ncode CLI Using ncode
The session screen
ncode opens one conversation of one project in full screen. The screen is built around the composer: letters always type into your draft, and nothing you type is taken as a command unless it starts with /.

Layout#
| Area | What it shows |
|---|---|
| Run tabs | the runs of this conversation; Alt+1 … Alt+4 switch to a tab as drawn |
| Transcript | your messages, the assistant's answers, and one row per tool call (a file read, an edit, a command) |
| Side panel | one row per agent at work, most urgent first; by default (auto) it stays out of the way until two or more agents work or something needs you; Ctrl+B cycles auto, full, compact and hidden (see Agents) |
| Composer | your draft; Enter sends it (see Composer) |
| Status line | the project's approval mode (read-only, auto or full access), always, the model with the effort in force (the conversation's own, else NCODE_EFFORT, else the default in Settings), the worker model (worker <model>) while it differs, how much context is used against the model's window (ctx 8k/1M), the cost, and hints for the keys that work right now; when the side panel is not showing them, the busiest model's tokens per second and the memory (142 tok/s · RAM 312 MB) |
Under 120 columns the side panel becomes a one-line strip, or disappears. A small 80 × 24 terminal still works. While a run retries a failed provider call the turn header says retrying 2/5 · HTTP 500 instead of thinking, and it says writing as soon as the answer starts to stream.
If you open the desktop app while a session is open, a one-line banner says so: The ncode app is open on the same database. Quit it, or stop your runs here and quit (Ctrl-C twice).

Reading long output#
A transcript row shows at most 8 KB of a prompt or an answer and 2 KB of a tool's output; the row says how much more there is. To read all of it, move to the row in select mode and press Enter (or o): it opens whole in a pager, even when nothing was cut. A tool row with nothing more to open says nothing more to open. A command that ran in the background shows how it ended: exit 0, killed at quit, still running or ended (exit not recorded).
Select mode#
Ctrl+T switches from typing to selecting transcript rows:
| Key | Does |
|---|---|
j / k | move down / up |
| Enter | open the row (the whole output, or the agent it belongs to) |
y | copy the row |
r | on a failed or stopped run, retry it (r retries · Ctrl-P Retry failed run) |
| Esc or Ctrl+T | back to typing |
Scrolling and the mouse#
- PgUp / PgDn scroll the transcript; on an empty draft Ctrl+U / Ctrl+D scroll half a page.
- The mouse wheel scrolls the transcript, an agent overlay or the pager, three lines a notch. By default the terminal keeps its own mouse selection, so you select text by dragging;
/mouse onsends wheel reports to ncode instead, and then Shift-drag (in Terminal.app and iTerm2, Option-drag) selects. See Terminal, themes and mouse. - Ctrl+L redraws the screen if it is garbled.
Switching conversations and finding things#
- Ctrl+P opens the palette: conversations, runs, the model, and the feature screens: Workflows, Deep Research, Scheduled Tasks, Settings, Usage, Changes, Checkpoints and MCP Servers.
/resumeopens a picker of saved conversations: each row shows its age (2h ago) and, under the title, the last prompt./newstarts a new one and keeps this one saved.ncode --resumewith no value opens the same picker at start./search <words>lists the messages of this project's conversations that match, as a picker; Enter opens the conversation./rename <title>,/fork(copy this conversation into a new one and open it) and/delete(asks first; files are not touched) manage conversations.- Ctrl+G opens the runs dashboard, Ctrl+R the run palette (in the composer Ctrl+R searches your earlier prompts instead; see Composer).
- F1 (or
?outside a text field) shows every key that works where you are. F2 opens Settings.
Quitting#
Press Ctrl+C twice within 1.5 seconds with nothing else to cancel. A single Ctrl+C first closes an open layer, then clears your draft (Ctrl+Z brings it back), then stops the running turn; a press that did one of those never counts towards quitting. When runs are still working, ncode asks:
Stop 2 live runs and quit?
Quitting stops this session's runs. Closing the terminal window (SIGHUP) or sending SIGTERM does the same cleanly: the runs stop, the session closes and the summary below is printed.
Back in your shell, the last turns of the conversation are printed into the terminal's normal scrollback, so you can scroll up to them and copy from them. Then comes a short summary: the runs it stopped, what you spent, and the line to reopen the conversation:
Last prompt add a --verbose flag to the CLI parser Spent 12.4k tokens · $0.03 · 6m 12s
ncode --resume <conversation-id>- The last turns. Up to the last 3 exchanges (
terminal.exit_transcript: 0 to 20,0prints none): your prompt after›, the assistant's reply as it was written, and a$line for a command you ran with!. Turns you rewound are left out. A long reply is cut at 4,000 bytes and the whole print at 24 KiB. - The Files changed line counts the net change of the session, with the first three names (
2 · NOTES.md, src/app.ex, then+Nfor the rest). A file that is back to what it was before the session, because you rewound it or reverted it by hand, is not listed; a file the session created and then removed is not listed either. The line is left out when nothing changed. - The Spent line adds tokens in and out (
12.4k,1.3M), the dollar cost, and how long the session ran (45s,6m 12s,1h 05m). The dollar part appears only when every run of the session has a known cost; below one cent it reads<$0.01. The line is left out when no run started. - Only the full screen prints these.
-pand--plainprint neither.
A Ctrl+C outside the full screen (while it starts, during a -p run, or after the summary) simply ends the program.

This screenshot was taken with 0.1.0, whose summary has no turns above it and no Spent line.