Skip to content

Agent Configuration

Agent configuration defines behavior, capabilities, and constraints. Each setting controls how the agent processes requests and interacts with tools.

Configuration Fields

Field Type Required Description
name string Yes Unique identifier (no /, \, .)
card string Yes Brief capability description shown in UI
model string Yes LLM model name (must be registered)
category string Yes Classification for organization
instruction string No System prompt defining behavior
max_actions integer No Iteration limit (1-200, default: 50)
mcp array No List of MCP server names
system_tools array No List of system tool names
tags array No Labels for search (max 20)

Field Details

name

Unique identifier used in API calls and UI selection.

Rules: - Cannot contain /, \, or . - Must be unique per user - Cannot be changed after creation

Examples: - ✅ code-reviewer - ✅ data_analyst - ❌ my/agent (contains /) - ❌ agent.v2 (contains .)

card

Short description displayed in agent lists and selection menus.

Best practices: - Keep under 100 characters - Focus on what the agent does, not how - Use action verbs

Examples: - Reviews code for quality, security, and best practices - Analyzes data and generates visualizations - Writes technical documentation from code

model

The LLM that powers the agent's reasoning and tool usage.

Requirements: - Model must be registered in Model Manager - Model must be active and accessible

Common choices: - claude-sonnet-4.5 - Best for complex reasoning - gpt-5 - Strong general-purpose performance - deepseek-v3 - Fast and cost-effective

category

Classification for organizing agents in the UI.

Common categories: - coding - Code generation, review, debugging - analysis - Data analysis, research - writing - Documentation, content creation - automation - Task automation, workflows - support - Customer support, Q&A

instruction

System prompt that defines the agent's personality, expertise, and behavior. Leave blank for general-purpose assistants.

Example: 

You are a senior Python developer specializing in data engineering.
When helping users:
- Prioritize pandas and numpy solutions
- Always include error handling
- Explain performance implications

max_actions

Maximum number of tool calls and reasoning steps before stopping.

Range Use Case
1-10 Single-step tasks, simple Q&A
10-30 Multi-step workflows, basic research
30-50 Complex analysis, iterative refinement (default)
50-100 Deep research, extensive code generation
100-200 Long-running automation, comprehensive tasks

mcp

List of MCP server names the agent can access. Each server must be configured and running in MCP Manager.

See: MCP Integration

system_tools

List of built-in system tools the agent can use.

See: System Tools

tags

Labels for search, filtering, and discovery. Maximum 20 tags.

Example: ["python", "data", "pandas", "analysis", "visualization"]

Runtime Configuration Override

When running an agent, you can override or extend configuration:

Override Effect
model_name Use different model
system_prompt Replace instruction
mcp_servers Add additional MCP servers
system_tools Add additional system tools
max_iterations Change iteration limit

Next Steps