AI support agent overview

Helpin's AI support agent answers customer questions directly from your knowledge sources, cites where answers came from, asks clarifying questions when a request is ambiguous, and hands off to your human team the moment it is out of its depth.

The AI-first model

When AI support is enabled, every new customer message flows through one pipeline:

1. Message arrives (widget chat, email, or internal/API) and enters the conversation pipeline.

2. AI evaluates the question against your connected knowledge sources — docs spaces, help center articles, and website content.

3. AI decides how to respond, based on confidence and configuration:

   • Reply directly — post an answer to the customer in the thread.

   • Add internal note — leave a draft/summary as a private note for agents instead of a public answer.

4. If confidence is too low or the request needs a human — the AI escalates: the conversation is marked "Escalated by AI", lands in the configured handoff inbox, and the customer sees your escalation message.

5. Human takes over — from that point the conversation is human-driven; AI is out of the loop for that thread unless the customer's later message re-enters as a fresh turn.

What the AI answers from

The AI answers only from content you connected:

• Docs spaces (internal or external-capable)

• Published help center articles

• Website content sources added in Knowledge settings

It does not invent answers from its own training data — when your content doesn't cover a question, the correct behavior is escalation, not improvisation. See Knowledge-base answers.

What customers experience

• AI replies appear in the same thread as human replies, identified as coming from your workspace's AI.

• Answers cite the sources used (see AI answer citations & source transparency).

• Customers can ask follow-ups; the AI keeps conversation context.

• A Talk to a person option (if enabled) requests a human at any time.

What agents experience

• AI-active conversations appear in the AI Handling view with an "AI Active" badge.

• Escalations arrive in the destination inbox with escalation context (what the AI tried, what the customer asked).

• Agents can take over any AI conversation at any moment — takeover wins.

• AI-resolved conversations appear in AI Resolved for review.

Core configuration (Settings → AI Assistant)

• Enable — the master switch.

• Support agent — which configured support agent persona handles conversations.

• When a message comes in — Reply directly (public AI answer) or Add internal note (private note for agents).

• Confidence — the threshold (10%–100%) below which the AI hands off instead of answering.

• Handoff after — how many AI follow-ups on the same problem (1–10) before a forced handoff; the count resets when the visitor raises a new problem.

• Escalation messages — per-scenario customer-facing copy: Default, Team busy, After hours.

• Routing & assignment — links to where handoffs go (see Inboxes / routing settings).

• Guidance — per-agent instructions for tone and behavior (see Support AI assistant configuration).

The sections in this collection

• Knowledge-base answers — how retrieval over your content works.

• AI answer citations & source transparency — how sources are shown.

• AI clarifying questions — disambiguation before answering.

• Handoff to human agents — escalation mechanics.

• Stuck detection & same-issue handoff — when the AI gives up.

• Escalation rules — instant handoff categories.

• AI response modes & follow-up limits — reply vs. note, limits.

• Support AI settings — where AI settings live.

• Support AI assistant configuration — personas and guidance.