48 specialists.
One AI office, on your laptop.

An AI team that does the work — not just answers questions.

48 specialists · 10 departments · all yours on every plan · runs locally

When most people think "AI", they think of a chat window: you type a question, it types back, and you copy-paste the answer somewhere. Useful, but you're still doing the real work — turning that text into a finished email, a posted caption, a reviewed pull request.

Krew flips that. Instead of one general chatbot, you get a team of named specialists — Arjun the chief of staff, Zara the caption writer, Neo the code reviewer, and 40 more. You give one a task in normal language, and it goes and does it: searches the web, reads your files, drafts the email, and brings back something finished. You're reviewing, not doing.

Every section below explains one part of how this works — read whichever ones you're curious about. If you only read one, read the next.

Each one has a name. Each name means a job.

Name.Role — like Zara.Content or Neo.Engineer — so you always know who's doing what

Krew isn't one chatbot you rename. It's 48 distinct specialists, each trained for a single job and shown in chat as Name.Role. Arjun.Boss runs the show and hands tasks to the right person. Zara.Content writes your captions. Vera.Engineer reviews your code. Nyx.Data digs up companies from public databases. All 48 are included on every plan — nothing is locked behind a higher tier.

Browse the full team. Filter by department, or expand to see them all:

It doesn't guess. It reasons, then verifies.

Reason → Act → Observe → repeat — up to 10 steps before it answers

A plain chatbot answers from memory, which is why it sometimes confidently makes things up. Every Krew agent works differently: it thinks about what it needs, uses a tool to go get it, looks at what came back, and only then continues. It's the difference between a confident guess and a checked answer.

Here's that loop on a real task — writing a launch post — shown exactly as it runs inside the app:

1. Up to 10 reasoning steps per taskComplex jobs get multiple rounds of think-act-check, not one rushed reply.
2. Real tools, in the loopWeb search, file reads, and app calls happen mid-task — the agent sees real results before continuing.
3. It stops and asks if it's genuinely stuckNo quiet guessing. If it can't proceed safely, it tells you instead of inventing an answer.

Not a chatbot. An actual agent.

Four systems run on every task — the reason Krew feels different from a chat window

These aren't clever prompts you have to write. They're built into how every agent works, quietly, on every task you give it.

Your agents know your tools.

Gmail · Google Suite · Notion · Slack · GitHub · Linear · Airtable · Brave + live browser

Connect the apps you already use, and your agents can read from them and act in them in real time — draft a reply in your inbox, check a Notion page, read a GitHub issue, post a result to Slack. Not a one-time "import": a live connection, with your approval before anything is sent.

How the AI itself is powered is up to you — run it free and offline, bring your own provider key, or use one simple adris.tech plan:

Not web search. The actual web.

Opens a real Chrome window, clicks, fills forms, reads pages, takes screenshots

Sometimes an answer isn't on a search results page — it's behind a login, inside a dashboard, or on a site that only works with full JavaScript. So Krew opens a real Chrome window on your machine and uses it like a person would: navigate to a page, click a button, fill a form, copy the content, take a screenshot. Not a scraper — the actual browser.

1. Reaches what search can'tLogged-in dashboards, SaaS tools, JavaScript-heavy pages — your own accounts, read on your behalf.
2. Does, not just readsFill a lead form, monitor a pricing page, pull structured data, research a competitor.
3. Runs on your machineIt's your Chrome, your logins, your network. No cloud browser, nothing handed to a third party.

They draft. You decide.

Every outside action — send, post, create — waits for your one-tap approval

An agent that can send emails and post to Slack is powerful — and that power needs a seatbelt. So every action that touches the outside world shows up as an approval card first. You see exactly what it wants to do, in full, and nothing irreversible happens until you tap approve. You can also edit the draft before sending — it's never all-or-nothing.

Your Krew. Your device. Your data.

Conversations on your disk · credentials in your OS keychain · zero cloud by design

This is the part that's genuinely different from any online AI tool. Krew runs on your laptop. Your chats live in a database file on your machine, your passwords are encrypted in the operating system's keychain, and we never see any of it. Not because we promise not to look — because there's nothing on our side to look at.

If a few hours of your week is repeatable work, Krew is for you.

You don't need to be technical. The people who get the most from Krew are: