | /plan add tests for auth. |
| /goal | Start or control a long-running objective with criteria and evidence. | /goal migrate js to ts. |
| /memory | Open, add, remember, pin, unpin, or delete runtime memory. | /memory pin use vitest. |
| /auth | Open authentication and provider state. | /auth |
| /mcp | Manage MCP servers. | /mcp install github. |
| /subagent | List, launch, fork, or inspect scoped subagents. | /subagent reviewer. |
| /replay | Open session, goal, provider, tool, approval, and memory evidence. | /replay latest. |
| /mode | Switch permission mode for the session. | /mode paranoid. |
| /provider | Switch model provider in one step. | /provider openai. |
| /model | Switch the routed model in one step across 1M-context frontier and local models. | /model qwen3-coder. |
| /worktree | Create, apply, or remove isolated git worktrees for parallel threads. | /worktree create feat-auth. |
| /diff | Show the workspace diff or compare against a base branch. | /diff main. |
| /review-branch | Review the current branch against main and surface regressions. | /review-branch main. |
| /verify | Run configured checks or plan verification. | /verify. |
| /undo | Back out the last accepted write batch when supported. | /undo. |
| /commit | Prepare a commit message after verification. | /commit. |
Use /plan when the task touches files, carries risk, or needs user review.
Use /goal when the work has clear completion criteria and may take multiple loops.
Use /memory to inspect what the agent will retrieve, and /replay to verify what the agent actually used.
Use /worktree to run parallel threads in isolated git worktrees and /subagent when investigation, review, tests, or docs can happen in parallel.
| Command | Use |
|---|---|
| /worktree create|apply|remove. | Isolate a thread in its own git worktree, apply changes back, or clean up. |
| /diff main. | Show the current workspace or branch diff against main. |
| /review-branch. | Run a structured local review of the current branch before opening a PR. |
| /verify. | Run configured checks or the checks named in the plan. |
| /undo. | Back out the last accepted write batch when supported. |
| /commit. | Prepare a commit message after verification. |
Mode commands change how much Picasso can do without asking. /model and /provider switch routes in one step across 1M-context frontier and local models. Sponsor commands explain or block labeled sponsor surfaces. Memory, replay, worktree, and subagent commands expose runtime state instead of hiding it behind chat.
/mode safe/mode auto/mode paranoid/model claude-opus-4-7/provider anthropic/worktree create feat-auth/diff main/review-branch/memory/replay latest/subagent/sponsors why/sponsors block <sponsor>The Mac app does not run a separate product. Its multi-thread sidebar with unread completion dots and native notifications, instant runs without project setup friction, Worktree composer pill, inline syntax-highlighted diffs, and Reasoning | Changes inspector share the same local runtime state as the CLI and TUI.
Run parallel threads with unread completion dots, native notifications, and per-thread Mac Worktree mode.
Inspect model reasoning, inline syntax-highlighted diffs, and edit-loop auto-diagnostics side by side.
Memory writes, goal ledgers, and write approvals keep the exact same policy semantics across CLI and Mac.
Non-interactive command equivalents.
Project-level policy around command execution.
Permission modes, sandboxing, and audit logs.
Picasso for Mac is almost here — a coding agent that looks the way serious tools should, and costs what creative freedom should: nothing. Leave your email and be first on the canvas.
Sponsors and labs — the early canvas is yours. Choose Sponsor or Lab above and we'll reach out before launch.