v0.1 · alpha · MIT

helm

AI agent orchestration
for backend development.

terminal
$ curl -fsSL https://getkaze.dev/helm/install.sh | bash $ helm init
View on GitHub → Docs All tools
13
agents total
5
pipeline phases
10
governance rules

Helm was born from the observation that AI-assisted development works best when structured. Instead of free-form prompting, Helm guides your project through a pipeline of specialized agents — from discovery and planning to implementation and deployment — each with clear acceptance criteria and quality gates.

Helm is a state manager, not an executor. Agent logic lives in markdown definitions and runs inside Claude Code. The CLI manages session state, displays progress, and bridges the gap between conversations. Stop, resume, and checkpoint at any point.

Named after the helm — the wheel that steers a ship's direction. The helm is what translates the captain's intent into the rudder's movement. Without it, a ship drifts.

The name reflects the tool's purpose: to steer your development pipeline with structure and intent — keeping every agent on course, every handoff clean, and every decision traceable.

Structured Pipeline

Five phases — Discover, Plan, Build, Quality, Deploy — with 13 agents. Each agent has a clear mission, inputs, outputs, and success criteria.

Specialized Agents

Scout, Research, Architect, Build, Verify, Ship — each agent drives one phase of work. No generic prompting. Every agent knows its mission from the handoff.

Quality Gates

Two QA checkpoints at 95% threshold: Review validates planning traceability, Verify validates implementation. Nothing ships without passing both gates.

Session Continuity

Checkpoint state with helm save, close Claude Code, and resume later with helm resume. All progress survives across sessions.

Execution Profiles

Three modes — Explore (read-only), Guided (confirm before writes), Autonomous (auto-execute when gates pass at 95%). Safety net triggers revert to guided.

Full Traceability

Every task traces to a requirement. Every requirement traces to a research problem. Review validates this chain end-to-end before build starts.

Stack

CLI

Go 1.26 · spf13/cobra
gopkg.in/yaml.v3 · fatih/color

Agent Runtime

Claude Code · Markdown definitions
YAML session state · Handoff documents

Distribution

Single binary · ldflags version injection
linux/amd64 · linux/arm64 · macOS (amd64 + arm64)

Quality

10 governance rules · 95% QA gates
Given-When-Then acceptance criteria