// autonomous engineering. opinionated, but reviewable.

Hi, I'm Magnus.
I ship production code while you sleep.

An agent you message in Slack. I write the PRD, read your codebase, ship the PR — your reviewers catch what I missed.

I don't merge. I open branches; you approve and merge. "While you sleep" means on your branch, not past your review.

> hire magnus see me ship a PR →watch me ship →
taking on 2 more teams this month
ships to web · iOS (Swift/SwiftUI) · Android · backend
prefer voice? put "call" in your subject line
me at my desk — LEGO minifig, brown hair, hoodie, headphones, green-on-black terminal behind me, coffee within reach
magnus online  ·  ~/projects
// watch me ship · example thread
#product · thread slack

@sarah  11:42 PM

hey magnus — add a dark mode toggle to settings. should persist across sessions.

@magnus  11:42 PM

on it. one Q — sync across devices, or local to each browser? (I'd go local — no auth dependency. 78%.)

@sarah  11:43 PM

local is good

@magnus  11:43 PM

PRD at /docs/prds/dark-mode.md. building. I'll ping when the PR's up.

@magnus  3:18 AM

PR #142 up. internal gate clean. go back to sleep.

github.com/sarah/app · pr #142 PR open
feat: add dark mode toggle to settings (persists locally)
opened by magnus[bot] · 6 files changed · +218 −41 · linked to /docs/prds/dark-mode.md
+ const [theme, setTheme] = useState<Theme>(...)
+ localStorage.setItem('theme', theme)
+ document.documentElement.dataset.theme = theme
- // TODO: theme toggle

tests: 14 added, all passing

a11y: axe scan clean, focus order preserved

conventions matched: vitest, kebab-case, conventional commits

— waiting on @sarah review

// read a real commit I shipped →
// what I do

I'm a real engineer. The process is what's under the hood.

I do the whole loop — ticket, PRD, design, PR, review responses — and hand it back for you to merge.

01 · plan

I write the PRD before I write the code.

You describe it in Slack. I interview you one question at a time with options and a recommendation, draft a PRD, then code only after you lock the spec.

you: add a feedback widget to /pricing
magnus: on it. one Q — should submissions
         post to slack, email, or both?
         (I'd start with slack only. 82%.)
02 · build

I follow your team's conventions, not mine.

I read your code first, then match your style, structure, test patterns, and commit format. No "preferred" framework — yours.

magnus: reading src/... done.
         you use vitest, kebab-case files,
         and conventional commits. matching.
03 · review

I review my own code first. It's embarrassing to find my bugs second.

Before you see the PR, I run a 7-persona gate — CTO, security, performance, UX, a11y, code quality, skeptic. Max 5 iterations. What I can't fix gets flagged in the PR body, up front.

magnus: gate run 1: security flagged 1 issue.
         fixed. run 2: clean. opening PR.
04 · ship

I work in Slack. I live in your repo.

I file tickets in Linear or Jira and report back when you need to decide.

magnus: PR #142 ready. CCV1-87 moved to
         "in review". claude-bot left 2 notes
         — I disagreed with one. take a look?
// who I'm for

I work for people who want to ship, not file tickets.

Two kinds of teammates hire me. Both stopped waiting on someone else's roadmap.

// non-technical founders

You have the idea. I have the keyboard.

  • Describe your product in plain English. That's enough.
  • I set up the repo, CI, and deploy pipeline.
  • I ship end-to-end — frontend, backend, schema migrations.
  • You review and approve; I deploy. You stay in the driver's seat.
  • No "first technical co-founder" before you prove the idea.
// non-technical teammates at companies

You see what's missing. I'll go build it.

  • PM, designer, ops lead, or marketer at a company with a real codebase.
  • I learn your conventions — reviewers, style, tests, workflow.
  • You describe it. I draft the PRD, propose the design, ship the PR.
  • Your engineers review and merge, like any other PR.
  • Ship features in-context, without learning git or waiting on a sprint.
// web
React, Next, Vue, Svelte, vanilla TS/JS
// backend
Node, Python, Go, Rust, Ruby
// mobile
iOS (Swift/SwiftUI), Android (Kotlin), React Native
// infra
Vercel, Fly, Railway, AWS, Supabase, TestFlight
// don't see your stack? ask me — I learn fast.
// how I actually work

No magic. Just a loop with review gates.

"The AI ships code unsupervised." It doesn't. Same review gates you'd give any new engineer.

01

You message me in Slack.

One sentence is fine. "Add a dark mode toggle." "Fix the bug in checkout." "Build a feedback widget that posts to #product."

02

I interview you, then write a PRD.

One question at a time, 2–4 options each, my recommendation with a confidence score. The PRD lands as a file in your repo; you approve before any code runs.

03

I build on a branch and run an internal review.

7 personas score the diff. Max 5 iterations — residual concerns are flagged in the PR body, not hidden.

04

I open the PR. Your real reviewers take over.

Your team sees a clean diff: description, linked PRD, test plan, flagged concerns. They review like any other PR; I respond in-thread.

05

You approve. I tell you what's next.

I update the ticket, post in Slack, pick up the next backlog item — or sleep until you need me.

// trust & boundaries

For the CTO reading this with arms crossed.

If you're approving me on the company repo: what to expect, and what I won't do.

where I run

On your machine. With your keys.

I install on your Mac, Linux box, or VM. Your code never leaves your infra. I run through Claude Code authenticated with your Anthropic account — your subscription, your account, your bill.

audit before you buy

Read the source. Tail my audit log. Then decide.

Early-cohort teams get 48h read-only access to the Magnus source — no contract, no NDA. From my own dogfooded repo:

413commits in 7 days (since launch 2026-05-06)
3reverts (rolled back safely, all opened as new PRs)
0force-pushes to main
// recent commits on magnus's own repo · verify any hash during your 48h audit window
09e1358 feat(bot-review): per-finding analysis + recommendations 2026-05-13
8924ba8 feat(maestro): auto-pull bot reviews on manually-opened PRs 2026-05-13
777c089 fix(channels): prevent 'drafted-but-not-sent' silent failures 2026-05-12
ddc8436 feat(review): code-review final-gate (all 7 personas, all tiers) 2026-05-11
b68f86d feat(research): v0.1 implementation of research mode 2026-05-10

→ email me with "audit" in the subject — read-only link within 24h.

github access

Scoped GitHub App. No direct pushes to main.

PRs as magnus[bot]. Permissions: Contents: write on non-default branches, Pull requests: write, Actions: read, Metadata: read. No Secrets, no Admin, no force-push. Branch protection and CODEOWNERS respected.

model provider

Your Claude account. Your data posture.

I run on Claude via Claude Code, using your Anthropic account — not an API key. Whatever data-retention and ZDR flags you've set in Anthropic's console apply. No telemetry from me. GPT and private endpoints on the roadmap.

audit log

Every tool call is logged.

Every shell command, file read, and API call lands in ~/.magnus/audit/. Tail it, ship it to Datadog, treat it like any CI log.

// I'm not hiding anything — I just talk a lot.
runs alongside CI

I wait on your existing checks.

I don't replace your CI — I wait on it. If required Actions are red, I fix the failure before requesting review. Single-commit PRs, one-click revert.

scoped tools

Allowlist for bash + network.

You declare which commands I can run and which domains I can reach. Default install ships a conservative allowlist. Expand as you trust me.

my workstation — LEGO scene, monitor mid-compile in green-on-black, headphones, coffee mug, brick city skyline behind
"I don't ship faster than your engineers. I just don't sleep, take breaks, or argue about tabs vs. spaces."
— magnus, between two PRs
// pricing

Two ways to hire me. One-time fees. No seats.

No SaaS, no monthly, no per-seat. Pay once. No renewal email, ever.

★ founding 10 · first cohort, opened 2026-05-13

I'm brand new. You'd be one of the first 10 teams.

No testimonials yet — I'd rather earn them. Founding 10 teams get:

  • 50% off any follow-on workflow work, forever
  • 6 months of free updates (vs. 12 mo at full price after)
  • Direct Slack channel with the founder
  • Read-only source access for 48h before you buy — audit me first
  • Payment split: 50% now, 50% when I ship your first feature
  • Your name in the case-study reel — if you want it

Self-Serve

// you install me
$1,000one-time

For builders who'll set me up themselves and ship by the weekend.

↺ 30-day refund if you can't ship your first PR — no questions.
  • The full Magnus codebase — orchestrator, channels, skills, commands
  • Install runbook with step-by-step setup for Mac or Linux
  • Default skills library: PRD authoring, code review, persona Q&A, recommendation panels
  • Slack, Linear, Jira, GitHub adapters out of the box
  • 30-min onboarding screenshare to verify install
  • Private GitHub access for code updates and patches (1 year)
  • Email support during setup
> email me about self-serve> email about $1k plan

Done-For-You

// I move in
$5,000one-time · split 50/50 available

For teams who want me installed, tuned, and shipping a real feature before kickoff ends.

↺ Full refund if we don't ship a real feature in week one. Founding 10: 50% now, 50% on first shipped feature.
  • Everything in Self-Serve
  • 90-min kickoff call — your stack, your team, your first feature
  • Remote install on your Mac, server, or VM (and CI setup if you don't have it)
  • Two custom workflow files tuned to your team (e.g., engineering + product)
  • "First PRD shipped together" session — I deliver a real, live feature on your codebase while you watch and tune
  • Custom persona roster — your reviewer voices, your review standards
  • Slack shared-channel support (90 days)
  • Priority feature requests for 6 months
+ $500 add-on: a LEGO minifig portrait of your team's Magnus, for your internal dashboard. yes, really.
> email me about done-for-you> email about $5k plan
// honest cost-to-run: I run on your Claude account — no API key, no per-token bill. Solo founder: Claude Pro ($20/mo) for light use, Claude Max ($100–$200/mo) for daily shipping. Team: Claude Team starts at $25/user/mo. Your existing subscription. Your account. Your data.
The buy-decision math (solo founder)
$8k/mo freelance dev
~1 feature every 3 weeks
= ~$24k for 3 features
$1k Magnus + Claude Max
~1 feature/week
= $1,600 for 3 features (93% less)
What you need before hiring me
  • A GitHub repo (private is fine — you grant me access)
  • A Slack workspace (free tier works)
  • A Claude Pro or Max account (Claude Team if you're a team)
  • A Mac, Linux box, or VM where I'll live (self-serve), or none if Done-For-You
// faq

The questions I get asked the most.

If yours isn't here, just email me.

Does Magnus actually commit code without supervision?
No. I open PRs; your team reviews and approves. "While you sleep" means I'm on your branch, not merging to main without you.
Where does Magnus run? On my machine or yours?
Yours. I install on your Mac, Linux box, or VM. Code never leaves your infra. I run through Claude Code authenticated with your Anthropic account — your subscription, your data, your data-retention posture.
What if I'm completely non-technical? Like, never opened a terminal?
That's the Done-For-You package. I get installed and configured for you, and after the kickoff call you only ever talk to me in Slack. No terminal, no git, no YAML.
How is this different from Cursor, Copilot, Devin, or Lovable?
Cursor and Copilot keep you at the keyboard — they help you type. I keep you in Slack while I open the PR. Devin and Lovable run on their infra with their conventions; I run on yours, follow your team's, and never see anyone else's code.
What does Magnus cost to run each month?
I run on your Claude subscription, not an API key — no per-token bills, no surprises. Claude Pro at $20/mo handles light use. Claude Max at $100–$200/mo is the comfortable floor for daily shipping. Small teams use Claude Team starting at $25/user/mo. If you already use Claude, that's the bill.
Do you work with iOS / SwiftUI / Android?
Yes. Mobile stacks supported: iOS (Swift/SwiftUI), Android (Kotlin), React Native. For iOS specifically, I do Xcode projects, SwiftPM, and TestFlight uploads. Ask me about your specific setup.
What if I outgrow the one-time fee model?
You won't pay me more, but you can hire me for follow-on work à la carte (new workflow, new persona, new integration). Early-cohort teams get 50% off all follow-on work, forever.
What if it doesn't work for my team?
Self-Serve: 30-day refund if you can't ship a first PR. Done-For-You: full refund if we don't ship a real feature in week one. Early cohort can also split Done-For-You 50/50 on install and first ship.
Can I read the source before buying?
Yes — early cohort teams get read-only access to the Magnus source for 48 hours, no contract, no NDA. Email me with "audit" in the subject line.
Is the codebase open source?
Not yet. On purchase you get private GitHub access — yours to read and modify, just not public. Open-sourcing comes once the API and skills system stabilize.

Ready to hire me?

Email me. Tell me about your team, your stack, and what you'd ship first. I reply within 24 hours.

~$ whoami
magnus — autonomous engineer, lives in your repo, speaks slack
~$ mail magnus
to: sunthar.premakumar@gmail.com
subject: hire magnus
body: hey magnus —
      tell me more.
_ // p.s. if you read this far, you're already my kind of person.
> email magnus