Harmont

Website · Docs · Slack

Warning

Harmont is in early alpha.

Today it's a powerful task runner -- like make or just, but with DAG-based parallel execution, Docker isolation, layer caching, and typed toolchain presets for many languages.

The cloud CI/CD platform at harmont.dev is under active development. APIs will change. We'd love your feedback -- join the community.

The performance of the hm CLI is not as good as I'd like it to be. I'm actively working on cross-run caching. The code quality is similar -- needs improving and is a work in progress.

hm will always remain open-source, and pluggable into any CI/CD provider.

What is Harmont?

Harmont lets you define CI/CD workflows in TypeScript or Python and run them instantly on your machine in Docker containers. No YAML. No commit -m "run ci" --allow-empty spam. Each pipeline step runs in an isolated container with built-in caching, parallel execution, and consistent environments.

tui-demo.mp4

Features:

• Pipelines as real code - Python or TypeScript, not YAML.
• Instant local runs - hm run executes in Docker on your machine.
• DAG-based parallelism - independent chains run concurrently.
• Layer caching - Docker snapshots are reused across runs; only changed steps re-execute.
• Typed toolchains - first-class presets for Rust, Go, Python, Java, C++, React, and more.

Quick Start

0. Install hm

curl -fsSL https://get.harmont.dev/install.sh | sh

Or via Cargo:

cargo install harmont-cli

1. Create a pipeline

Save this as .harmont/pipeline.py (or .harmont/pipeline.ts):

Python

import harmont as hm
from harmont.python import PythonToolchain

@hm.target()
def project() -> PythonToolchain:
    return hm.python(path=".")

@hm.pipeline(
    "ci",
    default_image="ubuntu:24.04",
    triggers=[hm.push(branch="main")],
)
def ci(project: hm.Target[PythonToolchain]) -> tuple[hm.Step, ...]:
    return (
        project.test(),
        project.lint(),
        project.fmt(),
        project.typecheck(),
    )

TypeScript

import { pipeline, push, type PipelineDefinition } from "harmont";
import { python } from "harmont/toolchains";

const project = python({ path: "." });

const pipelines: PipelineDefinition[] = [
  {
    slug: "ci",
    triggers: [push({ branch: "main" })],
    pipeline: pipeline(
      project.test(),
      project.lint(),
      project.fmt(),
      project.typecheck(),
      { defaultImage: "ubuntu:24.04" },
    ),
  },
];

export default pipelines;

2. Run it

hm run ci

If the repo declares only one pipeline, the slug is optional - just hm run.

Browse the example projects for idiomatic pipelines in Rust, Go, Python, Java, C++, React, Next.js, and more.

GitHub Actions

Use harmont-dev/actions-hm to run your pipelines in GitHub Actions with automatic Docker image caching:

name: CI

on: [push, pull_request]

permissions:
  contents: read
  packages: write        # needed for Docker image caching via GHCR

jobs:
  ci:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: harmont-dev/actions-hm@main
        with:
          pipeline: ci

The action installs hm, runs your pipeline, and caches Docker images in GitHub Container Registry so subsequent runs skip unchanged steps. No actions/cache configuration required.

Multiple pipelines

jobs:
  lint:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: harmont-dev/actions-hm@main
        with:
          pipeline: lint

  test:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: harmont-dev/actions-hm@main
        with:
          pipeline: test
          parallelism: 4

Without caching

- uses: harmont-dev/actions-hm@main
  with:
    pipeline: ci
    cache: 'false'

See the action repo for the full input reference, sub-actions, and caching details.

Documentation

For the full pipeline reference, richer examples, and more - see the docs.

License

The CLI is dual-licensed under either of

• Apache License, Version 2.0 (LICENSE-APACHE)
• MIT license (LICENSE-MIT)

Motivation

The reason I started this project is because every other CI/CD tool I've used in my life has sucked.

I've worked at Tesla, Bun, Mesa and never did I find a CI/CD system that was easy to use and was also fast.

At Tesla, we used Jenkins -- executors are finite, so your builds are stuck in queues.

At Bun, we used Buildkite -- large shell pipelines, and really pricy service, and a TS SDK that's only slightly better than YAMLs.

At Mesa, I migrated everyone to use BuildBuddy and Buildkite. Bazel is awesome, but the mental overhead required to use it is way too high. We, sadly, ended up reverting to plain Buildkite.

I asked myself a couple questions:

• Why can't I run my CI/CD pipelines locally? act is an awesome project, but it's surprisingly slow (not to the author's fault -- but rather GHA's model).
• Why is my CI/CD system not just a Makefile? Why is there no hm run command that is shared between local dev and CI/CD?
• Why can't I get preview environments for Haskell, Rust, Zig or whatever? Vercel does an awesome job with next.js preview environments, but there is no good way to do this for arbitrary environments.
• Why do we have to write YAMLs for our pipelines? All my pipelines end up being YAML documents from hell. I think we can do better.
• Why do I need artifacts-upload and artifacts-download everywhere? I don't need it locally, so why do I need it in CI/CD? In other words, why aren't our CI/CD systems stateful? If my build scripts can write an openapi.json in the local directory, why do I need some magic to transfer it between individual steps?

Harmont's goal is to make all these questions obsolete. CI/CD can be better, and that's what Harmont wants to be -- a CI/CD that sucks a lot less.