Create and configure an agent

This guide shows you how to hire and configure an agent in your workspace.

Hire an agent

Navigate to Agents in the sidebar
Click Hire agent
Choose a role — see roles below
Fill in the configuration fields
Click Hire

Alternatively, ask your CEO agent to hire one: @ceo hire a coder named alice.

Choose a role

Role	Best for
CEO	Coordination, hiring, delegation
Coder	Writing and modifying code
Reviewer	Reviewing pull requests
QA	Testing and quality assurance
Designer	UI/UX implementation
Security	Security review and auditing
DevOps	Infrastructure and deployment
PM	Product management, planning
General	Flexible, multi-purpose

The role determines the default persona, memory access grants, and reporting hierarchy.

Configure the agent

Name and title

Name: a short identifier used in @-mentions (e.g. “alice”, “reviewer-bot”)
Title: a human-readable label (e.g. “Senior coder”, “Security reviewer”)

Adapter and model

Adapter: the CLI backend (Claude Code, Codex, Pi) — Control Center auto-detects installed CLIs
Model: the specific model within the adapter (e.g. Claude Sonnet, GPT-4o)

If your adapter isn’t detected, register it manually in Settings → Adapters.

Skills

Skills are free-form identifiers that inject relevant context into the agent’s prompt. Add skills that match your project’s technology stack:

flutter, dart — Flutter/Dart conventions
rest-api — REST API patterns
typescript, react — TypeScript/React conventions

Skills are case-insensitive.

Effort

Controls reasoning effort:

Low: fast responses, less thorough — good for simple tasks
Medium: balanced — good for typical work
High: thorough, detailed — good for complex tasks

Persona

The persona is free-text that shapes the agent’s behavior. Write in natural language:

You are a senior Flutter developer who prefers simple, readable solutions.
Avoid over-engineering. Write tests for all new code.

Capabilities

Capability flags control what the agent can access:

Push to repo: allow pushing commits
GitHub API: allow GitHub API calls
Ticketing: allow ticket creation and updates
Network egress: allow outbound network requests

Disable capabilities you don’t need to minimize the agent’s attack surface.

Budget

Set a monthly budget (in cents) to cap the agent’s spending. When the budget is exceeded, the agent is blocked from dispatching until the next month.

Import from disk

Control Center scans for AGENTS.md files that define agent configurations. If found, agents appear as discovered in the registry and can be imported with one click.

Verify the setup

After hiring, run Agent doctor from the agent’s context menu to verify:

The adapter CLI is installed and accessible
The model is available
The sandbox backend is working
Network connectivity is fine