Cetus

Cetus

Human in the loop is the bottleneck.

Agents can work for an hour on their own. Then they wait for you. Cetus is one desktop client for reviewing and managing the work of every agent you run.

DownloadGitHub
acme-web1 needs you1 working1 done
Today
The checkout e2e test fails about 1 in 5 runs on CI. Find out why and fix it.
Fix flaky checkout test2:41 PM

Root cause was a race: the test clicked Pay before the cart refetch settled. It now waits for the cart query. 50 runs, 0 failures.

Done613·1 reply
Add tables for invoices and payment methods, then migrate the dev database.
Add billing tables3:05 PM
Codex wants to run pnpm db:migrate
Needs you43
Move the settings page to server components. Keep the form interactive.
Settings page to server components3:12 PM

Settings now renders on the server; only ProfileForm and ThemeToggle stay client components. Saves go through a server action.

Needs review1273·3 replies
Check for outdated dependencies and open a PR for safe minor upgrades.
Nightly dependency check3:20 PM

Running pnpm outdated…

Working21

Start a new thread — @ picks the runtime.

acme-webCetus
Fix flaky checkout testacme-web
The checkout e2e test fails about 1 in 5 runs on CI. Find out why and fix it.
Grepcheckout e2e/
Reade2e/checkout.spec.ts
Readapp/checkout/cart-summary.tsx
Bashpnpm playwright test checkout --repeat-each 20

Reproduced: 4 of 20 runs fail with element is disabled. Pay is disabled while the cart refetches, and the test clicks before that finishes.

Edite2e/checkout.spec.ts
Bashpnpm playwright test checkout --repeat-each 50

Root cause was a race: the test clicked Pay before the cart refetch settled, so the button was briefly disabled. The test now waits for the cart query to finish. 50 local runs, 0 failures.

Message the agent. Press / to focus.

acme-webClaude CodeOpus · High

Artifacts

launch-next-steps.md

Markdown · 338 B

Launch next steps

Decisions from the launch planning meeting.

Next steps
  • Website — finish the product walkthrough.
  • Feedback — group early user notes by workflow.
  • Release — review the checklist before publishing.
Ready for review

Share the updated walkthrough with the team and collect final feedback.

Interactive preview · Example content

The thesis:
Agents aren’t the slow part.
Waiting on you is.

Claude Code, Codex, and DeepSeek Harness can work for an hour on their own. What slows them down is the person they are waiting on: a permission prompt, a clarifying question, a finished task nobody has reviewed yet.

Cetus starts from that. Every part of it is there to make your side of the loop shorter.

Several agents.
Several clients to check.

Most people who use agents seriously run more than one: Claude Code in one window, Codex in another, scheduled jobs that each report back in a different client. Every switch costs attention, and every minute you don’t notice an agent waiting is a minute it sits idle.

Cetus gives all of them one client, with one place to approve, answer, and review what they do.

Your accounts.
Your tokens. Your choice.

You already have a relationship with your model providers: API keys, CLI logins, preferred models, and a token budget you decide how to spend. That should be the starting point.

Cetus runs the CLIs you already have installed and signed in to, and reuses their logins, models, and tools. There is no second account to set up.

Terminals are for agents.
Not for reading.

Some tools, like Superset, put different runtimes side by side, but in a TUI. A terminal is a fine place for an agent to run and a poor place for a person to read: everything is monospaced text, plans, diffs, and questions all look alike, and whatever scrolls off is gone.

Cetus is a real desktop app, so every agent’s work is easy to read, easy to answer, and easy to review.

Notice

Threads and the board show which sessions are working, which need you, and which are done, across every runtime.

Respond

Approvals, questions, and plans arrive as cards. Answer in place, steer a running turn, or summon any runtime with a hotkey.

Explain

Screen context, meeting transcripts, selected text, and memory come along, so you type less background.

Step away

Automations run on a schedule and land on the board for review. Your agents keep working while you are away.

The models keep getting faster.
Your attention doesn’t.

So Cetus spends it carefully: notice sooner, answer faster, explain less, and step away.

Any runtime, one hotkey away.

Hold both Command keys to summon an agent over any app. Ask when the thought arrives, without hunting for the right window.

Your screen, selected text, and browser URL can come along as context. Keep what helps. Remove what doesn’t.

⌘+⌘

Less waiting. Less explaining.

Answer agents from one place, hand them the context they need, and leave work for later.

Never keep an agent waiting.

Approvals and questions from every runtime, in one place.

Say what’s on your mind.

On-device dictation, wherever you type.

Pick up the context.

Opt-in screen history, searchable by text and app.

Remember the conversation.

Turn meetings into searchable, on-device transcripts.

Leave a task for later.

One-off or recurring work, on your schedule.

Carry decisions forward.

Memory you and your agent both edit.

Website polish
Polish the website and review the details.
Updated 3 files

The layout and interactions are ready for review.

website/index.html
website/styles.css
Finished
Ask a follow-up…
Codex

Room for every thread.

In Threads, every message starts its own session on whichever runtime you @mention. Open one beside the feed, and follow the badges to the next session that needs you.

Choose a workspace.

Keep each conversation grounded in the right project.

Keep the conversation.

Carry context and background terminals across replies.

Review what’s ready.

See which tasks are running, need review, or are done.

Explore another approach.

Run parallel solutions and keep the one that works.

A shorter way into your flow.

Reach your assistant, show it your screen, or speak a thought — from the app you’re already in.

Quick Launcher

Hold both Command keys to summon your agent.

⌘ + ⌘

Quick Launcher + screenshot

Hold both Option keys. Select an area, or click to capture the full screen.

⌥ + ⌥

Dictation

Hold right Command and talk. Text lands where your cursor is.

Right ⌘

Meeting notes

Start an on-device transcript for an in-person meeting.

⌘ ⇧ M

Keep the agents you already trust.

Your preferred runtime, models, and tools. A consistent desktop home around them.

Make room for Cetus.

Open source. Built for your Mac. Ready for your next idea.

Download Cetus macOS 13+ Cetus Swift Beta · macOS 15+

Both builds run on Apple Silicon and update themselves. Your chosen model provider’s usage charges may apply.