Work in the TUI
Press ctrl+t to open /toolcalls and inspect top-level calls during a run.
Use ↑/↓ to choose a row, then enter or click it to open its detail;
press esc to return to the list. The inspector also works when revisiting a
session transcript. Running Edit and Write cards show their request diffs inline;
settled calls keep their full arguments and results in the inspector. When
mecatui receives a result before its final update, it keeps the card open
and shows result received · finalizing.
A Subagent call's detail shows one-line ✓/✗/pending summaries of up to
128 retained child tool calls, without child drilldown. Each summary shows
call-side intent (such as Read · path), not the result body. An early
observed result shows result received · finalizing (or failed · finalizing)
until the canonical result confirms or changes its outcome. A call without a
result stays pending. The server scrubs controls and caps preview length;
previews are neither complete child calls nor guaranteed secret redaction.
Results without a matching retained call are omitted rather than shown as
new rows. A reconstructed session may have no Subagent activity preview and
notes that history may be incomplete. Press f6 for the Agents view to
navigate Subagent children, Parallel branches, and Team members, including
their task and findings views. Each lane retains up to 128 tool calls and
12 independent message previews. Child previews with invalid UTF-8 or IDs
longer than 256 bytes are omitted; child execution and the parent session
are unaffected.
Press f9 to reveal conversation details, including reasoning summaries,
per-turn usage, permanent error details, and changed files. Tool results stay in
/toolcalls. Both shortcuts can be remapped.
f9 also temporarily reveals retained benign guardrail notices in live and
replayed conversations. These are completed, acceptable checks that allowed an
action or released a result; findings, failures, unresolved reviews, and
approvals remain visible. To keep benign notices visible, configure
hook_notices.show_benign.
Start a session with an execution choice
Enter /execution while idle. Use ↑ and ↓ to choose Deployment default,
None, or an eligible template, then press enter to create a new session.
The existing session stays intact if creation fails. None binds no execution
filesystem or built-in Shell; the default uses the deployment's configured
execution. Template rows show operator-provided display names and exact pinned
revisions, not paths or container settings. The server checks the selected revision
and your access again at creation time; a stale or full template never falls back
to the default. The success status and /session details report whether the new
session actually has execution files and a built-in Shell.
When catalog discovery is disabled or temporarily unavailable, the picker still
allows default and none. An empty catalog has no templates eligible for your
account. If a selected template becomes unavailable, reopen /execution to refresh
and retry. Debug chats stay file-less and do not offer this switch.
Attach a local file
Type @ to find a file on the machine running mecatui. Use ↑ or ↓ to
select a result, then tab or enter to insert its path. While completion is
open, enter selects a path and esc dismisses the list.
| Mention prefix | Search location |
|---|---|
@path or @./path | The client workspace. |
@../path | A parent of the client workspace; repeated ../ components reach higher parents. |
@~/path | The client process's home directory, including when no client workspace is configured. |
Completion preserves the prefix, lists files only, excludes hidden files and
directories, and limits the number of results. Without a client workspace,
non-home completion has no current-directory fallback. After ~/, leading ./
and ../ components are also supported.
When you send the prompt, mecatui reads mentioned regular files and uploads
their content. Text is inlined; supported media becomes an attachment. The file
stays on the client machine. Its path does not become a server workspace path,
and uploading it grants no filesystem or command access. This also applies to
remote, containerized, and no-filesystem sessions. Check what you are sharing,
especially when selecting home-directory files.
Only a literal leading ~/ expands to the client home directory. ~user is an
ordinary workspace-relative path when that file exists. Quoted paths, embedded
tildes, and paths containing whitespace remain prose rather than a single
mention. If the client 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.
While a run is active and the prompt is available, /clear replaces the
session. 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
File mentions can attach 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
The server recovers transient failures before model output becomes visible,
within its configured recovery limits.
If recovery ends in a terminal failure, use /retry; mecatui does not start
another run automatically. 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
During a permission request, the active Toolcalls binding (ctrl+t by
default) opens that request's details instead of /toolcalls. See
Approve or deny a request for
approval keys and the scope of "allow always".
Get editor notifications
Run mecatui in a terminal provided by a supported editor and the editor can
tell you when the agent needs an approval and when a run ends. You configure
nothing in mecatui for this.
mecatui reports four lifecycle points to the editor's agent hook: the start of
a turn, an approval request from the main session, the operator's answer to that
request, and the run's terminal state. The answer returns the editor's status to
working immediately instead of leaving the approval notification active until
the run ends. The editor decides how to present these reports. Approval requests
raised by a subagent stay out of the report, so delegated work does not compete
with the main session for your attention.
Superset is the supported editor. The reports reach it only once Superset
registers mecatl as a hook-emitting agent. Until that registration ships,
mecatui sends nothing and no editor notifications appear. In any other
terminal, mecatui skips the report and its behavior is unchanged.
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 |
| Refresh direct MCP tools or broker workspace services | /mcp-refresh | Use learning and memory commands |
| Inspect named agent definitions | /agents | Named agents |
| Inspect available skills and the active soul | /skills and /soul | Skills, commands, and soul |
| Inspect saved memory | /memory | 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 the independent effective checker state. Off includes setup guidance; unavailable or older-server status is unknown. 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 tool calls, f9 to reveal conversation 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.
Inspect the broker catalogue
On a broker-only mecak8s connection, /mcp shows the current session's
enrollment state, connector names, catalogue state, and tool counts. The panel
requires a verified authenticated caller who owns the session. Opening it or
refreshing its display reads local state; it does not probe upstream services,
refresh credentials, or enroll connectors.
When the session is idle and the operator has enabled enrollment, /mcp (or
ctrl+o) offers Connect tools. Use it for a new session or to reconnect
after completed turns, a previous connection, a failed attempt, or a broker
restart. After a restart, saved connector names do not prove connectivity: the
panel reports broker state unavailable and protected tools remain unavailable
until you explicitly refresh that session.
Reconnect tools
Run /mcp-refresh to start broker setup. Reconnection withdraws the existing
broker tools for the whole bundle. Cancellation or failure leaves them
unavailable. Setup shows Setup in progress and Cancel setup; prompts and
a second refresh are blocked until it settles. A running session or one awaiting
approval cannot refresh. The browser authorization flow continues without a
reopen-browser action.
/tools-connect is a deprecated broker-only alias for /mcp-refresh.
/tools-cancel cancels pending setup.
ToolHive owns upstream OAuth authorization, callback state, credentials, token refresh, and grant reuse. Mecatl handles enrollment through opaque references. The operator configures MCP network access and credentials.
Broker-only and direct-MCP configurations are mutually exclusive. Broker-only sessions do not expose direct MCP resources, prompts, or groups. The catalogue is an inventory, not an upstream health check.
Next steps
- Manage sessions to resume, inspect, fork, or clear a chat.
- Use learning and memory commands when the server provides learning features.