What an agent is
An agent is a named AI employee configured for a job. It combines identity, instructions, model settings, knowledge, skills, capabilities, channels, and triggers into one reusable assistant.
Main parts
| Part | Purpose |
|---|---|
| Profile | Name, description, avatar, welcome message, and system instructions. |
| AI model | Provider, model, and tuning parameters that control response behavior. |
| Response settings | Tone and output format, including structured output when needed. |
| Capabilities | High-level permissions such as file access, code execution, workflow management, or API access. |
| Skills | Reusable playbooks or procedures the agent can load for specific work. |
| Knowledge | Knowledge bases the agent can read from or write to. |
| Interfaces | Custom forms or UI surfaces that collect structured input before chat. |
| Integrations and triggers | External systems and events that start or support agent work. |
| Channels and web embed | Places users can reach the agent outside the builder. |
How agents differ from normal chat
Normal chat is flexible. An agent is repeatable. Use an agent when the same role, rules, tools, and knowledge should apply every time.