Skip to main content
Version: 2025.6

Chatbots

Chatbots (AI Agents) are intelligent assistants that can help learners with personalised guidance and instant answers throughout their experience in Reflectis.

Overview​

Agents leverage the Research Archive (Vector Stores) to provide context-aware responses based on your organization's knowledge base.

Creating a Chatbot

Chatbots are created from templates and then tailored with parameters (name, audience, tone, topics, grounding, etc.).
Templates and fields may vary; what follows shows the typical flows.

1) Pick a template​

From Create New Chatbot choose a starting template.

  • Basic templates → quick setup (usually just Label and Description).
  • Structured templates → full configuration (identity, audience, tone, constraints, topics, grounding…).

bo_create_new_chatbot_from_template

Templates are examples. Your workspace may expose different names/models (e.g., “Product expert”, “Tutor”, “Support bot”).

2) Basic template flow (quick)​

Fill in:

  • Chatbot Label – the visible name.
  • Chatbot Description – short purpose statement.
  • Enable Research Archive – allows the agent to search uploaded documents.

Click Create. You can edit later.

bo_chatbot_basic_creation

3) Structured template flow (full)​

The exact fields differ by template; these are the most common.

A. Identity & Knowledge​

bo_new_chatbot_first_part

  • Chatbot Label – public name.
  • Chatbot Description – purpose.
  • Enable Research Archive – let the agent search uploaded documents.
  • agent.name – internal agent identity.
  • domain.subject – primary domain focus (e.g., museum education, cyber security, product Q&A).

B. Audience & Style​

bo_new_chatbot_first_part

  • audience.levels – beginner, intermediate, advanced.
  • tone.formality – casual / neutral / formal.
  • style.verbosity – succinct / balanced / elaborate.
  • style.creativity – conservative / medium / high.
  • response.depth – shallow / balanced / deep.

C. Constraints & Topics​

bo_new_chatbot_first_part

  • constraints.wordLimit – max response length (characters).
  • topics.allowed – comma-separated whitelist of topics the bot should focus on.
  • topics.restricted – topics to avoid/refuse.

D. Grounding & File Search​

  • grounding.filesearch – controls how the agent retrieves information:
    • ON (retrieval-first): The agent queries the attached vector store first and bases answers on retrieved documents plus the conversation context. Recommended for factual, document-grounded responses (policies, manuals, FAQs).
    • OFF (reasoning-first): The agent relies only on its internal knowledge and the current conversation context, without searching the vector store. Better for general assistance or creative tasks.

4) Review & Create​

  • Use Go Back to change the template if needed.
  • Click Create Chatbot to save. The agent appears in your list and can be edited anytime.

Notes & Best Practices​

  • Field variability: each template exposes different parameters—fill what you see.
  • Grounding strategy:
    • Enable grounding.filesearch (ON) when you need accurate, document-based answers (product specs, company policies, training materials).
    • Disable it (OFF) for general conversations or when your vector store doesn't contain relevant information.
  • Audience first: set audience.levels and tone to calibrate explanations.
  • Safety: use topics.restricted to fence off unwanted areas.
  • Iterate: test with real questions, then refine topics, tone, and constraints.