Simply AskingDocs

AI Agents

Create AI agents that draw from your entire knowledge system: documents, entity relationships, and graph intelligence.

AI Agents are customizable chatbots that draw from your entire knowledge system, including documents, entity relationships, and graph intelligence. Each agent has its own persona, behavior settings, guardrails, and can be deployed to different parts of your website.

Per-Agent ConfigurationEach agent has unique persona, grounding mode, and guardrails
Lifecycle ManagementDraft → Active → Paused → Archived progression

The Create Agent wizard guides you through four steps. Go to AI → Agents → Create Agent.

1

Basic Info

Name: A friendly name for your agent (e.g., "Support Assistant")
Slug: URL-safe identifier, auto-generated from name (e.g., "support-assistant")
Description: Brief description of what this agent does

2

Persona

Choose a personality template or write a custom persona. This defines how your agent communicates. You'll also set a welcome message, the first thing users see when the chat opens.

3

Configuration

Grounding Mode: How strictly the agent sticks to your documents. See Agent Configuration for details.
AI Model: ChatGPT or Gemini. Requires your API key. Response style and length are managed automatically for optimal quality.

4

Data Access

Control which documents the agent can access. You can scope by All documents (entire knowledge base), Specific departments (limit to certain team's documents), or Tagged documents (only documents with specific tags).

Pro Tip: After creation: Your agent starts in Draft status. Configure guardrails, test thoroughly, then activate when ready.

Agents move through a lifecycle. Each status determines whether the agent is receiving live traffic.

Draft: In DevelopmentNot live. Configure settings, set up guardrails, and test in the sandbox. No external traffic.
Active: LiveReceiving traffic from your website. Conversations are logged. Changes take effect immediately.
Paused: Temporarily DisabledNot receiving traffic. Widget won't load for visitors. Use when you need to make significant changes.
Archived: Soft DeletedHidden from the main list. Data preserved for analytics. Can be permanently deleted from this state.

Click on any agent to open its detail view. You'll see these tabs:

OverviewQuick stats (conversations, satisfaction, escalation rate), setup checklist for drafts, recent conversations, and usage summary.
ConfigurationPersona, grounding mode, AI model, response settings, and data access. See Agent Configuration guide.
RulesGuardrails and restrictions. See Agent Guardrails guide.
Chat TestTest your agent with real conversations before going live.
ConnectionsWebsite widget setup, API keys, and embed codes. See Website Widget guide.
ActivateStart receiving live traffic. Agent must have a valid API key configured.
PauseTemporarily stop traffic. Widget stops loading on your website.
DuplicateCreate a copy of an agent with all settings. Useful for A/B testing or creating variants.
ArchiveSoft-delete the agent. Preserved for analytics. Can be restored or permanently deleted.
DeletePermanently remove an archived agent and all its data. This action cannot be undone.

The Chat Test tab lets you have real conversations with your agent using the same pipeline as production. Always test before activating.

Test with real customer questions from your support logs
Verify guardrails trigger correctly on restricted topics
Check that responses cite the correct documents
Test edge cases: ambiguous questions, multi-topic questions, out-of-scope requests
Preview the widget branding in the Chat Test panel

Before activating, verify these items:

LLM API key is configured (OpenAI or Google)
Persona and welcome message are set
Guardrails are in place for sensitive topics
Tested with at least 10 representative questions
Website widget embed code is installed (if using web deployment)
Agent won't activateCheck that an LLM API key is configured and has credits. The activation button is disabled without a valid key.
Agent gives wrong answersCheck grounding mode (try Strict), verify relevant documents are in the knowledge base, and narrow data access scope.
Widget not showingVerify the agent is Active (not Draft/Paused), embed code is installed correctly, and API key origin restrictions include your domain.
Too many escalationsConfidence threshold may be too high. Adjust in Operations → Settings → Threshold Tuner.
Was this helpful?