Agents

Creating and Configuring Agents

Create agents from a prompt, a template, or a skillset, then configure the name, system prompt, model, and effort level.

Creating an Agent

From a Prompt

The fastest way is to type a prompt directly into the creation form. Gamut derives a name from the prompt, creates the agent, and starts a first session with your message.

  1. Open the "Create Agent" view.
  2. Type a description of what the agent should do (e.g., "Monitor my GitHub PRs and summarize them daily").
  3. Click Create Agent or press Cmd+Enter / Ctrl+Enter.

From a Template

Templates are pre-built agent configurations shared through Skillsets. Installing a template creates a new agent with pre-configured instructions, skills, and optionally an onboarding session that walks through setup.

  1. Click Browse Templates in the creation aids section.
  2. Select a template and follow the install dialog.

From an Import

Agents can be imported from a previously exported .agent file. Double-click the file, and Gamut restores the full agent workspace, including instructions, skills, and optionally secrets.

Agent Name and Description

  • Name: shown in the sidebar, agent home, and session headers. Rename an agent at any time by clicking the name on the agent home page or editing it in Settings > General.
  • Description: an optional short summary of what the agent does, stored in the CLAUDE.md frontmatter.

System Prompt

The system prompt defines how the agent behaves. It is stored as the body of a CLAUDE.md file in the agent's workspace directory, in standard Markdown. The default template looks like this:

# Agent Instructions
 
You are a helpful AI assistant.
 
## Preferences
 
<!-- Add your preferences here -- the agent will also append what it learns -->
 
## Project Notes
 
<!-- Add project context here -- the agent will also add notes as it learns -->

To edit the system prompt:

  1. Open the agent home page.
  2. Click System Prompt in the Extras panel on the right, or open Settings and select the system prompt view.
  3. Edit the Markdown content in the dialog.
  4. Click Save.

The system prompt is appended to the default Claude Code system prompt. Use it to specify the agent's personality, domain knowledge, preferred tools, coding standards, or any behavioral rules.

Model Selection

Each message can target a specific model family. The model selector appears in the composer toolbar next to the effort selector:

ModelDescription
Opus 4.8Most capable. Best for complex, long-horizon work.
Fable 5Newest model family.
Sonnet 4.6Balanced speed and capability.
Haiku 4.5Fastest and cheapest. Good for quick or simple tasks.

The first session of a new agent defaults to Opus. After that, Gamut remembers the model last used per session and seeds the selector accordingly. A default model can also be set in the app-wide settings, used as the fallback when no prior session context exists.

Effort Level

The effort level controls how much thinking the agent does before responding. It is selected per message alongside the model:

LevelDescription
LowFastest. Minimal thinking, terse answers.
MediumBalanced thinking and response depth.
HighDefault. Thorough planning and explanations.
Extra HighDeep reasoning for long-horizon work (Opus only).
MaxHighest effort (Opus only).

Extra High and Max are only available when the Opus model is selected. Switching to a model that does not support the current effort level resets it to High.

Settings Dialog

Open agent settings by clicking the gear icon on the agent home page. The dialog has several tabs:

General

  • Agent Name: rename the agent.
  • Template Status: if the agent was installed from a skillset template, shows whether it is up to date, locally modified, or has updates available. Force-sync, submit changes back to the skillset, or publish as a new template from here.
  • Export: export the agent as a shareable template (instructions and skills only) or as a full .agent archive (includes secrets and workspace data).
  • Session Auto-Delete: override the app-wide auto-delete policy for this agent. Inactive sessions older than the configured number of days are removed automatically; starred sessions are always preserved.
  • Danger Zone: permanently delete the agent and all its data.

Secrets

Manage environment variables injected into the agent's container. See Secrets.

Agents (Cross-Agent Policies)

Configure how other agents can invoke this agent with the multi-agent tool. See Multi-Agent.

Access (Auth Mode Only)

When authentication is enabled, the Access tab lets owners manage role-based permissions:

  • Owner: full control, including settings, secrets, and deletion.
  • User: can create sessions and send messages.
  • Viewer: read-only access to session history.

Agent Home Page

The agent home page is the central hub for interacting with an agent:

  • Composer: type a message to start a new session. Supports file attachments, voice input, and drag-and-drop.
  • Sessions list: all past sessions, sortable by newest or oldest, with search filtering. See Sessions.
  • Bookmarks: links and files pinned by the agent during conversations.
  • Triggers: scheduled tasks and webhook triggers for automation. See Automation.
  • Connections: connected OAuth accounts for external services. See Integrations.
  • Skills: installed skill extensions.
  • Volumes: mounted host folders. See Volumes and Mounts.