Using the TUI
The TUI (terminal interface) brings the full agent into your terminal, running locally with access to your files. This page covers day-to-day use. To install, see Install TUI; for the dry reference of every command, see TUI commands.
Start
Open the TUI with:
imaginne
imaginne chat is equivalent. On the first run, with no setup wizard, the login opens in your browser automatically — complete it and return to the terminal. The TUI requires a terminal with a TTY; if you see requires a TTY-capable terminal, use a real terminal emulator.
The DEV channel uses the imaginne-dev binary and the ~/.imaginne-dev folder (it does not share config with PROD). Use the build your organization points you to.
The layout and the status bar
The screen has a header (TUI — [N]umbers), the conversation viewport, the > prompt with the text box, and, at the bottom, the status bar:
EXECUTE · nnumbers · 1.2k tokens · ctx:▓▓▓░░ 38%
The bar shows the current mode (EXECUTE / PLAN / ASK), the model, token usage, and the context bar (used/limit). See Sessions to understand context and compaction.
Send and write multiple lines
- Send — Enter.
- Multi-line — paste text with multiple lines to send it all at once.
Working modes: Exec, Plan, Ask
The agent operates in one of three modes, switched with Shift+Tab (or with the commands below):
| Mode | Behavior |
|---|---|
| EXECUTE | Plans and acts (edits files, runs commands and skills). |
| PLAN | Plans only; does not run actions with effects. |
| ASK | Talks/answers, without changing anything. |
Equivalent commands: /exe (/execute), /plan, /ask.
Slash commands
Type / to open the menu. Full list:
| Command (aliases) | What it does |
|---|---|
/help (/h, /?) | Lists the commands. |
/quit (/q, /exit) | Exits the TUI. |
/clear (/cls) | Clears the conversation. |
/context clear | Clears the context (restarts the model's reasoning). |
/model (/m) | Opens the model picker. |
/autonomy (/a) | Adjusts the autonomy level. |
/verbose (/v) | Adjusts the output verbosity level. |
/version | Shows the version. |
/update | Updates the binary. |
/diagnostics | Collects diagnostic information. |
/status | Shows the session state. |
/project (/p) | Project operations. |
/skill (/skills) | Manages skills (see subcommands). |
/sessions | Manages sessions (see subcommands). |
/plan · /exe (/execute) · /ask | Switches the working mode. |
/lang (/language, /idioma) | Changes the language. |
/remote | Remote session pairing (see below). |
/init | Initializes the project. |
/iterate (/iter) | Iterates on the last output. |
/agent | Subagents (see subcommands). |
/login · /logout · /whoami | Identity. |
/policy | Shows the effective policy. |
/memory | Opens memory. |
/feedback (/fb) | Sends feedback. |
/todo | Session task list. |
/skill subcommands
| Subcommand | What it does |
|---|---|
list | Lists the available skills. |
info <name> | Details of a skill. |
validate [path|key] | Validates a skill's manifest. |
groups [--search q] | Lists skill groups. |
publish <key|--skill-path p> --group <slug|id> [--mode] | Publishes a skill. |
status | Skill sync state. |
/sessions subcommands
| Subcommand | What it does |
|---|---|
list | Lists the sessions. |
new <title> | Creates a session. |
switch <id> | Switches sessions. |
archive <id> | Archives a session. |
rename <id> <title> | Renames a session. |
You can't switch sessions while the agent is responding, and you can't archive the active session. See Sessions.
/agent subcommands
The TUI can delegate tasks to specialized subagents:
| Subcommand | What it does |
|---|---|
list | Lists the subagents. |
spawn <type> <task> | Spawns a subagent for a task. |
status | Subagent state. |
cancel <id> | Cancels a subagent. |
metrics · cost | Metrics and cost. |
Available types: explorer, analyzer, reviewer, implementer, tester, refactorer, documenter.
Mentioning files: @
Point to a file with @ and the TUI reads its contents and injects them inline into your message:
summarize @report.md and produce an executive version
Accepted syntaxes:
@path— relative path;@"path with spaces"— quoted;@/abs— absolute path;@~/home— from your user directory.
Directories, binaries, and files larger than 100 KB are ignored. Use Tab to complete (path mode and fuzzy mode) and ↑/↓ to navigate the suggestions. See Files & artifacts.
Model
/model (/m) opens the picker, filtered by the organization's policy. The default is nnumbers. The TUI always auto-routes per turn — selecting a model works as a hint/override. See Models.
Autonomy
Adjust it with /autonomy (/a) or Ctrl+A:
- low — asks about everything;
- medium — balanced;
- high — full auto.
The preference is saved in ~/.imaginne/prefs.json. The workspace vault stays active at any level. See Autonomy & permissions.
Verbosity level
/verbose (/v) or Ctrl+X cycles between silent, medium, and debug (default debug). Use lower levels for cleaner output.
Voice
Voice is speech→text: Ctrl+G or F2 starts capture (depends on a build with the voice tag; today, the darwin-arm64 release). The transcription is appended to the prompt and is not sent on its own; Esc cancels. Configurable via IMAGINNE_VOICE_* variables. See Voice.
Remote session
/remote shows a pairing code valid for 30s. On the other end, open <host>/remote in the browser, sign in, and paste the code — the prompts run on your computer and the stream comes back to both screens. Use /remote status and /remote stop to manage it. Requires login. See Remote session.
Keyboard shortcuts
| Shortcut | Action |
|---|---|
| Enter | Send |
| Esc | Cancel (double Esc clears) |
| Ctrl+C | Quit |
| Ctrl+G / F2 | Voice |
| Ctrl+A | Autonomy |
| Ctrl+T | Task panel |
| Ctrl+X | Verbose |
| Ctrl+K | Clear screen |
| Ctrl+O | Clear context |
| Ctrl+Y / F1 | Help |
| Shift+Tab | Cycle mode (Exec/Plan/Ask) |
| Tab | Complete |
| PgUp / PgDn | Scroll the conversation |
Configuration
The TUI reads ~/.imaginne/config.yaml (YAML, 0600 permissions). The main keys:
llm:
model: nnumbers # default model
max_tokens: 16384
temperature: 0.7
memory:
storage_path: ~/.imaginne/memory
You normally don't edit this file by hand — login fills in the identity. Use imaginne config init to generate a base file.
Environment variables
| Variable | Effect |
|---|---|
IMAGINNE_DANGEROUS_BYPASS | Enables dangerous execution mode (equivalent to starting with --dangerous-bypass). |
IMAGINNE_DEBUG | Turns on debug logs. |
IMAGINNE_LEGACY | Legacy compatibility mode. |
IMAGINNE_TUI_LOCAL_STORE=off | Disables local session storage. |
IMAGINNE_GLOBAL_SKILLS_DIR | Overrides the global skills folder. |
IMAGINNE_VOICE_* | Voice configuration. |
The full list of client variables is in Environment variables.
See also
The full reference of commands, subcommands, and shortcuts.
Open →Autonomy & permissionsThe autonomy levels and the workspace vault.
Open →Remote sessionDrive your local session from another screen.
Open →VoiceDictate requests by speech→text.
Open →SkillsWhat the agent can do and how to publish new skills.
Open →Environment variablesThe client variables that affect the TUI.
Open →Was this page helpful?
Report a problem on this pageDo not send passwords, keys, tokens, or customer data.