Docs / Agents/ Creating Custom Agents

Creating Custom Agents

Build your own AI agents in CRHQ — configure instructions, models, skills, and behavior for your specific workflow.

Custom agents let you tailor AI workers to your exact needs. Define their role, personality, capabilities, and constraints.

Creating a New Agent

  1. Go to Settings → Agents
  2. Click Create Agent
  3. Fill in the configuration:
FieldDescription
NameDisplay name (e.g., "SEO Specialist", "Data Analyst")
DescriptionShort summary of what this agent does
IconVisual identifier — choose from available icons
Default ModelHaiku, Sonnet, or Opus
InstructionsThe system prompt that shapes behavior
  1. Click Save

Cloning System Agents

The fastest way to create a custom agent is to clone an existing system agent:

  1. Go to Settings → Agents
  2. Find the system agent you want to start from
  3. Click Clone
  4. Modify the name, instructions, and settings as needed
  5. Save

The clone is fully independent — hub updates to the original won't affect your copy.

Writing Instructions

Instructions are the most important part of your agent. They define everything: role, tone, approach, constraints, and domain expertise.

Structure

A well-structured instruction set typically includes:

# Agent Name

**Role:** One-sentence description

## Capabilities
- What this agent can and should do
- Tools and skills it should use
- Domains it specializes in

## Communication Style
- Tone (formal, casual, technical)
- Response format preferences
- Verbosity level

## Process
1. How it should approach tasks
2. Steps it follows
3. Quality checks it performs

## Constraints
- What to avoid
- Security rules
- Boundaries

Best Practices

  • Be specific — "Write in active voice, max 3 sentences per paragraph" beats "write well"
  • Include examples — Show the agent what good output looks like
  • Set boundaries — Define what the agent should NOT do
  • Reference skills — Mention specific skills the agent should use for certain tasks
  • Iterate — Start simple, test, refine based on results

Version History

Every time you save changes to an agent's configuration, a new version is created automatically. You can:

  • View the full change history
  • See what changed and when
  • Restore any previous version with one click

This means you can experiment freely — if an instruction change makes things worse, just roll back.