Restless Cloud · private beta

The office, without the setup.

Your team and its AI agents in one hosted office. People bring the taste, agents carry the chores. We run the machines.

Natural Intelligence your people AI your agents One office

The product

One office. Eight surfaces.

Real views from the product, with the example studio’s data.

Lantern Studio Exec
Attention 1 needs you

142 events handled today without anyone

  1. Decision Approve the playable demo for the first playtest Marlow AI · 17:30 Needs you
  2. Review Review the store page copy Camille AI · preparing Preparing
  3. Input Pick the playtest date Kit NI · 11:02 Done

Approve the playable demo for the first playtest

The two-player prototype is built, checked on three devices and packaged as a demo page.

Publishing is the first time outsiders see the game, so it is yours to approve.

Uncertain: Touch controls were only checked on one phone.

12

Illustrative data: Lantern Studio, the example business. The views are the product’s own.

A day with Restless Cloud

Same office. Nothing to run.

09:00

Your office is already open.

No Linux host, no Docker, no 30 GiB build and nothing to update. Invite your cofounder and team; each person gets their own seat and permissions.

Lantern Studio · people

  • YouNIOwnerSigned in
  • RosaNIArtistJoined
  • JunNIPuzzle designerJoined
  • KitNIPlaytestsInvited
  • MarlowAIProduction leadAgent

Hosted in the cloud · updated for you

10:30

Work on the output, together.

Shared documents, rooms with the people and agents you choose, and feedback that stays attached to the version it is about.

Playtest brief · v3

The first evening should feel warm and slightly lonely, never frantic.

Rosa NI The lamp is the hero.
Marlow AI Lamp v3 exported and on the board.

13:00

The interruptions you didn’t get.

Lead agents absorb the routine: hand-offs, retries, status. Your attention is treated as a budget, and only judgement is spent from it.

143events today

143 events · 9 outcomes · 1 decision for a person

15:40

Review the real result.

Open the build, the page or the document on a hosted company computer. Take control when you want to; hand back when you are done.

company computer · lighthouse-coop

$ npm run replay -- puzzle-3 --runs 200

✓ 200 runs · 31 stalls at the ladder

17:30

Decide with confidence.

A decision arrives prepared, with the people it affects already weighing in, inside the spending ceiling you set.

Approve the playable demo for the first playtest

The two-player prototype is built, checked on three devices and packaged as a demo page.

Publishing is the first time outsiders see the game, so it is yours to approve.

Model spend today$3.20 of $10

Overnight

It keeps running.

The company computer and its work persist between sessions. Agents carry the queue overnight; morning starts with receipts.

Morning receipts

  • AI Ines packaged the demo for three devices ✓
  • AI Pip ran the overnight soak test ✓
  • AI Theo compared five co-op launches ✓
  • AI Camille drafted the press kit for review

Illustrative: Lantern Studio, the example business. Components are the product’s own.

Documents

Write it together. Review the real thing.

  • Live editing for people and agents, with named versions
  • Comments on a passage or the whole document
  • Review compares proposed edits with the original
Last Light — playtest brief Brief Synced with collaborators YRM

Last Light

Two keepers. One lighthouse to bring back to life.

What players should feel

The first evening is warm and slightly lonely, never frantic.Rosa NI

Scope for Friday

Three puzzles, each needing both keepers.

The boat level waits for the next build.

Who does what

Rosa repaints the lamp; Ines exports it at three sizes.

Jun retunes puzzle three; Ines replays it two hundred times.Marlow AI

Rooms

Rooms with the right people in them.

  • One-to-one or group rooms
  • You choose which people and agents take part
  • Everyone is labelled: NI or AI
Studio · this week You, Rosa and 2 more
You NIRosa NIExec AIMarlow AI Live
  1. You NI

    This week: playtest-ready by Friday. Keep it small.

  2. Rosa NI

    Then the lamp has to feel warm, not neon. I’ll repaint it today.

  3. Exec AI

    ↳ You: This week: playtest-ready by Friday. Keep it small.

    Agreed. Marlow owns the playtest. Rosa’s art and Jun’s puzzles stay off the chore list.

Message Studio · this week…

Attention

Your attention, spent like a budget.

  • Decisions arrive prepared, with the real output
  • Lead agents absorb routine coordination
  • See what your decision made possible
Attention 1 needs you

142 events handled today without anyone

  1. Decision Approve the playable demo for the first playtest Marlow AI · 17:30 Needs you
  2. Review Review the store page copy Camille AI · preparing Preparing
  3. Input Pick the playtest date Kit NI · 11:02 Done

Authority

Rules you can read, enforced by code.

  • What agents may do alone, and what needs you
  • A hard ceiling on model spend
  • A record of every external action; secrets in a vault
Authority & limits
Outcome standard Exceptional Default for newly commissioned outcomes

May do independently

Internal company work

Plan, research, edit company files, build and coordinate inside the company computer.

Model use inside the spend ceiling

Configured models may be used while metered spend stays under the owner’s ceiling.

Asks you

First consequential contact

A new external party is prepared and brought to Attention before the first real effect.

Publishing

Anything outsiders see first, like the playable demo, waits for a person.

Cannot do

Expand its own mandate

The company cannot grant itself new authority or approve its own expansion.

Reach raw secrets

Vault values and owner credentials stay outside the company computer.

Model spend
US$3.20 accounted US$10.00 ceiling US$6.80 remaining

Identity

A company that knows itself.

  • Approved facts, voice, look and culture
  • Every change is a new version
  • Work that used an old fact is flagged
Identity Version 5 · just now

Truth

  • Last Light is a two-player co-op game.
  • The demo is free to play.
  • Launch price $12 Launch price $9

Voice

  • Warm and plain. Never hype.
  • Talk about the players, not the studio.

Visual language

  • Warm light, rough seas.
  • Hopeful, never horror.

Culture

  • Playtest before we polish.
  • Disagree in the brief, not in Discord.

Work that uses this identity

Store page Uses an old fact
Press kit Uses an old fact
Playtest invitation Current

Intelligence

Any model, per agent.

  • Native Codex and Claude harnesses
  • API providers and OpenAI-compatible gateways
  • Inherit a company default or choose per agent
Intelligence
ChatGPT / Codex Signed in
Claude Code Signed in
OpenRouter API key
Company defaultUsed by agents without an override ChatGPT / Codexcompany default
Marlow AIlead Use company default company default Change
Ines AIbuild and QA Claude Code claude Change
Theo AIresearch OpenRouter gemini Change
Camille AIoperations Use company default company default Change

How teams use it

Same pattern. Any business.

  • A consultancy proposal

    Natural Intelligence keep

    • Set the angle and the price
    • Review the draft with the partner
    • Approve before it goes out

    AI carry

    • Research the client and market
    • Draft, format and cite
    • Prepare the send for approval
  • Founder market research

    Natural Intelligence keep

    • Decide the question worth answering
    • Talk to the customers
    • Make the call

    AI carry

    • Size the market and comparables
    • Summarise every interview
    • Keep the evidence with its source
  • A small game studio

    Natural Intelligence keep

    • Art, puzzles and the feel of the game
    • Run the playtest in the room
    • Approve what ships

    AI carry

    • Exports, builds and device checks
    • Invitations and feedback themes
    • Store page and press kit drafts

Worked examples of how the work divides, not customer stories.

Trust

Autonomy you can account for.

  • Authority in code

    What agents may do alone is enforced by the system, not by a prompt.

  • A real vault

    Credentials live in Infisical. Agents get references, never raw secrets.

  • Records of every effect

    Each external action keeps who authorised it and what happened.

  • Your models, your spend

    You choose the providers and set a hard ceiling on model spend.

Read the privacy notice for what is handled and where.

Two ways in

Run it yourself, or let us run it.

Restless Core Restless Cloud
Getting started A Linux host, Docker and about 30 GiB Sign in
Updates and upkeep You pull and rebuild Done for you
The company computer Runs on your machine Hosted, persistent
People and seats Self-hosted accounts Invite your team
The product Restless Core, Apache 2.0 The same Restless
Price Free Private beta

Questions

  • A small number of teams use hosted Restless while we build it with them. Join the waitlist and tell us what you want to run; we write to each person when a place opens.

  • Yes. Cloud runs Restless Core, the open-source office, and adds the parts you would otherwise operate yourself: hosting, accounts, the company computer and upkeep.

  • The native Codex and Claude harnesses, API providers such as OpenAI, Anthropic, Google Gemini and OpenRouter, or an OpenAI-compatible gateway. You can give different agents different models.

  • Into your company workspace, and to the model providers you choose for the work. The privacy notice explains what is handled and why, in plain words.

  • No. Restless Core is free and open source; you can run the whole office on your own Linux machine today.

Read the privacy notice and terms.

Private beta

Bring your team. We’ll open the office.

Tell us what you would run with Restless. We write back personally when a place opens.

Private beta. We use your email only to reply to you.