Agents
What is an Agent?
An Agent is a self-contained AI entity that perceives user inputs, reasons about goals, and executes actions to complete tasks autonomously. Unlike static APIs or simple chatbots, an Agent can:
- Chain multiple steps — e.g., read a document, extract data, validate it against rules, and generate recommendations.
- Persist context across interactions through short-term chat history.
- Adapt dynamically based on instruction changes, new actions, or external triggers.
- Delegate work to other Workflows, or external APIs via configurable Actions.
Key Terms Glossary
| Term | Definition |
|---|---|
| Orchestrator | The primary LLM node responsible for intent understanding and action routing. |
| Planner | An optional LLM node that decomposes complex requests into a step-by-step sequence of actions. Conversational interaction and parameter collection are handled by the Orchestrator. |
| Formatter | An optional LLM node that converts raw responses into markdown or other human-readable formats for cleaner, more structured output. |
| Renderer | An optional LLM node that converts raw responses into rich chat output such as interactive forms, cards, charts, and quick replies. |
| Action | A callable tool (Workflow or API) that the Agent can invoke to interact with external systems. |
| Strict Mode | A per-action flag ensuring the Agent will not invoke that action until every required input parameter has been explicitly provided by the user. |
| Input Schema | A JSON structure defining what arguments an Action expects. The Agent uses this to extract parameters from the user's message. |
| Output Schema | A JSON structure defining what data an Action returns. |
| History Context | The number of recent conversation turns passed to the Orchestrator as context for each new request. |
| Chat Session | An isolated conversation thread in the Builder's test chat. Multiple sessions can be created to test different scenarios. |