Every image below is a capture of Rem0te running against an isolated demo
database, all taken at v0.53.2, at 1440×1000 or, where marked, 390×844 (phone width), both at
deviceScaleFactor: 2.
Nothing is mocked, drawn, or AI-generated. The one exception is marked: the tray icon sheet is the icon files
themselves, not a screenshot.
All businesses, people, devices and activity shown are fictitious. The RustDesk IDs are synthetic and registered with no rendezvous server, so none of these devices can be reached. Enrollment tokens, RustDesk IDs and public IP addresses are replaced with sample values in the browser immediately before each capture — see Regenerating these.
↩︎ Back to the README · ▶ Watch the walkthrough (v0.53, 14 min 35 s) · highlight · Quick start · Architecture · More MSP tools at mspreboot.com
| Section | What it covers |
|---|---|
| Overview and dashboards | The dashboard, search, the customer list |
| Visual insights and monitoring | Client map, alerts, live sessions, history, audit, host health |
| Devices | The device list, bulk actions, saved views, one device, Linux and macOS |
| Device tabs | Performance, processes, services, network, patches, the latest install and scheduling one, tasks, event log, Ask AI and Create ticket, actions, chat |
| Everyday workflows | Adding a device, downloads, Quick Connect, launching from Tactical RMM |
| Reports, automation and patching | Reports, the script library, the winget catalog, client-update windows, patch policies, monthly windows, Automation → Tasks |
| Access and administration | The three-level model, capabilities, accounts |
| Configuration | Platform settings, branding, AI, PSA / ticketing, releases, documentation |
| Regenerating these | How to rebuild the gallery without touching production |
Most features appear once, in whichever theme suits them; the dashboard, device list and a device page are also shown at phone width.
What needs a person, first. A site outage is one item rather than three offline alerts; every tile is a count with what it is counted from; Work on next says why each item is there.
The same view in dark. Both themes ship with the product and are switched from the sidebar.
At 390 px the sidebar is a drawer behind the menu button, and the page scrolls instead of fitting the screen. The outage strip is capped only on the fitted desktop layout, so on a phone it shows its whole sentence.
Ctrl+K from any page: devices, businesses, sites, people and open alerts, each group only if you may open its list page.
Every customer business you manage, with its computer and people counts. This is the top of the hierarchy: each business owns its own computers, users and history.
One customer's record, including the per-business Quick Connect switch that gates temporary support access for their users.
Managed computers plotted from the address each last checked in from, clustered by city with a count. Useful for spotting a site that has gone dark.
Every open alert, deduplicated: a problem that repeats is one alert with a count, and a site outage gathers the computers it took down.
Everything online right now, grouped as connectable cards. One click launches the technician's local RustDesk client against that machine.
What was done, by whom, on which machine and for how long — the record you go back to when a customer asks what you did last Tuesday.
An append-only trail of every action: sign-ins, session launches, credential reveals, capability changes. Failures never crash the request path, so the log is what actually happened.
The same trail with the actor and resource filters, for narrowing to one person or one machine.
Host health and the services Rem0te depends on — including RustDesk's own hbbs
and hbbr — so you can tell a platform problem from a customer problem.
The platform's own security surface, in one place.
From 0.48.1 the list fits the window from 1280 to 2560 pixels with the sidebar open. Each wider column appears at its own width: Last seen from 1440 pixels, Agent from 1536, OS from 1700, Site from 1860 and Logged-in user from 2000. Alerts is off by default because Health already shows the worst open alert.
Every device you may see, across every business. The server does the searching and filtering, so it covers every page, not just the one on screen.
Tick devices for group membership, an inventory refresh, a client update, an agent update or upgrade, or an export of just those rows.
Filters, sort, columns and density saved under a name, for yourself or shared in a business. A shared view still shows each person only what they may see.
Scoped to a single customer — the view you use when they phone up.
One device: its state, what is wrong with it, and the tabs for everything else. An action that is not available says why underneath.
Reachable, but not reporting. RustDesk still sees the computer, so Connect works; the agent has not checked in for 30 minutes or more, so nothing it measures is current. The dashboard counts it apart from online and offline, and the map draws it amber.
Policy by device group: the Remote depots group raises its Offline monitor to 60 minutes because the depot VPN drops, and the tab says which scope each rule came from.
CPU, memory and system disk, sampled on the computer's heartbeat rather than read from the six-hourly inventory pass — so these are minutes old, and the card says so instead of letting a needle imply "now". Amber past 80% and red past 92%, the point at which a Windows volume starts failing updates.
What a managed computer reports about itself, collected by its agent on a heartbeat rather than fetched live — so every card says how old its contents are. Hardware, storage and network come in every six hours, pending Windows updates every twelve, and who is signed in on every beat.
Three separate things can be out of date here — the Rem0te agent, the RustDesk client, and the machine's own Windows updates — so each is named rather than merged into one "up to date" claim.
Linux and macOS computers run read-only agents: heartbeat, inventory and pending updates, and nothing the console sends.
What a Business User sees — only the machines they were granted, and a Connect button. No RustDesk IDs to type, no passwords to pass around.
Machines that enrolled but have not landed in a business yet.
Per computer (0.46): Inherit, On or Off. The card says what is actually in force; on this demo server the process switch is off, and it wins over the computer's On.
A computer that has gone quiet offers a reinstall link under its buttons and in More.
The link re-attaches to the same record: history, site, groups and settings kept, no duplicate.
Metric history: 24 hours of every sample, then hourly means with their range for 7, 30 and 90 days. Hatched stretches are gaps, where no samples arrived.
A snapshot, not a live view: it says when it was taken and who asked. From 0.45 a row's End process asks the full agent to end that process as an approved action; Windows, security, Rem0te and RustDesk processes show Protected instead.
The plan names the process id, its name and when it started; the computer refuses unless all three still match, so a process id Windows has reused is never ended. It has been tested against a simulated Windows only.
Display name, name, state, start mode and account. From 0.45 each row offers only the operations allowed for that service, each a signed approved action.
Only the state or the startup type ever changes; the program, account, password, recovery actions and dependencies never do. Tested against a simulated Windows only.
What the evidence is and how old, what is offered, and, when deploying is not possible on this computer, why.
What the last install did, update by update, in the words Windows gave (0.52). When it needs a restart the card says so and offers Schedule restart; the updates stay listed as installed, waiting for restart until the computer restarts and scans again. Dark: latest-install-dark.png.
An install can start at a date and time in a named zone, or at the next opening of the computer's own patch-policy window (0.52, 0.53). Nothing is handed to the computer before then. The form was not submitted. Dark: schedule-install-dark.png.
Everything sent to this computer and where it is: scheduled, waiting for approval, queued for the next check-in, handed out, running or done (0.52). Light: device-tasks-light.png.
Read a page of a managed computer's Windows event log without connecting to it. From 0.48 each entry has Ask AI and Create ticket; the second entry here already has a ticket. Five logs are available — Application, System, Security, Setup and Windows PowerShell — and nothing else: the allowlist is enforced by the API and again by the agent that opens the log.
There is no channel to reach into an endpoint, so this is a request rather than a query: it is collected on the machine's next heartbeat, up to about three minutes away. Reading logs is its own permission, separate from viewing the computer, and every request is audited with the log, the window and who asked.
The entry is redacted on the server before anything is sent; the dialog names the provider and model and how many values were removed. The answer shown came from a loopback stand-in on the demo server, not a real provider.
The ticket goes to the business's own connector, or the platform default. The demo server's connector is a loopback stand-in; no real PSA was called.
From 0.47.1 the Spooler action shows the service's last known state. When it is already running the button reads Nothing to do.
Pictures are shown only when their first bytes say PNG, JPEG, GIF or WebP; anything else is a download. The tray's chat window has not yet run on real Windows.
Unread replies from every open chat you can see, from anywhere in the console.
The icon files that tools/windows-installer/cmd/rem0te-tray/icons/gen-icons.mjs generates (0.51) — normal, a chat open, an unread message, the portal unreachable — drawn at 8× on dark
and light. This is not a Windows screenshot. The tray, its menu and the
0.53.1 menu header naming the business and provider have not been photographed on Windows; the header's text is in
the tray guide.
Business, site, who may connect, then the operating system. All of it is fixed into the single-use install link: the machine that runs it cannot place itself somewhere else.
Choosing a business narrows the access list to that business's people.
The clients Rem0te hands out, including the preconfigured RustDesk build that already knows where your server is.
What a Tactical RMM URL Action lands on. Right-click an agent in TRMM and Rem0te resolves it to a computer and connects.
Here it will not: two customers both have a machine the front desk calls
reception-pc, which is most MSPs rather than a contrived example. Rem0te
refuses to pick — guessing opens a remote session on somebody else's machine —
and remembers whichever you choose, so the next launch goes straight through.
Temporary help for a machine that is not an enrolled device. They run a client, read you an ID and a one-time password, and closing it ends the session. No computer record is created.
Answers in the product, not just the repository — how connections are brokered, why a Connect can fail, and what to do about it.
From stored scans only: a device with no fresh scan is a coverage gap, never counted as compliant or not.
Measured from the Offline monitor's episodes. There is no availability percentage, because the time each computer was monitored is not recorded.
Scripts written, versioned and reviewed in the portal, and run only as signed actions. The notice says what has been proven and what has not.
The reviewer approves an exact hash, and cannot be the person who wrote the content.
Tick up to 25 packages and add them to a catalog in one go; each is pinned to its latest version and installer hash. The demo server holds a 36-package sample of the index.
Outdated RustDesk clients updated by the server, only inside a window outside working hours, skipping computers with a session open.
The pre-flight reads each requirement from the real server. The demo server runs a scratch approval signer and release, so only the policy sweep fails: it is off, and the page says exactly that.
One policy for every computer set to Inherit (0.50). The portal builds the rings and proposes a new revision when computers join or leave; computers it cannot patch are listed with the reason. The same page in light: company-policy-light.png.
Inherit, Exclude or Use policy X, per computer. Dark: patch-policy-setting-dark.png.
Second Tuesday plus 2 is the Thursday after Patch Tuesday (0.53). The next three openings are worked out by the code the approval signer holds every ring to. The policy was not saved. Light: monthly-window-light.png.
Every task on every computer you can see, with what it is waiting for and, when it is late, the likely reason (0.52). Dark: tasks-dark.png.
A toast when something you sent finishes, wherever you are in the console; the top bar counts what is still in progress. The demo marked the install succeeded to show it. Dark: task-toast-dark.png.
Three levels and no reseller hierarchy. A Business is the security boundary, enforced on the server rather than by hiding buttons.
Who can do what, per business. Owners hold everything; each Business User shows the capabilities actually granted — not a role name that hides the detail.
The short list of people who run the service itself.
Every account on the platform, across all businesses.
Your own profile, password and two-factor enrolment, each on its own tab.
RustDesk infrastructure, the Quick Connect master switch and MFA policy — the settings that apply to the whole platform.
Name and logo shown to the businesses you support.
Which changes ask you to confirm it is you, function by function (0.49). Light: confirm-it-is-you-light.png.
The business's setting (0.46). A computer's own setting, a site or a device group wins over it.
Platform Admins only. The key is write-only. The demo server's key is a fake one and its provider a loopback stand-in.
Where Create ticket sends. The saved URL is never shown again, only its host; on the demo server that host is a loopback stand-in.
The running version and the release history, parsed from CHANGELOG.md.
The same guides that live in docs/, searchable inside the app.
Captures run against an isolated demo stack — never production, and never a
database with real records. The capture script refuses an https:// target or
port 3000/443 outright.
# 1. Throwaway database (the demo-data module refuses any other name)
sudo -u postgres createdb reboot_remote_docs
cd apps/api
DATABASE_URL=postgresql://…/reboot_remote_docs npx prisma migrate deploy
DATABASE_URL=… npx tsx prisma/seed.ts
DATABASE_URL=… ENCRYPTION_KEY=… npx tsx prisma/_docs-demo-data.ts
# 2. Run the API and web against it on non-production ports (4001 / 4000).
# `output: 'standalone'` bakes the API rewrite in at build time, so patch
# the copied build rather than relying on INTERNAL_API_URL. See CLAUDE.md.
# 3. Re-anchor the demo clocks, then capture. Online/offline is derived from
# how recently a device checked in, and the API sweeps the flag as data
# ages — without this the dashboard counts stop matching the inventory.
DATABASE_URL=… npx tsx prisma/_docs-demo-monitoring.ts # monitors, alerts, a site outage, API keys
DATABASE_URL=… npx tsx prisma/_docs-demo-workbench.ts # 0.31–0.36 pages, see below
DATABASE_URL=… DOCS_AGENT_SHA256=… [DOCS_CHAT_IMAGE=…] \
npx tsx prisma/_docs-demo-dispatch.ts # 0.41–0.48 pages, see below
DATABASE_URL=… npx tsx prisma/_docs-demo-roster.ts # 0.46–0.50 pages, see below
DATABASE_URL=… npx tsx prisma/_docs-demo-almanac.ts # 0.51–0.53 pages, see below
DATABASE_URL=… npx tsx prisma/_docs-demo-refresh.ts # last, right before capturing
# Run the demo API with every sweep off, or it acts on the demo data:
# MONITOR_SWEEP=off CLIENT_UPDATE_SWEEP=off PATCH_POLICY_SWEEP=off \
# METRIC_SWEEP=off RETENTION_SWEEP=off REPORT_SCHEDULE_SWEEP=off \
# SCRIPT_OUTPUT_SWEEP=off CHAT_RETENTION_SWEEP=off AUDIT_SEALER=off \
# STALE_SWEEP=off SNAPSHOT_SWEEP=off TRAY_SWEEP=off \
# AUTO_MIGRATION_SWEEP=off WINGET_INDEX_SWEEP=off AGENT_AUTO_UPDATE=off \
# AUTOMATION_SWEEP=off
# GEOIP_DB_PATH=docs-geoip.mmdb # from apps/api/scripts/docs-geoip-fixture.mjs; unset, it reads this
# # host's GeoIP, and a missing file leaves the maps empty
# For the 0.41–0.48 pages the demo API also needs (see below):
# NODE_ENV=test AI_TEST_BASE_URL=http://127.0.0.1:<stub> # loopback AI stand-in
# APPROVAL_SIGNER_INTERNAL_URL=… UPDATE_REPO_DIR=… # a scratch signer and release repo
# STALE_SWEEP=off matters: the demo's agent-silent computer is isOnline with
# an old heartbeat, and the stale sweep would mark it offline.
# The client-update and patching pages then say their sweep is off; that is
# what the screenshots show, and it is true of the demo server.
WEB_URL=http://127.0.0.1:4000 EMAIL=… PASSWORD=… \
node apps/web/scripts/capture-media.mjs # gallery -> docs/images/github/
WEB_URL=… EMAIL=… PASSWORD=… ONLY=task-toast-light,task-toast-dark \
TASK_FINISH_CMD='DATABASE_URL=… npx tsx prisma/_docs-demo-almanac.ts finish' \
node apps/web/scripts/capture-media.mjs # the toast (changes the data; re-seed after)
WEB_URL=http://127.0.0.1:4000 EMAIL=… PASSWORD=… TASK_FINISH_CMD=… \
node apps/web/scripts/capture-video.mjs # recording -> media/raw/
scripts/build-video.sh # MP4 + poster + highlightdocs/images/github/manifest.json records the route, theme and viewport behind
every image. Credentials are passed as environment variables and are never
written to a script, a manifest, or the repository.
apps/web/scripts/screenshots.mjs is a different tool and is still in use:
it produces the numbered-callout images embedded in the in-app documentation
under docs/screenshots/. This gallery and that guide do not share files. It
runs against the same stack and refuses any other database:
DATABASE_URL=…/reboot_remote_docs WEB_URL=http://127.0.0.1:4000 \
node apps/web/scripts/screenshots.mjs # guide -> docs/screenshots/
ONLY=guide/alerts,guide/api-keys … # just some of themEach guide image is followed in the markdown by a numbered list explaining its callouts; keep the two in step when a callout is added or moved.
_docs-demo-workbench.ts (0.34, extended in 0.36) seeds what the 0.31–0.36 pages need, and can be
re-run: it deletes what it owns first. It refuses any database not named
reboot_remote_docs, like the other demo modules. It writes:
- a Linux and a macOS computer on the read-only agents;
- 30 days of metric history (raw samples for the last 24 hours, hourly rows and six-hourly disk samples before that, with one deliberate gap);
- process, service and network snapshots, run through the API's own sanitiser and size cap;
- device groups and shared saved views;
- resolved alert history, which the alert and offline-time reports count;
- scripts with an approved version and a version in review;
- client-update policies with a site override, a device-group policy on Remote depots, and two held updates;
- a device-group Offline monitor on Remote depots (60 minutes);
- two patch policies, one awaiting approval and one active;
- a fictional relay host (
remote.example.net) and Quick Connect turned on.
_docs-demo-dispatch.ts (0.48.1) seeds what the 0.41–0.48 pages need, and can be re-run. It refuses any database
not named reboot_remote_docs. It writes:
- a 36-package sample of the winget index, for the catalog browser (no four-part version numbers: the masks below would print one as an IP address);
- HARB-DISPATCH-01 and -02 on the signed agent at 0.48.0 with the whole action catalog, so the Services, Processes
and Actions tabs offer their buttons; their agent digest is
DOCS_AGENT_SHA256, which must be thewindows/rem0te-agent.exedigest in the scratch release repositoryUPDATE_REPO_DIRpoints at; - an open, recorded chat on HARB-DISPATCH-01 with emoji, a picture (
DOCS_CHAT_IMAGE, or a generated one) and a text file, the last reply unread, and the chat window reported open for the next 45 minutes.
The Actions, Services and Processes buttons need an approval signer, so the demo stack runs a scratch one
(dist/approval-signer/main.js serve, its own APPROVAL_SIGNER_DIR, policy set --business default --approval direct) and a scratch release repository built with rem0te-release. Nothing is ever approved or
executed: the captures stop at the dialog. Ask AI and Create ticket point at loopback stand-ins: the API runs with
NODE_ENV=test and AI_TEST_BASE_URL (honoured only under test, loopback only) and an obviously fake key, and
the PSA connector is a webhook to a loopback listener. Never point a demo stack at a real provider, PSA, SMTP server
or webhook. Ask AI caches an answer for a day, and a cached one shows a cached badge, so clear AiAnswer before
capturing it.
_docs-demo-roster.ts (0.50.5) seeds what the 0.46–0.50 pages and the walkthrough need, after the dispatch
module, and can be re-run: it clears Harbor's patch-policy series and computer settings first. It refuses any
database not named reboot_remote_docs. It writes:
- Harbor's company-wide series, Harbor — every computer, Tuesday night, at revision 1 (canary HARB-DISPATCH-02, then HARB-DISPATCH-01), with a last check that is settling because HARB-GATEHOUSE is rejoining, and three computers left out with their reasons; the hand-made workbench policies become revoked and rejected;
- an assignable series, Harbor depots — Thursday after Patch Tuesday (a monthly window from 0.53.2: two days after the second Tuesday, 02:00–05:00 Pacific), used by the two depot computers, and HARB-WAREHOUSE-2 set to Exclude;
- a Windows Update scan (patch posture) on HARB-DISPATCH-01, -02 and HARB-GATEHOUSE, so the Patches tab offers updates to tick;
- uninstall keys on HARB-DISPATCH-01's programs, so Uninstall is offered (Office is refused, RustDesk protected), and three global catalog entries for Install;
- the per-computer and per-business agent auto-update settings (HARB-DISPATCH-01 On, HARB-GATEHOUSE Off; Harbor
Any time, Northwind Off). The demo API keeps
AGENT_AUTO_UPDATE=off, so the cards read Effective: Off; - Confirm it is you set to Don't ask for the everyday functions.
_docs-demo-almanac.ts (0.53.2) seeds what the 0.51–0.53 pages and the walkthrough need, after the roster
module, and can be re-run: it clears Harbor's action jobs, deployments and pending scans first. It refuses any
database not named reboot_remote_docs. Every job is a direct-mode plan built the way the portal builds one. It
writes:
- HARB-DISPATCH-01's last update install, answered 21 minutes ago: three installed, one failed (
0x800f0922), restart required, so the Patches tab shows the Latest install card with its restart banner; - an install on HARB-DISPATCH-02 scheduled for the next opening of its policy window, one on HARB-GATEHOUSE handed out six minutes ago, a failed one on HARB-DEPOT-BOI, a Spooler restart queued on HARB-DEPOT-BOI, a log collection on HARB-DEPOT-SPO that is overdue, a pending Windows Update scan on HARB-WAREHOUSE-1 and a finished Firefox install, so Automation → Tasks has a row in every state;
- a two-ring deployment, Defender and MSRT — depots, with nothing requested yet, for the deployment's Schedule canary.
_docs-demo-almanac.ts finish marks the HARB-GATEHOUSE install succeeded. That is the one change a capture makes
to the data: the toast shots and the walkthrough's Tasks chapter run it through TASK_FINISH_CMD so the toast
appears, and the module is re-run afterwards.
The seeded revisions are not on the scratch signer, so a revision's own page says not reachable for its approvals; the captures stay on the Patching page.
_docs-demo-refresh.ts keeps one computer, NWD-BILLING-RMT, deliberately agent
silent: still online to RustDesk, heartbeat three days old, so the amber state
is on the dashboard, the map and its own page. Every other computer reads agent
silent too once it has gone 30 minutes without a heartbeat, so re-run the
refresh every few minutes while capturing; it only moves timestamps.
screenshots.mjs deletes the audit rows its throwaway admin wrote when it
finishes. The audit chain on the docs database then reports those rows as
deleted; that is expected on a scratch database. Capture the gallery (whose
audit and security pages show the chain) before the guide, or reset the
database between them.
The PNGs are written as Playwright produces them; nothing in the pipeline recompresses them.
Masking happens in the browser DOM immediately before each shutter and never writes to the database:
| Value | Shown as |
|---|---|
| Enrollment / claim tokens | SAMPLE-ENROLLMENT-TOKEN |
| Long hex secrets | SAMPLE-TOKEN-VALUE |
| RustDesk IDs | ••••••••• |
| Public IP addresses | 198.51.100.24 (RFC 5737 documentation range) |
| API key prefixes | rk_SAMPLE |
Host names outside mspreboot, github, rustdesk and example, and the AI providers' fixed api.anthropic.com and api.openai.com |
remote.example.com |
Private ranges and loopback are left as they are — they carry nothing.
Rem0te is one of several free tools for MSPs published at mspreboot.com. ↩︎ Back to the README

















































































