Agents
What is an Agent?
An agent is an AI assistant that belongs to an agency. You configure each agent with a personality, language model, and a set of capabilities. Users interact with agents through conversations.
Agent Properties
Identity
| Property | Description |
|---|---|
| Name | The agent’s display name |
| Description | What this agent does |
| Avatar | A profile image for the agent |
| Start Message | The first message shown to users when they open a conversation |
| Custom Instructions | Instructions that define the agent’s behavior and tone |
Language Model
| Property | Description |
|---|---|
| Model | The AI language model powering the agent |
| Temperature | Controls response creativity (lower = more predictable, higher = more creative) |
| Max Tokens | Limits the length of the agent’s responses |
Features
| Feature | Description |
|---|---|
| Knowledge Retrieval (RAG) | When enabled, the agent searches its associated documents before answering |
| Suggested Options | Shows reply suggestions to users during a conversation |
| Prompt Processing | Applies pre-processing to user inputs before sending them to the model |
| Events | Allows the agent to react to and emit Evolvable.ai events |
Agent Capabilities
Tools
Agents can be equipped with external tools that allow them to fetch data or perform actions during a conversation. Tools are added and managed from the agent’s configuration.
Workflows
Agents can trigger workflows — automated sequences of steps that run in response to a conversation. Multiple workflows can be assigned to a single agent.
Guardrails
Guardrails are safety and compliance rules that constrain what an agent can say or do. They can block specific phrases, enforce tone, or restrict topics.
Knowledge (RAG)
When knowledge retrieval is enabled, the agent pulls relevant content from its associated documents before generating a response. Documents are organized using document tags — associating a tag with an agent gives it access to all documents under that tag.
Top-Level vs. Internal Agents
Agents can be set as top-level or internal:
| Type | Description |
|---|---|
| Top-level | Users can start a conversation directly with this agent |
| Internal | Only reachable from other agents; not visible to users as an entry point |
Agent-to-Agent Communication
Agents within the same agency can hand off conversations to one another using conversation flows. This enables multi-agent pipelines where a front-facing agent routes users to specialized agents behind the scenes.
Rules:
- Only top-level agents are visible to users as starting points.
- Internal agents can only be reached from other agents.
- A top-level agent cannot be the destination of another agent’s handoff.
Conversation flows are managed from the agency’s flow configuration.