MITOpen source and self-hosted

Your CircleCI config. Your runners.

ForgeCI runs your existing .circleci/config.yml on machines you control. Push to GitHub and Docker runners pick up the jobs, stream the logs live and report back to the pull request.

Get started View on GitHub

.circleci/config.ymlv2.1

# The config you already have.
version: 2.1

jobs:
  test:
    docker:
      - image: cimg/ruby:3.3
    resource_class: spot
    parallelism: 4
    steps:
      - checkout
      - run: bundle install
      - run: |
          circleci tests glob "test/**/*_test.rb" |
            circleci tests run --split-by=filesize \
              --command="xargs bin/rails test"

workflows:
  ci:
    jobs: [test]

Config: CircleCI 2.1
Runners: Outbound only
Built with: Ruby + Rails
Per-minute fees: None

Features

What you use in CircleCI. On your hardware.

A Rails control plane, a pure-Ruby config compiler and zero-dependency Docker runners.

Config

Bring the config you have.

Executors, commands, parameters, workflows, requires, matrix, branch filters and contexts. The compiler follows CircleCI's own compile rules.

Runners

Run jobs anywhere.

Runners poll over outbound HTTPS, so they work behind NAT. Route jobs with resource_class: checks on Spot instances, deploys on an on-demand host.

Tests

Split the suite.

parallelism: 4 and circleci tests run give each runner a share of the files, balanced by size.

Logs

Watch it live.

Step output streams to the dashboard as it runs, with secrets masked before they leave the runner.

GitHub

Report to the pull request.

A GitHub App starts pipelines on push and pull request, clones private repos with short-lived tokens and posts commit statuses.

How it works

From push to a green check. Three steps.

No inbound ports on your runners and no hosted control plane.

01
Push
The GitHub App gets the webhook and reads .circleci/config.yml at that commit. Without the App, start a run from any script with forge-ci trigger.
02
Compile
The config is validated and compiled into workflows and jobs. A config with errors fails fast, with the error on the commit.
03
Run
Runners claim queued jobs, run each step in Docker and stream logs back. A job starts as soon as the jobs it requires pass.

Compatibility

What works today. What doesn't yet.

An unsupported feature fails with a clear error instead of running differently.

Supported

Docker executors and service containers, executors, commands, parameters
Workflows, requires, matrix, branch filters, contexts
Caches, workspaces, artifacts, when conditions, timeouts
parallelism with circleci tests glob, split and run
Bash steps, $BASH_ENV and CIRCLE_* variables

Not yet

Orbs, because there is no orb registry yet
machine, macos and windows executors
Approval jobs, dynamic config and scheduled pipelines
Splitting by timings and setup_remote_docker
Docker layer caching and SSH re-runs

Quickstart

Running in five minutes. One compose file.

Postgres, the control plane, the worker and a runner.

$ git clone https://github.com/azmi2409/forge-ci && cd forge-ci
$ cp .env.example .env          # S3 bucket and credentials, secrets
$ openssl rand -hex 32          # paste into .env as RUNNER_TOKEN
$ docker compose up --build -d
$ docker compose up -d runner
# then open http://localhost:3000 and sign in

Create the S3 bucket first with aws s3 mb s3://<bucket>. For production, with HTTPS, the GitHub App and runners on their own hosts, follow the deploy guide.

Open source

Stop paying per minute.

Point ForgeCI at the config you have and run it on hardware you already own.

Get ForgeCI on GitHub