Airgentic Help
This guide is for Airgentic staff who run onboarding sessions. It provides session structure, timing, adaptation guidance, and facilitator notes. Customer-facing onboarding content is in the other pages in this section.
| Duration | 60–75 minutes |
| Format | Remote video call with screen sharing |
| Attendees | Platform admins, content owners, operations staff, and (optionally) managers or sponsors |
| Objective | Equip customers to operate their Airgentic service confidently and independently |
Before the session:
Purpose: Set the stage and confirm objectives.
Purpose: Build a shared understanding of how Airgentic works.
Cover the major building blocks:
| Component | What it is |
|---|---|
| Knowledge sources | Where information comes from (web crawl, uploaded documents) |
| Search | How the AI finds relevant content |
| Agents | Specialist AI personalities for different question types |
| Prompts | Instructions that control agent behaviour |
| Functions | Actions agents can perform (search, email, APIs) |
| Curated answers | Human-defined overrides for specific questions |
| Human handoff | Monitoring and intervention capabilities |
| Insights | Analytics on performance and usage |
Explain how these work together: user asks question → agent selected → search retrieves content → AI generates answer → conversation logged.
Reference: Platform Mental Model
Purpose: Show the platform in action with a realistic walkthrough.
Demonstrate a typical flow:
If the customer has specific agents or functions configured, include those in the demonstration.
Facilitator note: Keep this practical and connected to the customer's actual service. Avoid getting lost in edge cases.
Purpose: Cover the practical tasks customers do most often.
Walk through:
| Task | Where to do it | Why it matters |
|---|---|---|
| Testing with Admin Chat | Admin Chat | Verify changes before going live; use Trace Log for debugging |
| Reviewing conversations | Insights → Threads | Monitor quality and spot issues |
| Checking flagged conversations | Insights → Needs Review filter | Find conversations that need attention |
| Running a data sync | Data Sync | Refresh content after website updates |
| Adding a curated answer | Curated Answers | Override or guide the AI for specific questions |
| Viewing usage | Usage Dashboard | Track consumption against allowances |
Mention but don't deep-dive:
- Agent and prompt editing (point to documentation)
- Search settings (point to documentation)
- Human handoff screen (show briefly if relevant)
Reference: Admin and Operator Essentials
Purpose: Help customers understand what to change safely and when to ask for help.
Explain the safety model:
| Safe to change | Change with care | Ask Airgentic first |
|---|---|---|
| Curated answers | Agent prompts | Service-wide prompts |
| Service messages (welcome, no-answer) | Agent roles | Identity provider configuration |
| Scheduling data syncs | Search settings | Advanced function code |
| User interface labels | Agent status (Testing → Live) | |
| Privacy retention settings |
Key messages:
- Test in Admin Chat before making changes live
- Prompts are version-controlled — they can restore previous versions
- Not every setting needs to be touched
- Contact support if they're unsure
Reference: Guardrails and Governance
Purpose: Provide a simple decision framework.
Quick overview of when to use each capability:
| Need | Use this |
|---|---|
| General question answering from your content | Default Frontline agent and search |
| Exact wording or specific source required | Curated Answers |
| Different behaviour for different question types | Specialist agents |
| Actions like email, API calls, or escalation | Functions |
| Restrict to authenticated users | Secure Services |
| Let customers speak instead of type | Voice mode |
| Monitor and intervene in conversations | Human Handoff |
Emphasise that most services work well with just the basics. Advanced features are available when needed.
Reference: What to Use When
Purpose: Address customer-specific questions.
Invite attendees to ask questions gathered before or during the session.
If questions require deep-dives that would take the session over time, note them for follow-up (and optionally schedule a deep-dive session).
Purpose: Provide clear actions and close the session.
Summarise key takeaways:
- They can now test, review, and make safe changes independently
- Documentation is available at help.airgentic.com
- Contact support for guidance when needed
Point to follow-up resources:
- Post-Session Next Steps
- Optional deep-dives for specific topics
- Contact Support for help
If follow-up sessions are needed (deep-dives, additional attendees), schedule them before ending.
| Scenario | Adjustment |
|---|---|
| Simple service (no specialist agents or functions) | Spend less time on agents and functions; focus on content and curated answers |
| Complex service (multiple agents, functions, integrations) | May need a longer session or schedule deep-dive follow-ups |
| Technical audience | Can go deeper on prompts, functions, and the Trace Log |
| Non-technical audience | Keep it high-level; focus on outcomes rather than implementation |
| Secure service | Include identity provider and authorisation overview |
| Question | Suggested response |
|---|---|
| How do I know if the AI is answering correctly? | Use Admin Chat to test; check Insights for conversation outcomes and user ratings; use the Trace Log to see what content was used |
| What happens if the AI gives a wrong answer? | Add a curated answer for that question; improve content and re-sync; or refine the prompt and test before going live |
| How often should we sync content? | After significant website changes; many customers also set a weekly or daily schedule in Data Sync |
| Can we change the AI's personality? | Yes — via prompts. Edit the agent's prompt (or shared prompt components); test in Admin Chat; use version history to roll back if needed |
| How do we handle sensitive or off-topic questions? | Use guardrails in the prompt to decline or redirect; add curated answers for specific intents; consider human handoff for escalation |