Work in the TUI
Use the conversation view to follow the response, inspect tool calls, and steer the agent without waiting for the current run to finish. Assistant text streams as it arrives. Tool calls appear as compact cards; Edit and Write cards include their diff.
Focus a tool card and press ctrl+t to view its complete arguments and output.
Press ctrl+t again to return to the preview.
Attach a local file
Type @ to complete and attach a file. Ordinary @path and @./path completion
search the client workspace, preserving a leading ./ on insertion. One or more
leading ../ components search the corresponding parent of that workspace and remain
in the inserted path, so they may select files outside the workspace. A token starting
with literal ~/ instead searches the home directory of the mecatui client process,
retains ~/ when inserted, supports leading ./ and ../ components after ~/, and
also works when the client workspace is empty. With no client workspace, non-home
completion has no implicit process-cwd fallback. Completion lists files only, prunes
hidden files and directories, and is bounded; use ↑/↓ then tab or enter to
select a result.
On send, mecatui reads a mentioned regular file and uploads its bytes into the
conversation: text files are inlined and supported media becomes an attachment. The
source path is not a server workspace path and is never mounted, materialized, or made
readable/editable through server tools. Therefore in a remote, containerized, or no-FS
session, ~ still means the machine running mecatui, not the server/container.
Traversal outside the client workspace still uploads content only and grants no
execution-environment filesystem access. Treat this as sharing local content and avoid
attaching sensitive home-directory files.
Only literal leading ~/ has this home-expansion meaning. ~user receives no
home expansion, but can still attach as an ordinary workspace-relative mention when
such a file exists. Quoted paths and embedded tildes remain ordinary prose; a path
containing whitespace cannot be one mention token. If mecatui cannot determine its
home directory, @~/… also remains prose.
Keep working while a run is active
Type your next instruction while the agent is running, then press enter.
mecatui steers the run at the next safe turn boundary when the server supports
steering. Otherwise, it queues the instruction as a follow-up. Images and other
supported staged media stay attached, including media-only input. Multiple
queued lines become one prompt.
Bare TUI commands stay in the client. For example, /help opens local help and
/clear can replace the session during a run or approval. Unknown slash
commands, workspace commands, and built-in commands with arguments go to the
model.
To revise queued input, empty the prompt and press ↑. This restores the
pending steer or queued follow-up with its staged media. Press ctrl+u to clear
the unsent draft and its attachments.
Pressing esc clears an active selection first. During a run, it cancels the
run but preserves your draft and queued input. When the session is idle with a
paused queue, it clears the queue and preserves the draft.
Add files and images
Type @ to find a file in the workspace, then select it with enter or tab.
mecatui inserts text files into the prompt and attaches supported image or
audio files as media. The selected model must support the media type; otherwise,
the client keeps the draft and reports the unsupported attachment.
Press ctrl+v to paste an image from the clipboard. If the clipboard does not
contain an image, ctrl+v pastes its text. Large text pastes appear as compact
placeholders in the editor and expand when you send the prompt.
See Multimodal input for model capability and validation behavior.
When a model stream fails
When the server reports a retryable + precommit failure, mecatui retries the
model step once without adding a prompt or removing queued messages. If the
automatic retry fails, use /retry. The same command retries a
retryable + visible failure, including one reopened from storage.
/retry preserves the prompt textarea and queued prompts. For visible failures,
scrollback marks the failed partial output as superseded. If no eligible failure
is pending, the command reports that fact and makes no changes.
Review approvals
When a tool needs permission, a modal shows what it wants to do. Read the
request, then allow it once, allow the exact action for this session when
offered, or deny it. Long arguments can be scrolled; ctrl+t opens a
full-screen view when needed. Mouse buttons activate the same choices as their
displayed keys.
Complete browser authorization
When workspace-service enrollment or an MCP tool opens a browser, complete the
authorization and return to the TUI. mecatui checks the pending request
automatically.
Each workspace-service request has a 30-second deadline. After a timeout, the
server's state is uncertain, so mecatui stops checking. Run /clear, then run
/tools-connect in the replacement session as the displayed message directs.
You can still cancel a pending MCP authorization from its card. Presentation links are opened or copied only for that interaction and are not retained in the conversation.
Browse available capabilities
Open the slash-command palette with /. mecatui shows only the panels and
commands supported by the connected server.
| Task | Open in mecatui | More information |
|---|---|---|
| Browse MCP servers, resources, and prompts | /mcp; press f8 to open MCP prompts directly | MCP client |
| Inspect named agent definitions | /agents | Named agents |
| Inspect available skills and the active soul | /skills and /soul | Skills, commands, and soul |
| Inspect the user model | /usermodel | Memory |
| Manage recurring and one-shot tasks | /schedule | Scheduled tasks |
| Review learning and maintain memory | /learning, /reflections, /reflect, and /dream | Use learning and memory commands |
The palette also includes workspace-defined slash commands. See Skills, commands, and soul for how the server discovers and expands them.
Monitor delegated work
Press f6 to open the agents overlay for Subagents, Parallel runs, and Teams.
The footer shows running and completed counts after delegated work begins. Use
the overlay to inspect bounded activity previews; /team opens the same overlay
on the Teams tab.
See
Subagents, teams, and parallel
for delegation behavior and the information available in mecatui.
Change conversation settings
Press shift+tab to switch the active permission mode. See
Choose a permission mode
for the available modes and their behavior.
| Command | Result |
|---|---|
/models | Starts a session on the selected model and keeps the conversation. Switching models clears model caches and can increase cost. |
/effort | Forks the conversation onto the selected reasoning-effort tier. The server reports unsupported tiers. |
/compact | Reduces model history while keeping the session and visible scrollback. Run it without arguments while idle. Creating a cascade summary can use model tokens. |
/clear | Creates an empty-history session with the same placement. It does not roll back workspace changes. |
/session | Shows path-free details for the active session. |
/posture | Shows the server's operator posture and active defenses. See Permissions and posture. |
If /clear cancels an active run or approval and then fails to create the
replacement, the original session remains selected and may be cancelled. Wait
for it to settle, then retry /clear.
A short key reference
Use ? on an empty prompt for the live help overlay. The everyday defaults are
enter to send or steer, shift+enter or ctrl+j to insert a newline, ctrl+t
to inspect details, pgup/pgdn to scroll, and / to open commands. If the server does
not support steering, enter queues a follow-up while a run is active. See
Keybindings for approval controls, remapping, and the
complete reference.
Next steps
- Manage sessions to resume, inspect, fork, or clear a chat.
- Use learning and memory commands when the server provides learning features.