Dashboard

Inbox & history

The Inbox is your daily queue: one card at a time, approve or ship the agent's work. History is the full record of every conversation the bot has touched.

Inbox

The Inbox is a single-card approval queue, not a multi-pane CRM. The agent does the work first (drafts a reply, proposes a fix, or flags a moderation action); each item surfaces as one card with a classification badge and a one-line summary. You approve, regenerate, or ship. When the queue is empty, the agent handled everything within its autonomy and nothing needs you.

Each card carries:

A classification badge (the action the agent chose) and a short summary of the thread.
View thread — open the full conversation in a modal without leaving the queue.
A Trace view of the agent's steps (the tool calls and reasoning events it logged).
The primary action for that card: approve the draft, Regenerate it, or Ship with Cursor / Fix with DuggAI when the fix is code.

Discord moderation lives here too

If your org has moderation enabled, flagged messages appear in this same queue as their own card type (Ban / Mute / Dismiss). See Discord moderation. There's one operator queue, not a separate surface per use case.

Keyboard shortcuts

The Inbox is built to clear fast from the keyboard:

⌘ ↵ — ship / approve the current card.
⌘ R — mark resolved.
⌘ G — regenerate the draft.

History

Every conversation the bot has touched, across the widget and Discord, lands in History (reached from the History link in the header, at /dashboard/conversations). It's a look-back log, separate from the Inbox queue. You can search by text and filter by status:

All — every conversation.
Open — still active.
Escalated — handed to a human.
Resolved — closed.

Conversation detail

Open any conversation to see:

The full transcript.
The user's identity when available (email, external ID, channel).
Page context the widget captured (the URL / context passed via setContext).
For each agent reply: the model and route used, latency, and the sources the agent pulled (search_docs hits with their similarity score).
Mark resolved to close the thread manually.

No confidence score

DuggAI doesn't expose a confidence number or slider — that was removed in favor of explicit reasoning. Instead of a score, each reply shows the actual sources and tool calls behind it. See How the agent decides.

Taking over a Discord thread

Marking a conversation resolved closes it. For Discord, replies the bot posts go out under the bot's identity so users see one consistent voice. The agent escalates (pinging your configured role) when it needs a human rather than handing you a half-finished draft to send manually.