rupu documentation

rupu is an agentic code-development platform. You describe the work; agents and workflows do it — from the terminal, autonomously on triggers, or from a visual control plane.

Don't code it, rupu it. Define agents as plain Markdown, compose them into workflows, run them by hand or on triggers, and review everything with full transcripts and approvals.

Install

rupu is open source. Grab the latest signed macOS binary, or build from source with Cargo (any platform with a Rust toolchain).

macOS (Apple Silicon)

Download the latest signed & notarized tarball from the latest release (or the Download section), then:

# unpack, make executable, put it on your PATH
tar -xzf rupu-*-aarch64-apple-darwin.tar.gz
chmod +x ./rupu
sudo mv ./rupu /usr/local/bin/rupu
rupu --version

From source

cargo install --git https://github.com/Section9Labs/rupu

Your first run

Initialize rupu in a repo, authenticate a provider, then run an agent:

# scaffold .rupu/ (agents + workflows) in the current repo
rupu init

# connect a model provider (Anthropic, OpenAI, Gemini, Copilot…)
rupu auth login

# run a one-off agent against a prompt
rupu run code-reviewer --prompt "Review the diff on this branch"

Everything rupu does is recorded as a JSONL transcript you can inspect with rupu transcript.

The four building blocks

Read these in order — each builds on the last.

Agents →

A single AI worker defined in Markdown + YAML: provider, model, tools, permissions, and a system prompt.

Workflows →

Multiple agents composed into steps — sequential, fan-out, parallel panels, gated review loops, approvals.

Autoflows →

Workflows that run themselves — fired by cron or SCM events, claiming issues and opening PRs autonomously.

Sessions →

Long-lived, interactive agent conversations — in the terminal or live in the web control plane.

Then go further

Quickstart →