Customize your Agent from content to interface
Understand every layer of Agent configuration and use a safe sequence to turn brand requirements into tested behavior.
Customizing an Agent is more than changing colors or tone. A reliable customer experience combines verified Knowledge and Catalog data, explicit response guidance, carefully enabled tools, a tested support path, and an accessible chat interface.
Use this guide as the configuration map. It explains what belongs in each Humind area, in which order to work, and how to avoid masking a data problem with another instruction.
How this fits into Humind
AI Agent is the configuration space for the customer-facing assistant. Knowledge and Catalog provide facts; Guidance shapes response behavior; Tools add actions; Escalation defines human support; Test surfaces let you review the result before deployment.
Changes can affect many conversations, so test representative buying and support scenarios after each meaningful update. A visually correct chat is not enough: verify the answer, product context, available action, and handoff behavior together.
Before you start
Access: Agent configuration pages require agent_config access. Knowledge, Catalog, company settings, and deployment actions may require their own section permissions.
- Confirm the correct company and Shopify market, if the store uses Shopify Markets.
- Collect the approved brand voice, customer policies, support rules, and visual assets.
- Identify a small set of real buying and support questions for acceptance testing.
Step-by-step workflow
Build reliable information first
Open AI Agent, then Knowledge and Catalog. Review public policies, private operational guidance, products, collections, promotions, and filters. Facts that change with the commerce platform should normally remain in the catalog source; durable explanations belong in Knowledge.
Resolve obvious conflicts before tuning behavior. An Agent cannot consistently compensate for two sources that disagree about the same return window, shipping rule, or product attribute.
- Publish only reviewed Knowledge articles.
- Keep snippets concise and scoped.
- Verify representative products and variants after a catalog sync.
Shape response behavior in Guidance
Open Guidance to choose personality, answer length, emoji behavior, optional language enforcement, answer rules, and moderation behavior. The page saves changes automatically, so make deliberate edits and watch the save status.
Use answer rules for stable operating principles, not for large policy documents. Keep facts in Knowledge so they can be reviewed and updated without rewriting behavioral instructions.
Review tools and escalation
Open Tools and enable only actions whose dependencies are connected and tested. Some tools are configured elsewhere, and the interface indicates when configuration is missing. Custom tool creation is restricted to Humind administrators in the current workspace.
Open Escalation to configure support channels, human-handoff criteria, pre-handoff questions, operator step-in, and optional satisfaction collection. Make sure Inbox operators and roles are ready before advertising live support.
Customize the customer surface
Use Chat interface for avatar, colors, bubble and button styling, optional Agent name and home text, and other presentation settings. Use AI widgets, then Entry point, to control how the launcher appears on desktop, mobile, and selected page types.
Preview is useful for fast review, but it does not replace a live storefront test. Check contrast, overlap with store controls, product pages, consent, and the complete support journey after deployment.
Run a controlled acceptance test
Start in Playground with the agreed scenarios. Use Test product for a focused product conversation and Batch testing for a repeatable question set. Record failures by source: content, catalog, guidance, tool, escalation, interface, or deployment.
Change one layer at a time and rerun the failing scenario. Once the acceptance set passes, deploy through Sales channels and repeat a smaller smoke test on the live storefront.
Permissions and important caveats
- Navigation and edit controls are permission-aware; a missing page can be an access decision rather than a product failure.
- Guidance should not duplicate rapidly changing catalog values or full customer policies.
- Tools may require separate integration or store configuration before they work.
- Preview and Playground are testing surfaces; deployment must still be verified on the actual allowed domain.
Verify the result
Use this checklist before considering the work complete:
- Knowledge and Catalog contain no known conflicting facts.
- Guidance, tools, escalation, and interface settings each have a named owner.
- Buying, support, unsafe-input, and unavailable-product scenarios pass in test surfaces.
- The installed Agent works on desktop and mobile without covering essential storefront actions.
Troubleshooting
A configuration page is missing
Ask a company administrator to review your role and section permissions. Operators and supervisors intentionally do not receive the full Agent configuration area by default.
The Agent keeps using an old fact
Find the authoritative source in Knowledge or Catalog, remove conflicting copies, confirm publication or sync state, and begin a new test conversation after the source is current.
The preview looks correct but the storefront does not
Verify the Sales channels installation, the active domain, the correct company or market, and the live embed status. Then test in a private browser window to avoid stale state.