Agents Overview
Agents are AI-powered assistants that learn about your application and interact with your users. They can answer questions, demonstrate workflows, and take actions directly in your product — all based on the knowledge and instructions you provide.
How agents work
Section titled “How agents work”Each agent belongs to an Application and can be configured with:
- Custom instructions — Natural language directives that define the agent’s behavior, tone, and focus areas.
- Knowledge — Product documentation, guides, and other content the agent references when responding.
- Voice — A male or female voice used for text-to-speech interactions.
Once configured, agents can:
- Answer user questions via embedded widgets on your site.
- Run simulations to autonomously explore your application and build understanding.
- Execute QA tests to validate workflows and catch regressions.
Agent status
Section titled “Agent status”An agent’s status tells you whether it’s ready to interact with users:
| Status | Meaning |
|---|---|
| Active | The agent is fully ready and can respond to users. |
| Learning | The agent is processing knowledge or simulation data. It can still respond, but may not have full context yet. |
| Error | Something went wrong during processing. Check the agent details for more information. |
Learning progress
Section titled “Learning progress”When an agent is learning, two progress indicators show how far along it is:
- Vector index — Tracks text search indexing. This powers the agent’s ability to find relevant passages in your documentation.
- Graph index — Tracks knowledge graph construction. This gives the agent a structured understanding of how concepts in your product relate to each other.
Both indicators must complete before the agent has full access to its assigned knowledge.