Meowbert Docs

Appearance

Task Follow-up ​

After a task starts, the Task Detail view is where you steer the work, review what the agent did, and continue the conversation.

The events stream ​

The events stream (the main area of Task Detail) shows everything that happened during the task in chronological order:

  • User messages — your original prompt and any follow-ups.
  • Assistant messages — the agent's responses, reasoning, and plans.
  • Tool calls — every shell command the agent ran, including the full command text.
  • Tool output — the output (stdout/stderr) from each command.
  • Context compaction events — when the conversation was summarized to save context space (see Context Compaction).
  • Errors and cancellations — if something went wrong, the details appear here.

Workflow panel ​

If the task uses Long Horizon or Model Council, Task Detail also shows a dedicated workflow panel above the main conversation.

Long Horizon panel ​

The panel shows the state of the workflow at a glance:

  • the saved plan,
  • the latest submission,
  • the current round,
  • the reviewer verdicts.

You can open the full plan, full submission, and full review text from the panel.

Model Council panel ​

The panel shows:

  • the worker list,
  • the channel list,
  • live channel popups,
  • live worker POV popups.

This makes it much easier to inspect the ongoing council discussion without digging through raw logs.

Sending a follow-up message ​

While a task is awaiting_input or even after it has completed, you can type a follow-up in the input box at the bottom and press Send. This continues the same conversation — the agent sees the full history and your new message.

Use follow-ups to:

  • Correct an error: "The output file is missing the header row, please fix that."
  • Ask for clarification about what was done: "Explain why you chose that library."
  • Extend the work: "Now also generate a chart from output/q1_summary.csv."
  • Request a retry with different parameters: "Retry the analysis but exclude January."

Starting a thread from an assistant reply ​

Sometimes you want to ask a narrow follow-up without derailing the main conversation. In Task Detail, hover an assistant message and click the thread icon to open a sidebar thread.

Threads are useful when you want to:

  • ask a side question about one part of the reply,
  • explore an alternative approach without cluttering the main task,
  • or keep a focused mini-conversation tied to a specific assistant message.

If you highlight text inside an assistant reply, Meowbert also shows an Ask in thread action right next to the selection so you can start a thread from that exact excerpt.

How threads behave ​

  • Threads open in a right sidebar so the main conversation stays visible.
  • The sidebar is resizable on desktop.
  • Thread replies stay inside that thread instead of appearing in the main task history.
  • Existing threads on a message can be reopened from the same thread icon.

UI preferences ​

Open Preferences from the app sidebar if you want to change how task conversations behave.

Right now, the UI tab includes:

  • Enable Threads popup — controls whether the inline Ask in thread popup appears when you select text in an assistant reply.
  • Send with Shift+Enter — changes message sending so Shift+Enter sends and plain Enter inserts a newline.

These preferences are saved to your account, so they follow you across workspaces and future sessions.

Editing a previous message (branching) ​

You can click the edit icon on any of your previous messages to change what you said. When you do:

  1. Meowbert saves a branch at that point in the conversation.
  2. The agent re-runs from that message forward using your new wording.
  3. The original branch is preserved — you can switch back to it at any time.

This is useful when your original prompt had a mistake, was ambiguous, or you want to explore a different direction without losing what came before.

The Events tab ​

Click the Events tab in Task Detail to see a detailed log view with timestamps. This is where you go when:

  • A task appears stuck — inspect the last few commands to see what's happening.
  • You want to understand exactly what the agent did and why.
  • You need to copy a command that was run for debugging purposes.

Cancelling a task ​

Click the Cancel button to stop a running task. The agent will receive a cancellation signal and stop after its current step. Already-written files are not automatically deleted.

Re-running a completed task ​

After a task completes, you can start a fresh run from the same prompt by following up with a new message, or by opening a new task in the composer and duplicating the prompt.

Watching for compaction events ​

For long-running or multi-turn tasks, you may see a context compaction event in the stream. This means the conversation history was automatically summarized to keep the context window within model limits. Your follow-up messages still have access to all relevant prior context — the summary preserves goals, decisions, and open items. See Context Compaction for more.