Configure an agent

Open a company in HQ Console, choose Agents, then open the gear menu for an agent. Saving settings updates the running agent when possible. A change that cannot be applied live is saved for the next start.

Configuration reference

Setting	Purpose	Notes
Handle	Stable agent identifier	Chosen during creation. It is used in company identity, email, and vault paths and is not a normal profile edit.
Display name	Human-readable name shown in HQ and connected channels	The Console enforces channel-safe length limits.
Avatar	Profile image	HQ validates the image and file size before saving.
Title	Short role label	Use a role people can recognize, such as Research Lead.
Description	One-line responsibility statement	State what the agent owns and when teammates should contact it.
Provider	Runtime family	Selected at creation from the providers enabled for your account. Provider migration is not available in general Agent Settings.
Model	Model used for new work	Must be compatible with the provider and authentication method.
Reasoning effort	Depth used for supported models	Console values are Low, Medium, High, and Extra high. Grok does not use this setting.
Speed	Standard or Fast service tier	Available for supported Codex configurations. Fast can increase provider cost.
Membership role	Company permissions granted to the agent	An agent can be a member or admin, but cannot become the company owner.
Owners	People responsible for the agent	The creator is the initial owner. Additional ownership is shown only when enabled for the company.
Groups	Company groups the agent belongs to	Use groups for organization and policy-driven access.
HQ MCP	Agent connection to HQ tools and context	Owners and admins manage it under Integrations > For Agents while the runtime is running.
Slack	Slack app connection and diagnostics	Attach, finish a pending install, inspect delivery, or intentionally detach the app.
Telegram	Telegram bot connection and caller group	Connect a managed bot or supply a token, then choose who can call it.
Runtime controls	Start, stop, reauthenticate, or remove	Removal deprovisions the agent and requires confirmation.

Model compatibility

HQ validates the model name against the selected provider:

• Codex supports OpenAI model families such as gpt-* and o*.
• Grok supports grok-* models.
• Claude supports Anthropic model families such as claude-*, sonnet, opus, and haiku.

The available list is the source of truth for your account. Some models require API-key authentication even when the same provider also supports subscription sign-in. The Console also prevents combinations that the runtime cannot honor. For example, gpt-5.6-sol cannot use Extra high reasoning.

Profile guidance

A useful agent profile answers three questions:

• What does this agent own?
• What requests should teammates send it?
• What should a teammate do when the request is outside its scope?

Keep the title short and make the description operational. Channel users should be able to decide whether to mention the agent without opening HQ Console.

Managed settings

The backend also supports lower-level controls for instance size, autoscaling, caller email allowlists, provider migration, and direct MCP state. These are API or control-plane fields, not standard Agent Settings controls. An invalid value can prevent the runtime from starting or reject channel callers.

Credentials are stored as vault references. Raw provider keys and channel tokens are never returned by the settings API.