Requirement 1
Python 3.12 or newer
Check it with python --version before you start the install.
The shortest supported path is: install the CLI, initialize it inside a git repository, verify the environment, then start one unattended run you can inspect in your normal review flow.
Fastest supported path
pipx install ralph-workflow
cd /path/to/your/project
ralph --init
ralph --diagnose
$EDITOR PROMPT.md
ralph
This is the same install-to-first-run path documented in the product docs, compressed onto one page.
Before you install
Ralph Workflow runs inside a git repository and reuses the AI agent CLIs you already have on your machine. The docs call out three prerequisites before the first run.
Requirement 1
Check it with python --version before you start the install.
Requirement 2
Most teams try Ralph Workflow in an existing repository they already care about, but a scratch repo works too.
Requirement 3
Usually that means claude, Codex CLI, or opencode. Ralph Workflow reuses those existing CLIs instead of asking you to re-enter credentials into a hosted product.
Install options
The docs and README support three clean paths: pipx, pip, or installing from source.
Recommended for first install
pipx install ralph-workflow
ralph --version
If you do not have pipx yet, the getting-started guide shows the bootstrap commands too.
Simple package install
pip install ralph-workflow
ralph --help
This matches the public README and is fine if that is already how you manage Python CLI tools.
For contributors or source builds
git clone https://codeberg.org/RalphWorkflow/Ralph-Workflow.git
cd Ralph-Workflow/ralph-workflow
pip install -e .
ralph --version
Use this path when you want the source tree locally or you plan to work on Ralph Workflow itself.
After install
Once the CLI is on your PATH, the onboarding flow is intentionally short: initialize the repo, verify the environment, write the task spec, then let the run start.
cd /path/to/your/project
ralph --init
ralph --diagnose
$EDITOR PROMPT.md
ralph
What each command does
Next move
Use a real but low-stakes repo first, keep the task small and concrete, then decide whether the handoff quality is good enough for higher-stakes work.