Basic settings
| Field | Description | Default |
|---|
| Name | Display name for the agent | Required |
| Description | Short description of what the agent does | Optional |
| Icon | Bootstrap icon class for visual identification | bi-robot |
| Status | Active, Inactive, or Archived | Active |
LLM Model
Choose which language model powers your agent. Available models vary by provider:
- Azure OpenAI — GPT-4o, GPT-4o Mini
- OpenAI — GPT-4o, GPT-4 Turbo, o1, o3
- Anthropic — Claude Sonnet, Claude Haiku, Claude Opus
Each model has different capabilities, context windows, and credit costs. See Billing & Credits for pricing details.
Behavior Prompt
The behavior prompt is the system message sent to the LLM. It defines the agent’s personality, instructions, and constraints. Write it as if you’re briefing a human assistant.
You are an expert data analyst. When given a dataset description:
1. Identify key patterns and anomalies
2. Suggest relevant visualizations
3. Provide actionable insights
Always cite specific data points in your analysis.
Be concise — limit responses to 500 words.
Be specific about output format, constraints, and edge cases. The more precise your prompt, the more consistent the agent’s behavior.
Execution settings
| Setting | Description | Default |
|---|
| Max Steps | Maximum tool-call iterations per run (1–12) | 8 |
| Conversation Memory | Include previous messages from the same conversation | Enabled |
| Auto Tool Selection | Let a ToolSelector system agent dynamically pick tools | Disabled |
Max Steps
Each “step” is one LLM call that results in a tool invocation. If the agent reaches the step limit without finishing, it returns whatever output it has so far. Higher limits allow more complex multi-step reasoning but consume more credits.
Conversation Memory
When enabled, the agent receives the full conversation history with each new message. This enables multi-turn interactions where the agent remembers context. Disable it for stateless, one-shot tasks to save tokens.
Knowledge Base
Attach a Knowledge Base to give the agent access to your documents. The agent automatically gets a SearchKnowledge tool that performs semantic search over uploaded documents.
Add tools that the agent can call during execution. These can be:
- Built-in functions — platform-provided utilities
- Custom tools — workflow automations you’ve built
- Integration tools — functions from connected services
- Other agents — invoke another agent as a sub-task
