Ralph turns a natural-language goal into a PRD, then implements the work story by story using an AI coding backend.
Supported backends:
- pi
- OpenCode
- Claude Code
- Cursor Agent (
agent)
- optionally ask clarifying questions
- generate a PRD
- review the PRD
- implement stories in priority/dependency order
- run tests and retry failed stories
- Go 1.24+
- Git
- One of the supported CLIs on
PATH:piopencodeclaudeagent(Cursor Agent)
go build -o ralph .
go install .ralph "build a todo app"
ralph "build a todo app" --dry-run
ralph --resume
ralph statusUse RALPH_RUNNER to select the AI runner binary. Ralph does not pass a model to the runner; configure model selection in the runner itself.
Supported values:
pi
cursor
claude
opencode
go test ./...
go build -o ralph .