Agent Configuration

This page describes Otto Builder, Otto's ticket-driven workflows that work from Linear or Jira tickets. It is separate from Otto Detective, the pull request reviewer, which never commits, pushes, or merges. Start with the Introduction for Otto Detective.

Fine-tune your agents for better code generation.

Creating an Agent

1. Go to Agents in the sidebar
2. Click Create Agent
3. Fill in the configuration
4. Click Create

Configuration Options

Name

A descriptive name for the agent (e.g., "Backend Agent", "Frontend Agent").

Model

Choose the AI model:

Organization Context

Information about your organization that helps Otto write better code:

## Tech Stack

- Backend: NestJS with TypeScript
- Database: PostgreSQL with Drizzle ORM
- Frontend: Nuxt 3 with Nuxt UI

## Conventions

- Use dependency injection for services
- All endpoints require authentication except /health
- Use class-validator for DTOs
- Tests are required for new endpoints

## Patterns

- Repository pattern for database access
- DTOs for all request/response bodies
- Centralized error handling

Triggers

Associate triggers that activate this agent. One agent can have multiple triggers.

Advanced Configuration

Custom System Prompt

Override the default system prompt for specialized behavior:

You are a security-focused code reviewer.
Always check for:

- SQL injection vulnerabilities
- XSS vulnerabilities
- Authentication/authorization issues
- Sensitive data exposure

Max Iterations

Limit how many times the agent can iterate (default: 15 for trigger agents, 50 for code agents).

Temperature

Control creativity vs. consistency (default: 0 for code generation).

Multiple Agents

Consider creating separate agents for:

• Different codebases — Frontend vs. backend
• Different purposes — Implementation vs. review
• Different teams — Each team's conventions

Best Practices

Provide Rich Context

The more context you provide, the better Otto's code matches your standards.

Start Conservative

Begin with a simpler configuration and add complexity as you learn what works.

Iterate on Context

Review Otto's PRs and update the context when you notice patterns to correct.