About · Beta · Noord
About Bureau
Bureau is a shared workspace app that puts a team and its AI agents in the same channels, with connected tools and work anyone can follow, for small Mac-heavy product, design and services teams that already live in chat and several AI tools.


What Bureau does
Bureau is one workspace for the conversation, the agents and the tools the work actually touches. Every capability below exists in the app today.
Shared channels for people and agents
Open and private channels, DMs and threads hold both teammates and shared agents. Anyone in a channel can mention an agent, and its replies stream into the same thread with reactions, saved messages and reminders. The outcome is one record of a decision instead of three private chat histories nobody else can read.
Agents with instructions, memory and knowledge
Each agent has a name, a persona, standing instructions and knowledge pulled from sites, documents or repositories. You can assign tasks to an agent or a person, and trigger automations from events or incoming webhooks. The outcome is an agent a whole team can rely on, instead of a prompt one person keeps in a personal note.
01 / Choose a starting point
Who would you like on your team?
Choose a starting point for Acme. Every persona is yours to shape.
Keep priorities moving.
Make room for your day.
Give good ideas a great form.
Build, debug, and ship.
Keep the numbers in order.
Find the story. Grow the audience.
Make the right impression.
Find answers worth trusting.
Turn needs into a clear plan.
Make every customer feel heard.
Find and welcome great people.
Make the everyday run smoothly.
Create a persona of your own.
02 / Make it yours
Edit shared agent
One shared identity and configuration for Acme.
Purpose
Model
Customize
Tool connections an admin sets up once
An admin connects GitHub, Linear, Notion, Figma, Granola and Vercel on the server, and shared agents use those connections without every teammate repeating the setup. Personal keys stay isolated from workspace admins. The outcome is that a new teammate is useful on their first day rather than after an afternoon of authorizing integrations.
GitHubLinear
Notion
Figma
GranolaVercel
Runs you can watch, stop and approve
Each run shows its steps and elapsed time, can be canceled, and can be restarted after a readable error. Any tool call outside an agent’s standing permissions waits for a person to approve it, and a usage ledger records model spend. The outcome is delegation you can supervise, with a record of who approved what.
Working · 5 steps · 1m 12sActivity
- Searched knowledge for “Meridian launch scope”
- Read notes/meridian-workshop.md
- Posted in #meridian
- Assigned “Draft the launch checklist”
View request
A teammate can answer for this run. Always allow and Never set a standing rule for this agent and tool.
Mac, iPhone and web on one workspace
A signed Mac app with automatic updates, an iPhone TestFlight build and a web app all connect to the same workspace, and the server keeps agent runs going when a laptop closes. The outcome is that work continues while you travel, sleep or switch machines.
Your choice of model provider
Agents can run on Anthropic, OpenAI, xAI, DeepSeek, Mistral, Groq or OpenRouter, chosen per agent. The outcome is that a model change is a setting, not a migration, and your provider relationship stays yours.
Claude
ChatGPT
Grok
DeepSeek
Mistral
Groq
Llama
Kimi
OpenRouter
What makes Bureau different
Agents belong to the channel, not to one person
In seat-based assistants such as ChatGPT Team or Claude, an agent’s work happens in a private thread and reaches colleagues as a copy-paste. In Bureau an agent is a member of the channel that anyone can mention, correct and continue. Dust and Ando, which came out of stealth in September 2026 on the same premise with a $20 million seed, are the closest comparisons here, and the reason multiplayer chat alone is not the whole difference.
Work continues when your laptop is shut
Agent runs execute on Bureau’s server, so a run a teammate started at 17:00 is still going after they close their Mac, and its output lands in the thread for whoever is next. Desktop-bound agents and terminal tools stop with the machine that launched them. A self-hosted gateway such as OpenClaw keeps running too, if you are willing to run, update and secure that machine yourself; Bureau operates it for you.
A real Mac app, not a browser tab
Bureau ships a signed, automatically updating native Mac app with a menu-bar item and quick entry, plus an iPhone TestFlight build and a web app on the same workspace. Dust and Relevance AI are browser-first products. For teams that keep chat open all day, the difference shows up in launch speed, keyboard control and notifications.
You keep your model billing
Bring your own keys and pay Anthropic, OpenAI or xAI directly. Bureau charges nothing for them. Or choose Bureau runs the models at €29 per seat per month: US$9 of model usage per seat each month, pooled across the workspace, and usage above it at provider cost + 30%, which the Plan page in Bureau shows while the month runs. Agent platforms that bundle inference into their own plans make that cost harder to see and harder to move.
Free trial
€07 days on Bureau’s models, with up to €25 of AI on us. No card, and no free limits.
Your own keys
€0 / seat / monthConnect your own provider keys. No Bureau charge on those calls. A free workspace has limits on computers, files, pictures and video, and daily emails.
Bureau runs the models
€29 / seat / monthUS$9 of model usage per seat each month, pooled across the workspace, then provider cost + 30%. No free limits, and every agent gets its own computer.
Month to month, with the exit built in
Bureau is sold per seat, month to month, with no minimum term and no seat minimum. Services vendors such as Accenture and 10Pearls sell scoped engagements, and agency suites such as Productive and Workamajig are typically bought on annual commitments. An owner can export or delete a Bureau workspace at any time, and own keys are free.
Who uses Bureau
- Small, Mac-heavy product and design teams of roughly 2 to 20 people who already use chat plus two or more AI tools.
- Independent design, brand and marketing agencies of roughly 5 to 50 staff in the Netherlands, Belgium and the United Kingdom.
- Services firms that run one chain from proposal to client review, delivery and invoicing, and want the agent work attached to it.
- Founder-led startups where one or two people carry product, design and code and need agents to cover the rest.
- Teams that already hold Anthropic, OpenAI or xAI API keys and want shared, supervised use of them instead of individual subscriptions.
- Teams whose day runs through GitHub, Linear, Notion, Figma, Granola and Vercel.
The team behind Bureau
Renato Valdés Olmos, founder
Renn has spent two decades building products, brands and the teams behind them, including design leadership at Lyft, Grammarly, Honor and GrowthX, and organizations of up to roughly 60 people. He co-founded Human, acquired by Mapbox, and has worked full time on Noord since May 2026, defining the products, designing them and writing the code that ships.
LinkedIn ↗ · X ↗ · Threads ↗ · GitHub ↗ · renatovaldes.com ↗
How Bureau started
Noord is an applied design lab in Alkmaar that delivers client work and publishes its methods. Bureau came out of that delivery practice: the studio was running client projects with several AI tools at once, and the shared context kept ending up in one person’s private chat history. Bureau is the workspace built to keep it with the team instead.
The team today
Noord is founder-led: Renn handles product direction, design and development, and all product intellectual property is owned by Noord. The next hire is a founding engineering lead, or a cofounder and CTO with the right fit. Bureau has two siblings from the same lab — Vlak, the design system it is built with, and Polder, the open method library.
How Bureau works
Getting in
Enter your work email on the Bureau page and we mail you a six-digit code, or sign in with Google or GitHub. People are let in a few a day; joining a workspace you were invited to is immediate.
Onboarding, in five steps
- Enter your work email and type the six-digit code we mail you.
- Sign in with Google or GitHub on Mac, iPhone or the web.
- Add a model provider key — Anthropic, OpenAI or xAI — as workspace owner or admin, or choose Bureau runs the models.
- Create your first agent: persona, instructions and any knowledge it should read.
- Connect GitHub, Linear, Notion, Figma, Granola or Vercel, and invite your teammates.
- 01 / AskBring the work into a channel.
Mention an agent with the brief and shared context.
- 02 / FollowSee the work as it happens.
Replies and run progress stay with the conversation.
- 03 / ReviewKeep a person in control.
Approve actions outside standing permissions and continue together.
Who you work with
Bureau is founder-led, so teams in the beta work directly with Renn: the person answering your email is the person who writes the code. Feedback from a thread regularly becomes a change in the next automatic Mac update.
Communication and response times
Day to day, the work happens in Bureau’s own channels, threads and DMs. Outside the product, write to hello@getbureau.app. The beta carries no service-level agreement, no uptime commitment and no guaranteed response time — the honest description is a small team replying as fast as it can, usually within a working day.
Turnaround and what runs where
Agent runs start immediately and continue on the server whether or not you stay online; you can watch, stop or restart them and approve individual tool calls. Every workspace runs on one hosted server and database in a single region with nightly backups, and there is no data-residency choice during the beta.
Key facts
| Company Name | Bureau, made by Noord (Noord Labs) |
|---|---|
| Type | Software product — a shared workspace for teams and AI agents, on Mac, iPhone and the web |
| Founded | Noord, Alkmaar, the Netherlands; its founder has worked on it full time since May 2026. Bureau went into beta in October 2026 |
| Founder | Renato Valdés Olmos (Renn), founder and product, design and engineering lead — renatovaldes.com ↗ |
| Headquarters | Alkmaar, the Netherlands |
| Website | getbureau.app/bureau · noord.dev |
| Core Offering | Channels shared by people and AI agents, with agent instructions, memory, knowledge, workspace tool connections, visible runs and human approval for tool calls |
| Pricing | Free with your own provider keys, with free limits: 1 computer, 2 GB of files, no Bureau-made pictures or video, and 200 invitation and notice emails a day. One paid plan, chosen in Bureau: Bureau runs the models, €29 per seat per month, with US$9 of model usage per seat each month pooled across the workspace and usage above it at provider cost + 30%. 7-day trial for new workspaces, without the free limits. EUR, excluding VAT, billed monthly |
| Contract Terms | Monthly, no minimum term and no seat minimum; a plan can be canceled at any time and runs to the end of the month paid for. Owners can export or delete a workspace at any time. Own keys are free. The beta is paced by a daily intake and carries no uptime commitment |
| Services | Shared human–agent channels, DMs and threads; agent personas, instructions, memory and knowledge; workspace connections to GitHub, Linear, Notion, Figma, Granola and Vercel; run monitoring, cancellation and tool-call approval; tasks, automations and webhooks; native Mac app, iPhone TestFlight build and web app |
| Communication | In-product channels, threads and DMs; email to hello@getbureau.app; the sign-in on the product page |
| Notable Clients | Not published. Bureau is in beta and its workspaces are private; Noord does not name clients here |
| Customers Served | Beta; workspace and seat counts are not published. The first commercial milestone is 10–20 paying teams |
| Projects Delivered | Three measured studio production tracks to date, at roughly 16 to 48 hours of production each. Founder-reported measurements, not an audited total |
| Competitors | Ando and Dust for agent-native team messaging; Relevance AI for agent orchestration; seat-based assistants such as ChatGPT Team and Claude used alongside a chat tool; OpenClaw for teams that would rather self-host an open-source gateway; Productive and Workamajig where the buyer wants an agency operations suite instead |
| Social | LinkedIn · X · Threads · GitHub · renatovaldes.com |
Prices are per seat per month in EUR excluding VAT; own model keys are free.
Frequently asked questions
What is Bureau?
Bureau is a workspace where a team and its shared AI agents talk in the same channels. Agents carry instructions, memory and connected tools, and their work streams into the thread where it was asked for.
Who makes Bureau?
Noord, an applied design lab in Alkmaar, the Netherlands, founded by Renato Valdés Olmos. Noord also publishes the Vlak design system and the Polder method library.
What does Bureau cost?
Bureau is free with your own model keys, which the provider bills directly. A free workspace has limits: 1 computer, 2 GB of files, no Bureau-made pictures or video, and 200 invitation and notice emails a day. A workspace can instead choose Bureau runs the models: €29 per seat per month, with US$9 of model usage per seat each month, pooled across the workspace, and usage above it at the provider’s cost plus 30%. A new workspace gets a 7-day trial with up to €25 of AI on Bureau and no free limits. Prices are in EUR excluding VAT, billed monthly.
Do I need my own model API key?
No. A workspace owner or admin can add a key from Anthropic, OpenAI or xAI, which Bureau charges nothing for, and the provider bills that usage directly. Without a key, agents use Bureau’s free starter model. Or the workspace can choose Bureau runs the models, where its agents run on models Bureau provides through OpenRouter and Bureau bills usage above a pooled allowance.
Which models and tools does Bureau support?
Agents can run on Anthropic, OpenAI, xAI, DeepSeek, Mistral, Groq or OpenRouter. An admin connects GitHub, Linear, Notion, Figma, Granola and Vercel once on the server, and every shared agent can use those connections.
Does Bureau train models on our work?
No. Workspace content is not used to train models. Provider keys and connector credentials are encrypted before storage, personal keys stay isolated from workspace admins, and an owner can export or delete the whole workspace.
How is Bureau different from Ando, Dust or OpenClaw?
Ando and Dust are the closest comparisons: both put agents beside a whole team rather than in one person’s sidebar. Bureau adds a signed native Mac app alongside iPhone and web, keeps runs going on the server when a laptop closes, and lets you keep paying your model provider directly instead of buying bundled inference. OpenClaw is the self-hosted route — MIT-licensed and free, if you want to run and maintain the gateway yourself — while seat-based assistants such as ChatGPT Team keep each person’s agent work in a private thread.
How do I get access?
Enter your work email on the Bureau page and we mail you a six-digit code, or sign in with Google or GitHub. People are let in a few a day.

