Agents

Agent Nine can create background agents to handle tasks in parallel — like having a team of AI assistants working simultaneously.

How It Works

When you ask for something complex, Agent Nine may create sub-agents that work in parallel:

Refactor the auth module and write tests for it at the same time

One agent explores the codebase, another writes the implementation plan

Each agent runs independently with its own context and reports results back to you.

Built-in Agent Types

Type	Purpose	Example
General Purpose	Any task — coding, analysis, debugging	"Fix the login bug"
Explore	Fast codebase search and navigation	"Find all API endpoints"
Plan	Architecture and implementation planning	"Plan the database migration"
Verification	Check implementations, run tests	"Verify the auth changes work"
Guide	Answer questions about tools and features	"How do hooks work?"

When Agents Are Created

Agent Nine decides whether to use agents based on the task:

Scenario	Agents?
Simple question or quick edit	No — handled directly
Complex task with independent parts	Yes — parallel agents
Research + implementation	Yes — one explores, one implements
Large refactoring	Yes — multiple agents handle different files

Monitoring Agents

Active agents appear in the status area. You can:

• See progress — live updates on what each agent is doing
• Cancel — stop a running agent if you change your mind
• Read output — view the agent's results when it finishes

Messaging Between Agents

Agents can communicate with each other via a mailbox system:

Send a message to the test-writer agent: "Auth module is ready for tests"

Check your inbox:

Read my messages

This enables coordination — one agent can wait for another to finish before continuing.

Agent Isolation

Each agent works in isolation to avoid conflicts:

• Separate context — agents don't interfere with each other's conversations
• Independent control — cancel one agent without affecting others
• Git worktree — agents can work on isolated copies of the repository, so file changes don't conflict

Examples

Parallel development

Create a team: one agent refactors the database layer,
another updates the API endpoints, third writes tests

Research and implement

Explore how authentication works in this project,
then implement OAuth2 support

Code review

Review the changes in the last 5 commits and
create a summary of potential issues