Skip to main content
Doppels is a CLI tool that turns one-off AI agent sessions into replayable, locally-executed automations. You explore a problem with Cursor, Claude, Codex, or OpenCode — then ask the agent to freeze what it just did. The agent writes a typed, validated Capability (the contract) and a Recipe (the steps) straight into your repo. Every subsequent run executes locally on your machine, consuming zero AI tokens.

Introduction

Understand what Doppels is, the problem it solves, and the core concepts: Capabilities, Recipes, and Runs.

Installation

Install the doppels CLI via curl, Homebrew, or build from source on macOS or Linux.

Quickstart

Initialize a Space, validate manifests, and run your first Capability in under five minutes.

How It Works

Dive deeper into the Capability–Recipe–Run model and how the freeze skill interacts with the CLI.

​
Get started in three steps

1

Install the Doppels CLI

Download the binary using the one-line curl installer:
Verify the install:
2

Install the freeze skill

Add the doppel-freeze skill to your agent (once per machine or project):
The skill teaches Claude, Cursor, Codex, or OpenCode how to write valid Capability and Recipe YAML and validate it with the CLI.
3

Ask your agent to freeze what it just did

Do your real work in the agent — deploy, migrate, fix prod, automate a report. When it works, tell the agent:
doppel freeze — turn what we just did into a Capability
The agent writes YAML into capabilities/ and recipes/, loops doppels validate until clean, and asks you to confirm before it’s done. Commit the YAML to git — that’s your reusable asset.
Zero tokens on replay. The freeze step happens once while the agent is working. Every subsequent doppels run executes the deterministic Recipe locally — no model calls, no re-prompting, no cost.
Recipes are plain YAML committed to your repo alongside your source code. Fork them, review them in PRs, and share them with teammates — execution always stays on the local machine.