Coding
Screenshot gate
It opens what it just built in the browser and compares it against the task before anything is allowed to close, so a task cannot be marked done on a page that never rendered.
Use your own Claude/Codex sub
The code editor for AGI.
Codus makes every person on your team exponentially more productive across most knowledge work, by giving each of them a crew of autonomous agents that run in parallel on their own Mac.
Download for Mac
Tour the disciplinesBy downloading you agree to our Terms and Privacy Policy. Codus runs commands on your machine, so use at your own risk.
CodingSEOVideoResearchSupport
£99 / seat / month · Modules £99 each · Unlimited usage
codus · 4 quadrants · 2 running
cmd+shift+p
Q1 · api · payments-svc
Add idempotency keys to /charge
Claude Coderunning
Q2 · web · checkout
Migrate cart to 5-col grid
Claude Coderunning
Q3 · infra · supabase
Backfill user_folders.label
Codexdone
Q4 · docs · marketing
Rewrite README with v0.1 features
Codexproposed
desktop · macsynced with cloud planner
Case studyFrom an empty repository
From an empty repository
to a live CRM in seven hours.
A client asked for property management software. The brief went in during the evening, a fleet of agents built through the night, and by morning Acorn was live: the app, the marketing site, the demo films the agents recorded of themselves using it, the domain and the DNS. The film runs three minutes and shows what was waiting when the founder woke up.
Recorded the morning after the run · 3:10
7h 3m
First commit to live in production
~40 min
Human input across the whole build
34
Database tables
15
Functional modules
Cursor, Windsurf and Claude Code run the agent. Codus runs the workflow.
The AI is the same one everybody else is using, and it runs on the Claude Code or Codex account you already pay for, so nothing here charges you twice for it. One agent in one window still leaves you deciding what happens next, stitching the pieces together and checking the result yourself, and that is where the day actually goes.
In parallel
A quadrant is one agent working on its own: its own window, its own copy of the project, its own place to run things. Four fit in a tab, and you can open as many tabs as your Mac will run.
Routed
The Brain is the manager sitting above all of them. It keeps track of what has to happen before what and gives the next job to whoever is free, so nothing sits waiting in a queue.
Verified
Screenshots checked, claims cited and bugs reproduced by a second agent before any of it reaches you.
How the checking works186
agent tools
59
bundled skills
20
live integrations
71
releases in two months
One app, five jobsOne operator,
One operator,
five jobs' worth of output.
Coding ships with every seat, and the quadrants, browser, planner and Brain that run it also run the disciplines below. Each one is a module you switch on per seat, so the person who shipped a feature before lunch can run a full SEO audit, cut the launch video and commission a data study after it.
Ships with every seat
Autonomous engineers, working in parallel.
Four quadrants run their own agent session on your Mac while the Brain routes the roadmap between them. Each one picks up a task, makes the change, runs your tests and hands it back for you to approve, and it checks its own screenshot and fixes what it got wrong before calling anything done.
- Four quadrants per tab, as many tabs as your Mac can run
- The Brain hands out the work and keeps the order straight
- Screenshots verified before a task is allowed to close
brain · roadmap routingdispatching
Q1 · api · payments-svc
queuedAdd idempotency keys to /charge
Q2 · web · checkout
queuedMigrate cart to a 5-col grid
Q3 · infra · supabase
queuedBackfill user_folders.label
Q4 · docs · handbook
queuedDocument the webhook retry policy
AccuracyNo agent marks
No agent marks
its own homework.
A single agent left to grade itself is confidently wrong often enough to cost you a client, so codus splits the work and makes verification somebody else's job: a second agent has to reproduce, measure or cite the thing before it ever reaches you. That cross-checking is what turns fast AI output into work you can put your name on.
One claim, four gates
Every number earns its place.
Watch one finding move through the checks it has to survive before an operator sees it, and the same discipline applies whether the output is a study, an audit, a patch or a finished video.
verification · one claimchecking
claim drafted
pendingOrganic clicks fell 38% after the March update
sample size
pendingBase checked across the full 1,204 queries
source
pendingCited to Search Console, your property, your dates
red team
pendingSecond agent argued seasonality, claim survived
Research
Fact-checker seat
A dedicated agent gates every claim in a study, checking what each number was measured against, how big the sample was and where it came from before a writer is allowed to quote it.
SEO
Red team pass
Findings are measured on the live Google results, cited back to your own connected data, then attacked by a second agent looking for the reading that would embarrass you in a client meeting.
Video
Watched like a viewer
An independent Quality Control agent watches the finished render the way a viewer would and gates every version against the client-feedback rules, so the agent that cut the video is never the one that signs it off.
Set how hard the checking bites on the rigor dials, anywhere from a quick sanity pass to nothing ships until the tests and the security review come back green.
UnattendedBrief it at midnight,
Brief it at midnight,
wake up to four live products.
You give it the idea in a sentence. A research crew works out what that market already has and where the gap is, coding agents write the entire product from the database to the landing page, and the same run then registers the domains, puts it live on Vercel with every setting and password already filled in, points the web address at it and sets up the billing. Every other agent tool stops at the code and leaves the rest to you, which is why so little of it ever reaches the internet, whereas what is waiting for you in the morning here is not a plan or a half-built draft but four finished products already serving real visitors.
One night, nobody watching
Seven hours you were asleep for.
The run below is the whole distance from a sentence to a real web address you can open, with no approval step in the middle waiting on you to come back to your desk.
unattended run · 4 productsrunning
23:40
queuedbrief
One line accepted, four products queued
23:58
queuedresearch
Competitors pulled apart, pricing and gaps mapped
00:12
queuednaming
Availability checked, domains registered at GoDaddy
01:35
queuedbuild
App, sign-in, billing and landing page written and reviewed
03:02
queueddeploy
Pushed live to Vercel, every setting and password filled in
04:18
queueddns
Web addresses pointed, security sorted, sites answering
05:50
queuedmonetise
Stripe products created, ad units and analytics wired in
It works out what to build
A research crew takes the idea apart before a line is written, reading the products already serving that market, what they charge, what their reviewers complain about and which corner of it nobody has covered, then comes back with the version worth building rather than the one you first described.
It writes the whole product
Not an empty starter: the app itself, the sign-in, the database, the billing, the landing page and the words that sell it, built by the same coding agents and put through the same review as any other codus work before it is allowed to deploy.
It buys the domain
Availability and price are checked across the registrars you have connected, then the name is registered and paid for on your account without waking you up to confirm it.
It deploys to Vercel
The project is created, the site is pushed live, and every setting and password it needs is filled in, so what is waiting in the morning is a working site rather than code somebody still has to put online.
It wires the money in
Stripe products, the payment page and the plumbing that tells your app when someone has paid, for the SaaS builds, ad units and analytics for the content builds, configured as part of the run rather than left as a to-do.
It runs on a schedule
Set it once and the run fires every morning on its own, so the question stops being whether you had time to build this week and becomes which of the four you want to keep.
Connect the accounts it spends and publishes on once in integrations, set the nightly brief, and the only decision left in the morning is which of them is worth keeping.
IntegrationsConnect it once, and
Connect it once, and
every agent can use it.
Twenty integrations connect in Settings and immediately become tools every agent can call, whether it is auditing a site, reconciling an invoice or putting a change live. Your keys stay on your machine, so an agent gets the capability and never the secret.
Code + hosting
- GitHub
- Vercel
- Cloudflare
- GoDaddy
- Supabase
SEO + search data
- Ahrefs
- Semrush
- DataForSEO
- Search Intelligence AI
- Serper
AI models + media
- Anthropic
- OpenAI
- OpenRouter
- Replicate
- ElevenLabs
Business
- Stripe
- Xero
- Meta Ads
- WordPress
- Slack
GitLab, Netlify, Linear, Jira, Notion, Sentry, Figma and six more are queued next. See the full catalogue
Your own tools
Your internal systems, wired in the same way.
Point codus at your own internal systems and they show up inside agents that are already working, so the warehouse system, the CRM or the billing service becomes something an agent can use halfway through a job, with nothing to restart.
- Your own systems appear without a restart
- Per-project scoping, so a tool only appears where it belongs
- The same tools, whichever kind of work you are doing
How it worksFrom one brief
From one brief
to finished work.
Whether the job is a shipped feature, an SEO audit or a cut video, the loop between asking for it and getting it back looks the same. A brief lands, the Brain routes it to whoever is free, agents work in parallel on your own Mac, and only the calls that genuinely need your judgment reach you.
- Step 01
The work lands in the planner
Sketch a roadmap on the canvas, brief a campaign, or queue a task from your phone. Codus breaks it into pieces small enough to hand out, each one knowing who owns it and what it has to wait for, across every kind of work the plan touches.
plan · q3-launch · 9 nodes → 21 tasks queued · 11 coding · 6 seo · 4 video
- Step 02
The Brain routes it to a free quadrant
Each task goes to a quadrant that can take it, with the right project already open in front of it, so a feature, an audit and a video are all in flight at the same time rather than queued behind one busy agent.
[Q1] api · payments-svc → Add idempotency keys [Q2] seo · money-cluster → Audit the pillar page [Q3] video · onboarding → Film and cut the walkthrough
- Step 03
Agents work in parallel, with real tools
Every quadrant drives its own editor, terminal and embedded browser, and reaches the accounts you have connected, so an agent runs your tests, checks where you actually rank on Google, or films the product itself rather than guessing at any of it.
[Q1] tests run · screenshot ok [Q2] live serp read · 20 results [Q3] flow filmed · 2m14s
- Step 04
You review, ship, repeat
Work reaches you already checked by an agent that did not do it: the change with its tests, the audit with its citations, the render watched back like a viewer. You stay in command, and the busywork is gone.
Pricing£99 per seat.
£99 per seat.
Agency packs to 10x your team.
The agents run on the Claude Code or Codex account you already pay for. Usage is unlimited and unmetered, and the cloud planner is included. Agency packs get cheaper per seat as the team grows. Larger organisations settle Enterprise pricing with us directly, which is where SSO and the audit log come in.
Solo
£99per seat / month
Everything in the desktop app, plus the cloud planner. 1–4 seats.
- Codus desktop app, all features included
- Four quadrants per tab · as many tabs as your Mac can run
- Runs on the Claude Code or Codex account you already pay for
- Unlimited usage. Codus never meters you
- Cloud planner that hands out work and keeps the order straight
- Plug in your own internal systems and tools
- Built-in browser that checks your site on a phone
- Email support, direct from the team
Agency
from £50per seat
Packs of 5, 10 or 20 seats: £345, £600 or £1,000 / month. The more seats, the less per seat.
- Everything in Solo, for the whole team
- Team management: seats, roles and central billing
- Real-time team sync. Push tasks to any teammate
- Work goes to whoever is best at it
- Two agents never edit the same file
- Shared cloud planner across the team
- Priority email support
Enterprise
Custom
For large teams whose needs don't fit the plans above. We'll build the right one with you.
- Everything in Agency, at scale
- Volume pricing beyond 20 seats
- SSO, and staff accounts set up centrally
- Audit log and security review
- Priority support with an SLA
- Onboarding and a named success manager
Compare everything that's included
Discipline modules£99 / seat / month each · on top of any plan
First month freeVideo and SEO are free for your first month.
A seat can put a real study or a real edit through before any of it costs anything, then continues at £99 a month unless you switch the module back off. Coding ships with every seat; the other disciplines are modules you toggle per seat from codus cloud, billed only for the days used, so you only pay for the people who use them.
All Subscription fees are non-refundable. Cancellation prevents the next renewal but does not refund any portion of the period you have already paid for, including unused months on an annual plan. See Terms § 8 for the full no-refund policy. Your statutory rights as a consumer under the law of your country of residence are not affected.
You set the direction.
Codus ships the work.
Codus is out now on macOS, and it runs on the Claude or Codex subscription you already pay for.
Download for Mac
By downloading you agree to our Terms and Privacy Policy. Codus runs commands on your machine, so use at your own risk.