Agent transparency

What this accomplishes

See the agent’s options before it acts, approve its plan, and monitor sub-agents.

Miton shows the agent’s decision-making at three levels:

1. Option matrix — the agent’s options before it chooses a path
2. Pre-flight plan — the plan before it executes
3. Sub-agent dashboard — live monitoring of running sub-agents

Current availability

This page documents behaviour verified in Miton development builds. Miton is still pre-release and is not publicly downloadable; invited beta users will receive a signed desktop build with automatic updates.

Before you start

• A Miton development build and an active project.
• A connected cloud or local model, plus any credentials or local runtime required by this task.

Option matrix

When the agent encounters a task with two or more meaningfully different approaches, it shows you an option matrix before acting.

Each option card shows:

Click a card to proceed with that option, or Dismiss to close the matrix and answer in plain text instead.

The option matrix only appears when the agent identifies two or more meaningfully different approaches. For a task with a single clear path, the agent proceeds without showing options.

Pre-flight plan and task checklist

Before executing a multi-step plan, the agent presents a pre-flight plan — a summary of what it is about to do:

• The active project milestone and how many steps it involves
• A task checklist: every planned action, listed in order

Uncheck any step you don’t want the agent to take — the agent adapts around a skipped step rather than treating it as a hard requirement. Click Apply all to approve the plan and start execution, or the close control to cancel the plan outright.

Sub-agent dashboard

The sub-agent dashboard is the right-side panel that shows all sub-agents spawned in the current session.

Opening the dashboard

Click the Agents icon in the right sidebar.

What the dashboard shows

Each sub-agent card shows:

• Role — Researcher, Explorer, Verifier, Debugger, Test Runner, or Orchestrator
• Status — running / completed / cancelled / failed
• Elapsed time — how long it has been running
• Cost so far — cumulative token cost (estimated)

Setting ceilings

Every sub-agent run has two hard ceilings that auto-cancel the run when exceeded:

Cost ceiling:

• $0.10
• $0.50
• $2.00
• Unlimited

Time ceiling:

• 30 seconds
• 2 minutes
• 10 minutes
• Unlimited

Set ceilings before spawning a sub-agent, or update them on a running agent from its dashboard card. When a ceiling is tripped, the sub-agent is cancelled immediately and a banner summarises what it completed.

Cancelling a sub-agent

Click Cancel on any running agent card. The agent is killed immediately. Changes it has already made to files are not reverted — use per-turn rewind to restore to a state before the sub-agent ran.

Model and routing chips — current model

The roster bar above the chat input shows the Model chip (the active model, with a colour-coded dot for the current routing profile) and the Reasoning chip (routing profile and reasoning effort). Per-turn routing details — model selection, category classification, and cost — are also visible in each turn’s tool cards in the transcript. There is no separate routing-history panel; if you need historical routing data, review the turn-by-turn cards in the transcript or the Insights dashboard in Features.

What success looks like

See the agent’s options before it acts, approve its plan, and monitor sub-agents. The result remains attached to the active project when you move to another workspace.

Common failures and recovery

• Confirm the correct project and model are active.
• Read the exact error and the requested permission before retrying.
• Retry only the failed step so completed project work remains intact.

Related next action

• Per-turn rewind — restoring to a state before an agent action
• Cost intelligence — session cost tracking and budget alerts
• Sessions — session lifecycle and the status bar