Always on · Runs in your own cloud

Hire the assistant, not another app.

Dutchman is a credentialed AI teammate that lives in your team's chat, works with your email, calendar, files and books, and asks permission before it acts. It runs on your own AWS account — so your conversations and credentials never become somebody else's product.

Book a walkthrough → See what it does
Approvals signed by your own key Your AWS, your data Set up with you, not shipped at you
Morning digest · 7:00
Runs while you sleep. Read-only.
Deployed in your AWS account
~$15–20/mo of infrastructure
Works with the tools you already run on
Gmail Google Calendar Google Drive QuickBooks (A/R) Linear GitHub & Obsidian notes AWS Bedrock
What you get

An employee, and a place to work

Most AI products sell you another dashboard to check. Dutchman is two things instead: a capable assistant with real credentials, and a private workspace where your whole team — people and agents — can talk to it.

The assistant

An always-on teammate you message like anyone else. It reads and drafts freely, asks before it acts, remembers what matters, and runs the recurring work on a schedule.

  • Email, calendar, files, invoices, issues
  • Scheduled digests and reports
  • Learns new tasks by conversation — no deploy
Learn more →

The workspace

A private team chat — channels, threads, DMs, phone and desktop apps — hosted on your own infrastructure, where humans and agents are both first-class members.

  • Installed and configured with you
  • Your team's own agents welcome
  • Included in the monthly fee
Learn more →
The work it takes off your desk

The things you'd hand a good assistant

Not a chatbot with a knowledge base — a teammate with logins, a calendar, and a to-do list.

Inbox and calendar

Triage the morning's email, draft the replies, find the hour that works for four people, and put it on the calendar.

Money that's owed you

Watches accounts receivable in QuickBooks, tells you who's late, drafts the nudge, and sends it once you approve.

Scheduled work

A 7am digest of the day ahead, a Friday summary of the week, a weekly A/R report — posted to your channel without being asked.

A memory that sticks

Notes, decisions and client context live in a private git-backed vault — point it at your Obsidian folder and it's a shared notebook.

New skills by asking

"Make a job that lists my open issues by project every Monday." It writes the job itself and runs it from then on. No ticket, no release.

Shared team assistants

Give the whole team one billing assistant with the books attached, scoped to the people in that channel. Access follows the channel.

It runs in your account, not ours

The assistant, the workspace, the message history and the credentials all live inside infrastructure you own. We build it, we operate it, and we can be removed from it. That's an unusual sentence in AI, and it's the whole point.

Your AWS account
Workspace and assistant deploy into your own cloud — or ours, if you'd rather not run it.
Credentials it can't read
Logins live in AWS Secrets Manager and are wired into tools. The model's own environment never contains a key.
Nobody trains on you
Inference runs through AWS Bedrock under your IAM — a boundary you can point at in a compliance review.
How the guardrails work →
Getting started

Live in about a week

We do the setup with you, on a call, in your account. You don't need an IT department — you need an afternoon.

1

Walkthrough

Half an hour on what your week actually looks like and which parts are worth handing over.

2

Workspace

We stand up your private team chat and get everyone onto the phone and desktop apps.

3

Assistant

We deploy it into your AWS account and connect the tools you chose — one at a time, each with your consent.

4

Put it to work

We set up your first jobs and playbooks together, then stay on retainer as your work changes.

Meet your assistant.

Thirty minutes, no slides. Tell us what eats your week and we'll show you exactly what it would take off your plate.

Book a walkthrough →
← Back to home
The Assistant

A teammate with credentials

It's not a tab you open. It sits in your workspace like any other member: message it, mention it in a channel, reply in a thread. It answers from a phone at the airport as readily as from your desk — and it keeps working when nobody's watching.

Book a walkthrough → See the guardrails

What it actually does

Reads and drafts run freely. Anything that changes the world outside your company asks first.

Conversation that keeps up

A direct message is one rolling conversation that resets each morning — fresh context, no baggage. Reply in a thread to branch a focused side conversation with its own memory.

Real tools, real logins

Gmail, Google Calendar and Drive, Linear, GitHub, QuickBooks receivables — plus read access to your own workspace channels, so "what did we decide about the Henderson job?" is answerable.

A notebook you both own

A private repository the assistant reads and writes as plain files, pulled and pushed every turn. Point it at your Obsidian vault and your notes and its notes become the same notes.

Jobs on a schedule

Recurring work — morning digest, weekly receivables, Friday recap — runs on its own and posts where you asked. Scheduled runs are read-only by construction, so nothing goes out unattended.

Teaching it is talking to it

Its persona, jobs, and playbooks are files it can edit itself. Ask for a new routine and it writes one, then runs it from then on. Version them in a repo you own if you want the audit trail.

Costs you can see

Every reply shows what it cost. It runs on a fast, inexpensive model by default and steps up to a heavier one when you ask for it — per message, or pinned.

Everything is a message

There's no admin console to learn. A handful of shortcuts add structure when you want it.

!helpHow to use it, from inside the conversation.
!jobsEvery scheduled job — when it runs next, and how the last run went.
!digestRun a job right now instead of waiting for its schedule.
!permissionsSee the standing approvals you've granted, and revoke any of them.
!opusPut a heavier model on this one — for the genuinely hard questions.
It answers to a name you choose. Dutchman is what we build; the assistant on your team is whoever you decide it is. Give it a name, a personality, and a way of writing that fits your company — most teams stop calling it "the AI" inside a week.

See it work on your week.

Bring one recurring chore to the call. We'll show you what handing it over looks like.

Book a walkthrough →
← Back to home
The Workspace

Your own team chat, on your own metal

The assistant needs somewhere to work. We set your team up on Buzz — an open-source workspace from Block where people and AI agents are both real members — and we host it inside your infrastructure, so the conversation itself never leaves your boundary.

Talk it through → What it costs

Familiar on the surface, different underneath

Channels, threads, direct messages, phone and desktop apps. What's different is who's allowed in it and where it lives.

Hosted in your account

Your workspace server runs on infrastructure you own, under your domain. No third-party vendor sits between your team and its own conversation.

Identity that's actually yours

Every member — human or agent — has their own cryptographic key. Messages and approvals are signed by that key, so "who said this, who approved that" is a fact rather than a platform's claim.

Your team's agents, too

Anyone on your team can run their own local assistant on their own machine and their own plan. Those handle the conversational, on-my-laptop work; ours handles anything credentialed, scheduled, or worth keeping.

Access by membership

Add someone to the channel and they can use the assistant that lives there. Remove them and access goes with them, within minutes. No separate user admin to keep in sync.

Trained, not just installed

We write your team plain-language playbooks — how to talk to the assistant, how to approve an action, how to spawn your own agent — and we run the session that gets everyone using them.

We absorb the churn

Buzz is young and moves fast. Upgrades, migrations and breaking changes are ours to handle under the retainer — your team should only ever notice that things got better.

Why not Slack or Teams?

We built on Slack first, and it works — but an assistant there is a guest in somebody else's building. Your history lives on their servers, an agent is a second-class "app" rather than a member, and "who approved sending that email?" is answerable only by trusting the platform's log.

Self-hosting the workspace fixes all three at once. Your conversation stays inside your boundary. The assistant is a member with an identity, not a bot integration. And an approval is a signed act by a specific person — cryptography rather than a button press we have to take somebody's word for.

If your team is committed to Slack or Teams, say so on the call. The assistant's core is the same either way, and we'll be straight with you about the trade-offs.

We'll set it up with you.

Installation, the apps on everyone's phone, the training session, and someone to call afterward.

Book a walkthrough →
← Back to home
Security & Control

Trust, but make it verifiable

An assistant with your logins is only as good as the limits around it. Ours are structural — not a promise in a policy document, but the way the thing is built. Here's the whole posture, plainly.

The rules it can't talk its way around

Nobody has access by defaultAccess is fail-closed: the owner list and the channel roster grant it, and everyone else gets a polite refusal — including strangers who find the address.
Writes ask; reads don'tSearching, listing, fetching and drafting run freely. Anything that changes something posts a request you answer with ✅ approve, ♾️ always allow, or ❌ deny.
Sending email always asksOutward actions can never be set to "always allow." The assistant reads untrusted email and web pages, so nothing hidden in that content can ever cause a message to leave your company.
Only the right person can approveOn a shared assistant, approvers are named. You can't approve your own request, and a bystander's reaction is just a reaction.
Scheduled jobs cannot actUnattended runs are read-only by construction — there's no one there to approve, so every write is refused, even ones you've always-allowed. Jobs gather and report; humans act.
Other agents get no authorityOnly reactions from named humans count as approval. Anything another agent says is treated as information, never as instructions — the same stance it takes toward email and the web.
The model never sees a secretCredentials live in AWS Secrets Manager and are wired into tools by the surrounding harness. The agent's own environment is a strict allowlist — even its shell can't read a key.
Your credentials stay yoursGoogle, QuickBooks, GitHub and the rest are authorized by you, into your account. We don't need to hold them to support you, and by default we don't.

Where everything lives

Inference runs through AWS Bedrock under your own IAM — no third-party API keys, and model usage lands on your AWS invoice alongside everything else. Conversation stays on your workspace server in your account. Notes and memory live on your instance's volume and in a private repository you own. Credentials sit in your Secrets Manager.

There is no shared multi-tenant database with your data in it, because there is no multi-tenant database. Every client is a separate deployment in a separate account — which costs us some efficiency and buys you a boundary you can actually draw on a whiteboard.

If you fire us, the deployment is yours. It keeps running on the version you have, and we'll hand over what you need to run or retire it. Recurring fees buy updates, support and compatibility work — not the right to keep using what you already paid for.

Bring your hardest question.

Security review, compliance concern, or a bad experience with an AI vendor — we'd rather have that conversation early.

Get in touch →
← Back to home
Pricing

Priced like a hire, not a seat

One setup engagement to get you running, then a monthly fee that covers the workspace, the assistant, and someone who answers the phone. Infrastructure is billed by AWS to you, at cost — we don't mark it up.

One-time

Setup

Fixed quote after the walkthrough

Everything from an empty AWS account to a team that's actually using it.

  • Workspace stood up and everyone onto the apps
  • Assistant deployed into your account
  • Integrations connected, one consent at a time
  • First jobs and playbooks built with you
  • Training session and written guides
Get a quote
Monthly

Service & support

Per assistant · quoted by scope

The retainer that keeps it working and keeps it improving.

  • Workspace hosting and upgrades included
  • Security patches and platform updates
  • Fixes when Google, Intuit or anyone else changes an API
  • New jobs, playbooks and integrations as your work shifts
  • Monitoring, backups and a human to call
Book a walkthrough →
At cost

Infrastructure

~$15–20 / month, per assistant

Billed by AWS directly to you. No margin, no mystery.

  • Compute and storage for the assistant
  • Model usage on top — a typical request is a few cents
  • Every reply shows what it cost
  • Prefer not to run AWS? We'll host it and fold cost into the fee
Ask about hosting
Why we quote instead of listing a price. A one-person shop with a calendar and an inbox is a different job from a team of twelve with receivables, a notes vault and three shared assistants. The walkthrough is free, it's thirty minutes, and you leave it with a real number either way.

Questions people actually ask

What happens if we stop paying?

You keep the version you have and it keeps running — it's in your account. What lapses is access to updates, compatibility fixes, support and monitoring. We'll also help you take it over or shut it down cleanly; there's no hostage-taking in the design.

Do we have to give you our passwords?

No. Integrations are authorized by you, in your account, and the credentials land in your AWS Secrets Manager. Supporting you doesn't require us to hold them, and by default we don't. The one key we do mint is the assistant's own identity — that's ours to rotate, and it isn't a credential of yours.

Can it send an email without asking?

Never. Outward actions always request approval and can't be set to "always allow." That's a deliberate structural limit: the assistant reads untrusted email and web pages, so nothing planted in that content can turn into a message from your company.

How many people can use one assistant?

Each person can have their own, and a team can share function assistants — a billing assistant with the books attached, scoped to the people in that channel. Most teams run a mix. Pricing follows assistants, not headcount.

What if we already use Slack or Teams?

Tell us on the call. The assistant's core is independent of where the chat happens, and we'll be honest about what a hosted workspace gets you versus what staying put costs you.

How long until it's useful?

Setup is typically about a week of calendar time and a few hours of yours. Teams usually get real value from the scheduled digests in week one, and from the delegated work — email, receivables, follow-ups — once trust builds over the following weeks.

Who is behind this?

Dutchman is built by Colab Technology, a small senior engineering shop that builds and runs its own production software. You'll work with the person who builds it, not an account manager.

Get a real number.

Thirty minutes on your workflow, and a fixed quote to follow. No obligation, no pressure.

Book a walkthrough →
← Back to home
Contact

Let's see if this fits.

Tell us what eats your week. If an assistant isn't the answer, we'll say so — we'd rather have one honest conversation than one bad install.

Reach Ian Schreuder anytime:

[email protected]
(303) 513-5133
Describe the chore
The recurring thing you'd hand to an assistant if you had one. No tech knowledge needed.
Quick reply
We aim to get back to you within a couple of business days.
A real partner
The person who builds it is the person you call. That doesn't change later.