Built-in agents for common workflows, plus how to create custom agents with any Gloo AI model.
GlooCode uses a multi-agent architecture. Each agent is a specialized mode optimized for a different type of work — planning, implementing, researching, or general coding. You can switch between agents mid-session, or let GlooCode select the right one automatically.
Type the agent command (e.g., /plan) in the TUI to switch. You can switch as many times as you want within a session. Your conversation history carries over, so the new agent has full context of what you’ve been working on.
Use plan when you’re starting something complex — a new feature, a migration, a refactor that touches many files. It will think through the approach, identify risks, and lay out a step-by-step plan before any code is written.Use build when you know what you want and need focused execution. You have a plan (or the task is straightforward enough to not need one), and you want code written, tests passing, and changes committed.Use explore when you’re unfamiliar with a codebase or need to understand how something works before changing it. It reads files, traces call chains, and summarizes what it finds.Use general when the task is self-contained and doesn’t need a specialized approach. This is the default — if you’re not sure which agent to use, start here.
You can define your own agents that use any model available through Gloo AI. Custom agents let you create specialized tools for your team’s specific workflows — code reviewers, QA testers, documentation writers, security auditors, or anything else.Custom agents are configured in ~/.config/gloocode/gloocode.json:
Permissions control what tools a custom agent can use. Set each to "allow" or "deny". If you omit the permission block entirely, all permissions default to "allow".
Permission
What it controls
glob
Search for files by name or pattern
grep
Search inside file contents
read
Read file contents
bash
Execute shell commands (tests, builds, scripts)
edit
Create or modify files
lsp
Language server features (go-to-definition, find references)
webfetch
Fetch content from URLs
websearch
Search the web
skill
Use skills and plugins
task
Create and manage task lists
question
Ask you clarifying questions
Best practice: Read-only agents (reviewers, auditors) should set "edit": "deny" to prevent accidental modifications. This is especially useful when you want an independent second opinion — the agent can analyze but not change anything.