Agents

Understanding AI agents in CRHQ — what they are, how they work, and how to configure them for your workflow.

Agents are the core of CRHQ. Each agent is an AI worker with a specific role, set of capabilities, and behavior defined by its instructions.

What is an Agent?

An agent is a configured AI identity with:

PropertyDescription
NameDisplay name (e.g., "Developer", "Researcher", "Content Writer")
DescriptionWhat this agent specializes in
IconVisual identifier shown in the sidebar and session headers
Default ModelThe AI model it uses — Haiku (fast), Sonnet (balanced), or Opus (most capable)
InstructionsSystem prompt that defines the agent's behavior, tone, and approach
SkillsCapabilities assigned to the agent (browse web, query databases, etc.)
RecipesProcedural workflows the agent can follow
MCP ServersAdditional tool servers for extended functionality

System Agents vs Custom Agents

CRHQ comes with system agents — pre-configured agents maintained and updated through CRHQ Hub. These cover common roles like development, research, writing, and analysis.

You can also create custom agents tailored to your specific needs. Custom agents are fully under your control and won't be modified by hub updates.

Tip: You can clone a system agent as a starting point for your custom agent. This copies all settings while giving you full control to modify them.

Agent Status

When viewing agents in the sidebar, you'll see status indicators:

  • Online (green dot) — Agent is ready and responsive
  • Processing (pulsing) — Agent is actively working on a task
  • Offline (grey) — Agent process is not running
  • Reconnecting (amber) — Agent is reconnecting after an interruption

Choosing the Right Model

Each agent has a default model that determines its capabilities and speed:

ModelBest ForSpeedCapability
HaikuQuick tasks, simple Q&A, data formattingFastestGood
SonnetMost tasks — coding, writing, analysisFastVery Good
OpusComplex reasoning, architecture, nuanced workSlowerBest

You can also override the model per-session if a specific conversation needs more (or less) capability.

Agent Instructions

Instructions are the most powerful way to shape an agent's behavior. They define:

  • Role — What the agent is and does
  • Communication style — Tone, verbosity, format preferences
  • Capabilities — What tools and approaches to use
  • Constraints — What to avoid, security rules, guardrails
  • Domain knowledge — Context about your business, products, or processes

Instructions are written in markdown and support any level of detail. They're versioned automatically — every change is tracked and can be restored.

Example Instructions

# Content Writer Agent

**Role:** Create and edit marketing content for SaaS products.

## Style
- Write in active voice
- Keep paragraphs under 3 sentences
- Use concrete examples over abstract claims

## Process
1. Ask clarifying questions before writing
2. Create an outline first
3. Write the full draft
4. Self-review for tone consistency

## Don't
- Use jargon without explaining it
- Make claims without evidence
- Include placeholder text in deliverables

Managing Agents

Navigate to Settings → Agents to:

  • View all available agents (system and custom)
  • Click an agent to see its full configuration
  • Edit instructions, model, or skill assignments
  • View version history for any agent
  • Check for hub updates (system agents)