Codemany
Codemany turns Macs and Linux machines you own into single-use self-hosted GitHub Actions runners: change one runs-on line and every job gets a fresh runner, with local caches, optional KVM VMs and a fleet dashboard. $20/month for 2 machines, 14-day free trial.
About Codemany
What Codemany is
Codemany is a desktop app and background service that turns Macs and Linux machines you already own into self-hosted runners for GitHub Actions. You sign in with GitHub, pick the repositories or the organization it may serve, and change one line in a workflow: runs-on: [self-hosted, codemany] (on Linux, runs-on: [self-hosted, Linux, codemany]). From then on GitHub still schedules the jobs and shows the logs, but the work happens on your hardware, and each job gets a fresh, single-use runner that is registered just in time, used once and then replaced.
It is built by Arpit Agarwal, a mobile developer who previously worked at startups including Beepi and Lumi Labs and at Google on Google Maps, Google Keyboard and Android Messaging. In a post on the Codemany site he explains why: he builds an investing app, Amaltash, with Claude Code, and as coding agents pushed more commits, its GitHub Actions usage went from about 2,300 Linux minutes in its first month of CI (August 2025) to 76,000 minutes in October 2026, and the monthly bill from $2.54 in August 2025 to $553 in October 2026. Codemany's answer is a flat price for machines you own instead of a per-minute bill. The first public release (0.1.0, a signed and notarized macOS DMG) shipped on October 5, 2026; version 1.0.0 followed on November 20, and the site showed v1.2.2 when we checked on November 22, 2026.
How it works
- Single-use runners. Every job gets a just-in-time runner that takes one job and is then wiped, so leftover files from one job cannot break the next. Each machine runs as many jobs at once as it has room for; the FAQ gives 1 concurrent job on an 8 GB Mac and 3 on a 16 GB Mac.
- Host, VM and user modes. By default jobs run as your user on the host, using whatever is installed: Xcode, Homebrew packages, Docker. Nothing is preinstalled. On Linux, an opt-in
vmmode gives each job its own KVM virtual machine laid out like GitHub's runner image, with Docker and sudo, thrown away afterwards (experimental on macOS). An experimentalusermode runs each runner as a separate OS user. - Tool labels. Codemany detects installed tools and adds labels such as
xcode-26.0anddockerthat a job can ask for, which helps iOS teams pin an Xcode version rather than followmacos-latest. - Local cache and mirrors. Caches, toolchains and package mirrors stay on the machine (and can be shared over the LAN), so installs and checkouts start warm.
- Fleet view. A menu bar app on macOS, a tray app on Linux, a read-only dashboard at
http://codemany.local:8825for any device on your network, and Codemany Console (console.codemany.com) with every job's runner logs and a plain hint when a job failed because a tool was missing. - Auto-updates. New GitHub runner releases and Codemany updates install themselves, after running jobs finish.
Requirements: macOS 14 or later on Apple silicon, or 64-bit Linux with systemd (installed with one command, including a Snap package), and a GitHub account that can add self-hosted runners (a repo admin or an organization owner). Codemany says your code and GitHub token stay on your machine (the token in the macOS Keychain); its servers receive your GitHub id, login and email at trial start, a daily plan check, runner summaries and job logs with secrets masked for Console, plus anonymous usage counts you can switch off with codemany usage-stats off.
What we checked: Codemany's own benchmark data
Codemany publishes its speed claims ("Run GitHub Actions 3x faster" is the launch tagline) with the raw data in an MIT-licensed repository, codemanyhq/bench. Thirteen open-source projects are pinned to exact commits and built on both sides several times a day: on GitHub-hosted runners and on two Codemany machines, a Minisforum UM870 mini PC (AMD Ryzen 7 8745H, 8 cores, 32 GB) and a 16 GB Apple silicon Mac mini. On November 22, 2026 we downloaded data/results.json (3,020 jobs from October 31 to November 21, 2026) and computed the median build time of every successful job ourselves, without using the site's charts:
| Project (build from scratch) | Codemany | GitHub 2-core Linux (private repos, $0.006/min) | GitHub 4-core Linux (public repos, free) |
|---|---|---|---|
| ripgrep (Rust, cargo build --release) | 13 s | 46 s | 29 s |
| Hugo (Go, go install) | 31 s | 138 s | 78 s |
| react.dev (Next.js build) | 82 s | 171 s | 117 s |
| Django (test suite) | 132 s | 532 s | 324 s |
| TypeScript (npm ci + build) | 257 s | 681 s | 424 s |
| Now in Android (Gradle assemble) | 540 s | 1,228 s | 750 s |
| Excalidraw (docker build) | 70 s | 117 s | 77 s |
| macOS project (from scratch) | Codemany Mac | GitHub macOS runner (3-core M1, $0.062/min) |
|---|---|---|
| Firefox for iOS (xcodebuild build-for-testing) | 232 s | 920 s |
| SwiftLint (swift build -c release) | 216 s | 907 s |
| The Composable Architecture (xcodebuild test) | 92 s | 285 s |
Three things the raw data shows that the headline does not. First, the gap depends heavily on which GitHub runner you compare with: against the free 4-core runner GitHub gives public repositories, Codemany's from-scratch medians are roughly 1.1x to 2.5x faster, while against the 2-core private-repo runner they are about 1.7x to 4.5x faster. Second, with saved caches the advantage often disappears: Hugo and Kubernetes builds take 2 to 4 seconds on every side, the warm Excalidraw Docker build was faster on GitHub's public runner (8 s against 19.5 s), and so was the warm SwiftLint build on GitHub's Mac (270 s against 310 s). Third, reliability: 125 of 318 Codemany Mac jobs and 73 of 1,063 Codemany Linux jobs in the file did not finish with success (failed or cancelled), against 27 of 314 and 12 of 1,049 on GitHub's runners. The benchmark excludes those jobs from timings by design, and part of the Mac side ran on the maker's own MacBook Pro during development, so read the failure counts as a sign of a three-week-old benchmark on a young product rather than a measured failure rate for your workloads. The hardware is also not like for like: an 8-core, 32 GB mini PC against GitHub's 2 or 4 vCPU runners is exactly the point of the product, but it means the comparison measures your hardware as much as Codemany.
Pricing (checked November 22, 2026)
| Machines | 1 | 2 | 3 | 4 | 5 |
|---|---|---|---|---|---|
| Per month | $20 | $20 | $30 | $40 | $50 |
Codemany Pro is $20 a month including 2 machines, plus $10 a month for each extra machine, with unlimited runners on each machine and no per-minute billing. There is a 14-day free trial with no card. A plan covers all repositories of the GitHub account or organization that subscribes; personal and organization repositories need one plan each. When a trial or plan ends, new jobs stop and running jobs finish, and a plan keeps working for 7 days offline.
Worked example: when does $20 a month pay off?
GitHub-hosted price per minute (private repos)
Linux 2-core $0.006 macOS $0.062
Minutes that cost $20 on GitHub
Linux $20 / $0.006 = 3,333 minutes a month (about 111 a day)
macOS $20 / $0.062 = 323 minutes a month (about 11 a day)
Example: 20 iOS builds a day x 15 min x 30 days = 9,000 macOS minutes
GitHub: 9,000 x $0.062 = $558 a month
Codemany: $20 a month + a Mac you already own
The break-even is low on macOS, where GitHub's per-minute price is about ten times Linux's, and the case is weakest for small Linux workloads that fit in GitHub's included minutes, or for public repositories, which build free on GitHub. The Codemany price does not include the machine, its electricity or the time to look after it; the benchmark's Linux machine is a consumer mini PC, and the changelog uses a $700 baseline for it. The usage figures above are our assumptions.
Codemany vs the alternatives
| Codemany | GitHub-hosted runners | Plain self-hosted runners / Actions Runner Controller | Managed per-minute runners (e.g. Blacksmith) | |
|---|---|---|---|---|
| Where jobs run | Your Mac or Linux machines | GitHub's cloud VMs | Your machines or Kubernetes cluster | The provider's cloud |
| Billing | Flat: $20/month for 2 machines, +$10 each | Per minute for private repos; free for public | Free software; you run and patch everything | Per minute |
| Fresh environment per job | Yes (single-use runner; KVM VM on Linux opt-in) | Yes (new VM) | Only if you build it (ARC ephemeral pods) | Yes |
| macOS and Xcode | Your own Apple silicon Mac, your Xcode versions | Shared images, 3-core M1 standard runner | Possible, fully manual | Varies by provider |
| Maintenance | Auto-updates, fleet view, logs | None | All yours | None |
For the hosted side see GitHub Actions vs GitLab CI. If the build is not the bottleneck but flaky tests are, test tools such as Currents, TestSprite and heal.dev work on any runner, including Codemany's.
Who it's for
iOS and macOS developers paying GitHub's macOS minute rate, small teams and solo builders whose CI volume grew with coding agents, and anyone with a spare Apple silicon Mac or Linux box who wants GitHub Actions on it without hand-maintaining the runner service, its updates and its workspace cleanup.
Limits and trade-offs
- A self-hosted runner is not a sandbox. In the default host mode jobs run as your user with access to everything your user can reach. Codemany's own docs say to run only workflows you would run in your own terminal; public repositories are off unless you enable them per repo. Use
vmmode on Linux for real isolation. - You own the hardware and its failures. If the Mac sleeps, loses network or runs out of disk, jobs queue. Capacity is fixed by your machines, unlike GitHub's elastic runners.
- Nothing is preinstalled. Jobs see your tools, not GitHub's runner image, so workflows that assume
ubuntu-latestpackages may need setup steps (or Linuxvmmode). - Young product. First release October 5, 2026, version 1.0 on November 20, and frequent releases since; the public benchmark shows a high share of failed or cancelled Mac jobs during its first weeks.
- Some data leaves the machine. Job logs (secrets masked) and runner summaries go to Codemany Console; usage stats are on by default but can be turned off.
- Uninstalling does not cancel the plan. Cancel first, then remove the app; the FAQ lists the Keychain items and folders to delete for a full clean-up.
Related on TechLogHub
- CI/CD and Docker explained.
- The Docker cheat sheet and Git cheat sheet.
- GitHub Actions vs GitLab CI.
- Render and Railway for deploying what CI builds.
- The cron expression generator for scheduled workflows.
Value & Audience
Value Proposition
GitHub Actions on hardware you already own, with fresh runners per job and no per-minute bill, for a flat $20 a month for two machines.
Problem Solved
GitHub-hosted CI minutes, especially macOS minutes, get expensive as commit volume grows with coding agents, while hand-run self-hosted runners need setup, updates and workspace cleanup.
Tech Stack
Social & Links
Loading ratings...
Loading comments...
Gallery

Feedback & Roadmap
Codemany FAQ
Common questions about Codemany's features, pricing, and use cases
Is Codemany free to use?
No. Codemany is a paid product, starting at Codemany Pro - $20/month - 2 machines, unlimited runners per machine, no per-minute billing, 14-day free trial without a card. Plans listed: Codemany Pro - $20/month - 2 machines, unlimited runners per machine, no per-minute billing, 14-day free trial without a card; Extra machines - $10/month each (3 machines $30, 4 $40, 5 $50).
What are some alternatives to Codemany?
Popular Cloud & DevOps alternatives to Codemany include Deadhand, Sentry and Ohuriya AI.
Codemany appears in
You May Also Like
Deadhand
A unified monitoring tool for cron jobs, website uptime, and SSL/domain expiry with flat pricing.
View ProductSentry
Application performance monitoring and error tracking for developers.
View ProductOhuriya AI
The safest way for non-DevOps founders to manage a server using plain English commands with mandatory human approval for every step.
View ProductVercel
Deploy and scale web applications instantly with AI-powered infrastructure.
View ProductOne email a week
TechLogHub tracks 923 tools and 617 open-source projects. Get the week's new launches, release roundups and open-source picks — nothing else.
Get the TechLogHub digest
One email a week: new and trending developer tools, fresh comparisons, and what shipped. Unsubscribe in one click.