Simply AskingDocs

Agent Configuration

Configure how agents access your knowledge system: personas, grounding modes, and response settings.

The Configuration tab in your agent's detail view gives you complete control over how your agent accesses the knowledge system. Grounding modes determine how much of the system the agent can draw from, while persona and response settings shape how it communicates.

Access configuration at AI → Agents → [Your Agent] → Configuration

Basic agent metadata that appears in the dashboard and widget.

NameDisplay name shown in the widget header. Keep it short and recognizable (e.g., "Support Bot", "Sales Assistant").
SlugURL-safe identifier used in API calls and embed codes. Auto-generated from name but can be customized. Cannot be changed after creation if agent has been activated.
DescriptionInternal description for your team. Not shown to users. Helpful for documenting the agent's purpose.
Welcome MessageThe first message users see when opening the chat. Set the tone and let users know what the agent can help with.

The persona defines your agent's personality, communication style, and behavior guidelines. This is the "system prompt" that shapes every response.

Persona Templates

Friendly AssistantWarm, approachable, uses casual language. Good for consumer products.
Technical ExpertPrecise, detailed, uses technical terminology. Good for developer tools.
Sales RepresentativeHelpful, solution-focused, guides toward action. Good for lead generation.
CustomWrite your own instructions from scratch for full control.

Writing a Custom Persona

A good persona includes:

Identity: Who is the agent? What's its role?
Tone: Formal or casual? Concise or detailed?
Boundaries: What should it refuse to do?
Escalation: When should it suggest human help?

Grounding modes control how strictly the agent sticks to your documents vs. using general knowledge. Choose based on your accuracy requirements and use case.

BalancedHelpful while grounded in your knowledge base. Will use documents as primary source but can fill in with general knowledge for context.
Best for: General support, product information, FAQs
CautiousPrioritizes accuracy over helpfulness. Will decline to answer if not confident, and clearly indicates when information comes from documents vs. general knowledge.
Best for: Technical documentation, compliance-sensitive topics
StrictOnly responds from your documents. If the answer isn't in your knowledge base, the agent will say it doesn't know and offer to escalate.
Best for: Legal, medical, financial information where accuracy is critical
ComplianceMaximum restrictions. Documents-only with mandatory citations. Automatically escalates uncertain responses. Detailed audit logging.
Best for: Healthcare, finance, legal, regulated industries
Pro Tip: Tip: Start with Balanced and adjust based on testing. If you see hallucinations or off-topic responses, try Cautious or Strict.

AI Model

ChatGPTOpenAI's fast, affordable model. Great balance of quality and speed.
Gemini (gemini-flash)Google's quick response model. Good for high-volume use cases.

Response style and length are managed automatically for optimal quality. You can influence tone through your agent's Persona settings.

Page Awareness gives your agent context about where visitors are on your website. When a visitor asks a question on your pricing page, the agent knows they are looking at pricing and can tailor its response accordingly.

Awareness Modes

Route OnlyThe agent sees the current URL path. Minimal setup, but the agent has no additional context about what each page contains.
SitemapDefine your site pages with titles, descriptions, and help topics. The agent uses this context to provide relevant, page-specific assistance. Pages can be added manually or imported from a sitemap.xml file.
Full ContextThe agent receives the full page content. Highest context quality, but increases token usage per message.

Importing from Sitemap XML

Instead of adding pages one at a time, you can import your entire site structure from a sitemap.xml file.

1

Get your sitemap

Most websites have a sitemap at yoursite.com/sitemap.xml. You can also export one from your CMS (WordPress, Webflow, Squarespace, etc.).

2

Click "Import Sitemap"

In the Sitemap Configuration section, click the Import Sitemap button. You can upload the .xml file directly or paste its contents.

3

Review and select pages

After parsing, you will see all URLs found in the sitemap. Select the pages you want your agent to know about. You can add to your existing pages or replace them entirely.

4

Add titles and descriptions

Imported pages have their URL paths pre-filled. Expand each page to add a title, description, and help topics so the agent has the richest context possible.

Pro Tip: Tip: After importing, prioritize adding descriptions to your most-visited pages first. The agent can still match by URL path for pages without descriptions, but rich descriptions produce significantly better responses.

Control which documents the agent can search when answering questions. Narrower scope = faster responses and more relevant answers.

All DocumentsAgent searches your entire knowledge base. Best for general-purpose agents.
By DepartmentLimit to specific teams' documents. Useful for department-specific agents (e.g., HR bot only sees HR docs, Sales bot only sees sales materials).
By TagsOnly documents with specific tags. Useful for product-specific agents (e.g., "Product A Support" only sees docs tagged "product-a").
Important: Note: Data access is enforced at query time. Changing scope affects future responses immediately. The agent cannot access documents outside its scope, even if asked directly.
Start narrow, expand as neededBegin with Strict grounding and narrow data access. Loosen settings only if the agent is too limited. It's easier to add flexibility than to constrain a too-liberal agent.
Test with real questionsUse the Chat Test tab with actual customer questions from your support logs. This reveals gaps in your knowledge base and configuration issues.
Match persona to audienceTechnical users prefer concise, jargon-friendly responses. General consumers need simpler language. Enterprise clients expect professional tone.
Keep responses appropriately sizedShort for quick FAQ answers, Medium for most support, Long for detailed explanations. Very Long is rarely needed and can overwhelm users.
Review and iterateUse the Operations portal to review conversations. Look for patterns in escalations and knowledge gaps. Adjust configuration based on real performance.
Was this helpful?