Getting Started

Helm is a CLI tool that manages AI agent pipeline sessions for backend development. It's a state manager — agent logic runs inside Claude Code, while Helm handles session state, checkpoints, and cross-conversation continuity.

Install

Run the install script to download the binary:

curl -fsSL https://getkaze.dev/helm/install.sh | bash

This installs the binary to ~/.local/bin/helm.

From source

git clone https://github.com/getkaze/helm.git
cd helm
make build

Binary is output to bin/helm. Add it to your PATH or run directly.

Quick Start

Initialize a new project:

helm init

This creates .helm/session.yaml (runtime state) and helm.yaml (project config).

Inside Claude Code, activate the orchestrator:

/helm

The orchestrator reads session state and routes you to the correct agent in the pipeline.

Check pipeline progress at any time:

helm status

Checkpoint your session before closing:

helm save

Resume in a new Claude Code session:

helm resume

How It Works

Helm guides projects through a structured pipeline of 13 AI agents across 5 phases:

1. Discover — Understand the problem (Scout/Survey + Research)
2. Plan — Define what to build (Planning, Architect, Roadmap, Breakdown)
3. Quality — Validate the plan (Review — 95% gate)
4. Build — Implement code (Build agent)
5. Deploy — Ship it (Verify + Ship)

Each agent produces a handoff document for the next agent. Every task traces to a requirement, and every requirement traces to a research problem.

Project Types