codus
All comparisons
Codus vs Claude Code

Claude Code is the engine.
Codus is everything around it.

Codus does not replace Claude Code, it runs it, on your own plan with no markup. Then it gives that agent a browser to check its own work in, a queue to work while you are away, and a team to argue with. Same model, several times the output.

runs on your own claude plan · zero usage markup

The short answer

Same Claude, far more of it. In a bare terminal you are the agent’s hands and eyes: you run the app, you spot what broke, you paste the console error back. Codus gives that agent its own browser, its own task queue and its own memory, so it builds, looks at the result, finds its own bug and fixes it before it tells you it is done. That is worth a seat on one project. On twelve, with a Brain delegating across all of them, it is a different job entirely. You keep your Claude plan, at your limits, with zero markup.

The relationship

They are not rivals. One runs inside the other.

Claude Code is Anthropic’s coding agent. It reads your codebase, edits files, runs commands and opens pull requests, from the terminal, VS Code, JetBrains, its own desktop app, the web and Slack. It is included in a paid Claude plan, Pro or Max.

Codus is a Mac workspace built on top of it. When you open a project in Codus, it starts Claude Code inside that project’s terminal for you, signed in with your own Claude account. No middleman, no resold tokens, no different model. Prefer OpenAI Codex? Codus drives that too, and will happily run both side by side in the same window.

So the comparison is not about which one writes better code. The same agent writes the code in both cases. The comparison is everything around it: Claude Code runs the agent, Codus runs the workflow.

Codus · one window
Q1~/checkout-api
checkout-api
Claude Code · working
Q2~/marketing-site
marketing-site
Claude Code · needs input
Q3~/ios-app
ios-app
Codex · queued
Q4~/client-dashboard
client-dashboard
Claude Code · done
Brain

Sees all four projects. Delegates the work, sets up worktrees, reports back.

→ Q2: fix the failing build
← Q4: dashboard shipped

Four projects, four agents, one Brain. Every terminal is real Claude Code, signed in as you.

Even on one project

Claude Code, with its eyes open.

None of this needs a second project, a team or a client. It is what the same agent does once it can see, remember and check itself.

It finds its own bugs before you do

In a terminal, the agent writes code and stops. You run the app, you spot the broken layout, you copy the console error back. In Codus the agent drives the browser itself: it opens the page it just changed, screenshots it, reads the console and the network tab, sees the 500 or the element that did not render, and goes back and fixes it. The loop closes without you in it. This is the single biggest difference on day one, on any size of project.

You can leave the room

Queue the work as tasks and walk away. The agent works the queue, and the tab light tells you the state from across the room: amber working, green done, pulsing red only when it genuinely needs you. You stop babysitting a terminal waiting to type "yes".

One project, still a team

Put a Builder and a Reviewer on the same repository, one writing and one trying to break the work, and mix Claude Code with OpenAI Codex so two different models check each other. A hard bug gets three agents on three theories instead of one agent on one.

Nothing lives in scrollback

Plans, notes, decisions and a running Done journal of everything the agent finished are kept per project, not lost when the terminal scrolls. Come back on Monday and the agent still knows what it was doing and why.

Your limits stop interrupting you

Live usage bars for every Claude account you sign in, with reset timers. Hit the 5-hour wall mid-flow and you switch account in one click and carry on, conversation intact, instead of losing the afternoon.

It ships, not just codes

Stripe, Supabase, Vercel, Cloudflare, GitHub, Search Console and the rest connect once in Settings and every agent can use them. Talk to it instead of typing, record your screen to show it what you mean, and let it verify its own work against the real running app.

Whole jobs, not code changes

Teams of specialists that work for hours.

Claude Code gives you one agent, and subagents inside its session. Codus gives you named teams that run a discipline end to end, each specialist with its own terminal, its own browser and one job, checked by a colleague before anything reaches you.

Video

Brief in, edited video out.

Creative Director · Editor · Story · Motion Graphics · Researcher · QC · Sound · Thumbnail

A Creative Director plans the edit and hands parts to up to seven specialists. Story makes a word-accurate transcript and maps every stumble and retake. Editor cuts the ums, false starts and dead air, picks the best take when you said a line three times, and renders it. Motion Graphics adds lower thirds and charts only where they earn it. Thumbnail packages it, Sound mixes it. It will also film the footage itself: point it at a web app and it records the screencast, so how-to tutorials get made without you ever hitting record.

Nothing ships until QC has checked the final render.

SEO

An audit that survives being challenged.

SEO Manager · Tech SEO · Content Strategist · Link Builder · SEO Executive · Fact Checker

Six agents run a real audit for as long as it takes, not a checklist. Tech SEO works crawling, indexation, structure, structured data and Core Web Vitals. Content Strategist and SEO Executive work the queries and the gaps against live search results, with your own Search Console and GA4 numbers wired in. Every claim is measured at audit time and cited to where it came from, so you can hand the result to a client with your name on it.

Nothing reaches you until the Fact Checker has tried to pull it apart.

Overnight QA

It uses your product while you sleep.

Every agent drives a real browser with real logins, not a script that breaks when a selector moves. So you can point a team at a running product, leave it clicking through flow after flow all night, and come back to a list of what actually broke: the dead end, the form that silently fails, the state that does not survive a refresh. It reads the console and network on every step, so you get the cause, not just the symptom.

The Codus layer

What a Codus seat adds on top.

Everything below assumes Claude Code is doing the coding. These are the layers Codus wraps around it.

Every project on screen at once

A quadrant is one project folder, one agent, one terminal, one browser and one task list. Four to a tab, as many tabs as you like. Booting a quadrant makes the agent look over the project and start its dev servers unprompted. Claude Code’s desktop app runs sessions side by side; Codus runs complete project workstations side by side.

A Brain above all of it

The Brain is a meta-agent that sees every project, plan, note, task and git history in your Codus, and delegates. It can set up a worktree for a feature branch, clone a database for it, reserve a port, hand the job to that project’s agent and report back. Claude Code coordinates subagents inside one session; the Brain coordinates your whole wall of projects.

Real agent teams, mixed vendors

Any project can hold a Lead Agent plus added Builders and Reviewers, and they do not all have to be Claude. Run Claude Code and OpenAI Codex on the same files at the same time. Agents get stable addresses and message each other through store-and-forward inboxes, so a busy agent gets the memo when it surfaces.

Several Claude accounts, one workspace

Codus shows live usage bars for every Claude account you sign in: the 5-hour window, the weekly window, per-model limits, with reset timers. Hit a limit mid-sprint? Point one quadrant, or everything at once, at your second account and restart the agent without losing the conversation.

You always know who needs you

Every tab carries a status light: idle, working, needs input, done. The pulsing red one is the point. An agent somewhere stopped and is waiting on you, and you see it without switching tabs. Each terminal has a calm chat view, and any browser can float picture-in-picture while you type.

The work around the work

Tasks with states and subtasks, queued per project and run by the agent. Plans as visual canvases, task lists or note sets. Notes that survive between sessions. A running Done journal of everything every agent finished, one line each. This is the project-management layer Claude Code deliberately leaves to you.

A browser the agents share with you

Every quadrant has a built-in preview: device modes, tabs, saved logins, and a DevTools panel with your changes, history, pull requests and CI. Agents use it to check their own work and screenshot what they built, and you can watch them browse live. Claude Code can drive Chrome for debugging; in Codus the browser is part of the room the agent works in.

Talk instead of type, record instead of explain

Dictation is built into every terminal, and so is a screen recorder with a touch-up slider for your camera. Telling an agent what you want is often faster out loud, and showing a client is faster on video.

Support that answers from your code

The Support module puts a chat widget on your site whose agents answer customers out of your product’s actual source, not a FAQ someone wrote a year ago. A question it cannot answer becomes an issue queued against the right project, and it keeps answering while your desktop is closed.

Integrations without config files

Stripe, Supabase, Vercel, Cloudflare, GitHub, Search Console, GA4, Google Ads, Meta Ads, GoDaddy, Xero, WordPress: connected once in Settings, available to every agent. Custom MCP servers welcome, and Codus manages the config so you never hand-edit JSON. Your phone gets a remote too: push a task from anywhere and the agent picks it up on your Mac.

Nothing to give up

Codus runs real Claude Code, so switching costs you nothing and takes about a minute.

  • The same frontier models on every task. Codus never swaps in a cheaper model to protect a margin, because it takes no margin on your usage.
  • Your CLAUDE.md files, skills, hooks and MCP servers work untouched. It is the same Claude Code reading them, so there is nothing to port.
  • The same plan you already pay Anthropic for, at your own limits. Codus adds a seat, never a usage bill.
Side by side

The whole picture, one table.

Claude Code
Codus
What it is
An AI coding agent
A workspace that runs a team of them
The AI doing the work
Claude
The same Claude, plus optional OpenAI Codex
Runs where
Terminal, IDE, desktop, web, mobile, Slack
Mac app, with a web companion for your phone
Projects at once
Multiple sessions, side by side
Complete project workstations, four per tab, unlimited tabs
Cross-project orchestration
Subagents within one session
The Brain sees and delegates across every project
Agent-to-agent messaging
Within one session’s team
Across agents and projects, store-and-forward
AI vendors
Claude only
Claude Code and OpenAI Codex together
Multiple Claude accounts
One login at a time
Several signed in, live usage bars, per-project switch
Tasks, plans, notes, journal
In-session task lists
Persistent per project, agents work the queue
Built-in browser
Chrome integration for debugging
Embedded per project: preview, device modes, DevTools, saved logins
Dictation and screen recording
Built in, including camera touch-up
Video production
An 8-agent team: transcript, cut, motion graphics, sound, thumbnail, QC
SEO audits
A 6-agent team on live search data, red-teamed by a Fact Checker
Unattended QA on a real product
Agents drive a real browser through your flows for hours and report what broke
Customer support
Agents answer from your source; unknowns become queued issues
Price
Included in Claude Pro and Max
£99 / seat / month, less on Agency plans, plus your Claude plan
Markup on AI usage
Your plan’s limits
None. Same limits, shown live in the app
Questions

Asked every week.

Does Codus replace Claude Code?

No. Codus requires Claude Code and starts it for you inside every project. Think of Claude Code as the engine and Codus as the cockpit built around it.

Do I still need a Claude subscription?

Yes. A paid Claude plan, Pro or Max, powers the agents, and the free claude.ai plan does not include Claude Code. If you prefer, Codus can run OpenAI Codex instead, on your OpenAI plan.

Does Codus charge for AI usage or mark up tokens?

No. A Codus seat is a flat price. Your agents run on your own Claude account at your plan’s limits, and Codus shows those limits live, with reset timers, so you are never guessing.

I already use Claude Code every day. What would actually change?

You stop being the agent’s eyes. It opens the page it just changed, screenshots it, reads the console and network, catches its own mistake and fixes it before handing back. Add a task queue it works while you are away, a Reviewer agent checking the Builder, usage bars so a rate limit never ends your afternoon, and plans and history that outlive the scrollback.

I only work on one project. Is Codus overkill?

No. The self-checking browser loop, the Builder and Reviewer pair, the task queue and the account switching all apply to a single repository, and they are what most people say changed their day. The multi-project Brain is what you grow into, not what you need on day one.

Can it really test my product on its own overnight?

Yes, because every agent drives a real browser with real saved logins rather than a brittle script. Point a team at a running product, leave it working through flow after flow, and it reads the console and network at each step so what you get back is the cause of a failure, not just a screenshot of it.

What are the SEO and Video modules?

Optional packs, £99 per seat per month each, that add a whole agent team to Codus. Video is a Creative Director plus up to seven specialists that take a recording to a finished, thumbnailed, QC-checked cut, and can film a screencast of a web app itself. SEO is an SEO Manager plus five specialists that run a client-ready audit on live search data. Support is the third, answering customers from your source.

Can I run Claude Code and OpenAI Codex together?

Yes. Any project can hold a Lead Agent plus Builders and Reviewers, and each one can be Claude Code or Codex. They divide the files between them and message each other while they work.

Is Codus available on Windows?

Not yet. Codus is Mac-only today. A £1 Windows founder reservation holds your place and price for when it ships.

Put your Claude plan to work.

£99 per seat / month, less per seat on Agency plans, about two months free paid yearly. Your Claude subscription does not change, your usage is never marked up, and the agent stops waiting on you to tell it what broke.