Introduction

What is ody?

ody manages the full lifecycle of AI-assisted development: planning tasks, executing them via an agent, validating results, and optionally committing changes. It works with multiple AI backends and provides a structured workflow for turning natural language descriptions into completed code.

Key concepts

Backends

ody supports multiple AI coding agents as backends:

Claude Code (claude) -- Anthropic's CLI coding agent
OpenCode (opencode) -- OpenCode CLI agent
Codex (codex) -- OpenAI's Codex CLI agent

Each backend implements a common harness interface, so ody can orchestrate any of them with the same workflow.

Task plans

Tasks are structured Markdown files (.code-task.md) with YAML frontmatter that tracks status. Each task includes a description, technical requirements, implementation approach, and acceptance criteria. The agent works through tasks one at a time, transitioning them from pending to in_progress to completed.

The agent loop

When you run ody run, it enters a loop that:

Picks the next pending task
Builds a prompt from the task and project context
Sends the prompt to the configured backend
Monitors for a completion marker from the agent
Runs any configured validator commands
Marks the task as completed
Optionally commits changes to git

The loop continues until all pending tasks are done or the iteration limit is reached.

Prompt building

ody constructs prompts that give the agent full context about the task, the project structure, and any validation commands to run. This structured prompting helps the agent produce focused, high-quality work.

Getting started

Install ody
Run ody init in your project to create a configuration
Run ody plan to generate task plans from your descriptions
Run ody run to execute the agent loop

See the Configuration guide for details on customizing ody's behavior.