Agents
Overview
An Agent is an AI assistant that handles conversations. Each agent is configured with a language model, custom instructions, knowledge sources, tools, and safety rules.
Creating an Agent
- Open an agency from the Agencies page.
- In the Agents tab, click New Agent.
- Fill in the configuration tabs and click Save.
Configuration Tabs
General Info
| Field | Description |
|---|---|
| Name | Display name of the agent. |
| Avatar | Optional image for the agent. |
| Description | Short description of what the agent does. |
| Model | The LLM used for responses (selected from configured models). |
| Custom Instructions | System prompt defining the agent’s behavior and persona. |
| Start Message | Opening message the agent sends when a conversation begins. |
| Max Tokens | Maximum tokens per response. |
| Temperature | Controls response randomness (0 = deterministic, higher = more creative). |
| RAG Enabled | Whether the agent uses uploaded documents as context. |
| Options Enabled | Enables structured option buttons in responses. |
| Events Enabled | Allows the agent to emit and respond to events. |
Knowledge
Assign document tags to the agent. The agent will use documents with matching tags as context for responses (RAG). Go to Knowledge to upload and tag documents.
Actions
Configure automated actions the agent can trigger during a conversation.
Tools
Attach OpenAPI tools to the agent. These are external API endpoints the agent can call as part of its reasoning (function calling). Tools are defined in Settings → OpenAPI Tools.
Workflows
Attach Workflows the agent can trigger. When the agent determines a workflow should run, it invokes it as part of the conversation.
Integrations
Connect the agent to external messaging platforms:
- Slack — Deploy the agent as a Slack bot.
- WhatsApp — Deploy the agent as a WhatsApp bot.
Memory
Configure persistent memory for the agent. Agents with memory can recall facts across conversations.
Guardrails
Assign Guardrails to the agent. Guardrails are checked on every message (prompt phase) or response (response phase) to enforce content safety rules.
Evaluations
View and manage evaluation runs for the agent to assess response quality.
Testing an Agent
Click the Playground Chat button on the agent page to open a chat interface and test the agent’s responses directly.