The editor Assistant¶

Status: Current | Last reviewed: 2026-09-18

The Assistant uses supported defaults for routine authentication and call ending. For protected member information, it selects native Customer Authentication while respecting your identity setup and the caller's consent. It keeps these controls in capability settings and the prompt, and asks about business choices that still need your input.

The Assistant tab in the agent and workflow editors lets you chat with an AI about the agent or workflow you have open — ask how it works, dig into why something failed, or just describe a change and watch it apply.

Don't see the tab? The Assistant uses the same switches as Generate with AI: the deployment feature flag, the tenant AI Generation setting, and a configured AI engine integration. If any is off, the tab simply doesn't appear.

Where it lives¶

Open any agent or workflow. The right-hand panel (where Preview / Tester lives) gains a second tab: Assistant. Switching tabs doesn't disturb a live test chat or call — your test conversation is still there when you flip back.

What you can ask¶

Questions — plain-language answers about the open agent/workflow:

• "What tools does this agent have, and what does each one do?"
• "Which model is this using and where's the system prompt?"
• "What happens after the router node if the score is low?"

Debugging — the Assistant can read this agent's past conversations (and a workflow's past runs) including transcripts, events, and errors:

• "Look at the last conversation — why did it end?"
• "Why did yesterday's run fail?" (it reads the per-node trace and quotes the failing node's error)
• "Did callers ever ask about something the agent couldn't answer?"

Edits — describe the change; the Assistant validates and applies it to the draft:

• "Make the greeting mention our opening hours."
• "Add an example to the prompt about handling refunds."
• "Rename the second step and make it wait 5 minutes."

When it saves, the canvas and inspector update immediately — you'll see a small "Assistant updated the draft" notice. Everything is a normal draft edit: review it, tweak it, or undo it like any of your own changes. The Assistant can never publish.

The Assistant can also inspect other agents in your workspace, available capabilities and their configuration guidance. It can research public sources and, when connected, previous TrueServe Library demos and transcripts. Ask it to explain the examples and defaults it used. Existing OpenAI chats keep their original tool set; choose New chat to use newly added research tools.

Conversations persist¶

Your chat with the Assistant sticks to that specific agent/workflow — leave the editor and come back, and the conversation is still there (for up to a week of inactivity). Use New chat to start fresh.

Each reply can run for up to 30 minutes by default. If it reaches that limit, saved draft changes remain available; review them before continuing the chat.

Tips¶

• Be specific: "shorten the greeting to one sentence" beats "improve the greeting".
• For debugging, mention which conversation/run you mean ("the one around 2pm", "the failed run") — otherwise it starts with the most recent.
• Big restructures are better done through Generate with AI or by hand; the Assistant is at its best for targeted changes and explanations.