A Lens in the AI’s brain

Overview

In Conduit, every AI message is explainable. You can see the agent’s reasoning, the sources it used, and the exact next step to teach, draft, or test—without leaving the thread. When an AI message is generated it will be one of three types:

• Sent message: The AI had high confidence and sent the reply automatically (only on autopilot)

• Proposed reply (Co-Pilot): The AI drafted a response for your approval.

• Urgent message: The message was tagged as urgent based on predefined conversation tags and the generation was escalated regardless

• Approve/Send

  • Proposed reply: Accept or edit, then send.

• Escalation – Needs Help: Take the required action, then reply (use ⌘J if you want a draft).

• Escalation – AI Missing Info: Teach with ⌘K, optionally draft with ⌘J, then send.

Inspect the AI’s thinking

• AI Reflection (ℹ︎): Press the “i” to view how the AI analyzed the conversation, what context it considered, and the plan it formed.

• Sources: Expand Sources to see every document, KB entry, or data point the AI referenced. Open any source to edit or improve it.

Act directly from the inbox

• Teach (⌘K)

• Use when the AI needs knowledge or a rule:

  • Create or edit Knowledge.

• Create or edit a Guardrail (policy/rule).

  • Saving updates immediately improves future replies.

• Draft (⌘J)

• Use when you want the AI to generate a draft reply in the composer:

  • The draft appears in the same component where you can edit or accept.

  • Works for both proposed replies and escalations where you want a jump-start.

Escalation: The AI held back and asked for help. Two subtypes:

• Needs Help: Action required (e.g., confirm a booking, approve a task).

• AI Missing Info: Knowledge is missing; you need to teach it.

Test the AI

After you’ve taken an action (e.g., teaching or editing a source), you can verify that the AI has learned properly in one of two ways:

• Regenerate the message: If the message is a Proposed reply or AI Missing Info (i.e., it hasn’t been sent yet), simply regenerate it in place to see the updated behavior.

• Recreate in Sandbox: If the message has already been sent or you want to simulate the exact scenario, press Recreate in Sandbox in the sidebar. This opens the conversation in a test environment with the same context and knowledge snapshot.

Calls

Calls allow for a centralized space where you can see inbound and outbound calls answered by your team or the AI.

You can click on a call line to get the full transcript, and an AI summary of the conversation.