Decide whether you need a mesh
before choosing which one

Most remote-access purchases are made before the problem is properly defined, and half a year later the tool is either too limited or barely used. This page does the subtraction first: three questions to decide whether you need a mesh at all, what each of the four common paths costs you, and a pre-launch checklist worth going through once.

Three questionsFour paths and their costs12-item checklist
Step 1 · Subtract

Three questions eliminate half the options

Start from your constraints, not from product features. The answers narrow the field quickly.

Question 1

Do you have a static public IP?

If yes and you can change router settings, port forwarding or a reverse proxy covers most "reach one system from outside" cases, and you do not need a mesh. If no (home broadband, mobile, most office lines), that path is closed and a mesh becomes relevant.

Question 2

How many people and devices, concurrently?

One or two people occasionally: a remote desktop tool or a commercial VPN is less work. Three or more devices, multiple concurrent users, always on — that is when deployment and maintenance effort pays off.

Question 3

Can the data leave your network?

If yes, and topology passing through a third party is acceptable, a hosted service is the least effort. If no, only self-hosted works — and you accept being your own operations team.

The combination maps to four paths

Public IP and one system → port forwarding. No public IP and two devices → remote desktop. No public IP, many devices, hosting allowed → hosted mesh. No public IP, data cannot leave → self-hosted mesh. None is better than another; only fit matters.

Step 2 · See the cost

Four paths, and what each one defers

All four work. The difference is who pays later for the part you postponed.

PathWhat you saveWhat you oweFits
Port forwarding / reverse proxyNo licence cost, lowest latency, no clientInternal services exposed to the internet; attack surface equals the service; one system at a time; no identity or auditStatic IP, publishing web services only
Remote desktop / traditional VPNFast to deploy, staff already know itRemote desktop hands over a whole machine; VPN needs certificate and gateway upkeep, and mobile support is poor1–2 people, short term, one computer
Hosted meshNothing to deploy, vendor support and invoicing, optimised domestic routingControl plane belongs to the vendor, so topology and policy are not yours; per-seat or per-device annual fees grow linearlyNo server, needs invoices and an SLA
Self-hosted meshControl plane and data stay in your network; cost tracks one serverNeeds a server and basic operational skill; nobody else will debug it at 2 a.m.Sensitive data, in-house ops, growing scale
Step 3 · Do the maths

Licence cost is the visible part

Four items decide whether you regret the choice three years later.

01

Server and bandwidth

Self-hosting needs a machine; if it runs in the cloud, bandwidth counts too. Relay traffic goes through it, so a thin link degrades everyone at once.

02

Operations hours

Estimate hours per year: about half a day to deploy, quarterly upgrades and backups, plus incidents. Hosted services convert this into licence fees; self-hosting converts it into your own time.

03

Downtime

When remote access breaks, work stops. This line is usually ignored until the first outage — multiply the cost of one lost day by the probability, and it often exceeds the licence.

04

Migration cost

Switching means reinstalling clients, rejoining devices, rewriting policy and retraining staff. Ask "what does it cost to leave in three years" before asking "what does it cost today".

A formula you can reuse

Three-year cost ≈ licence × 3 + server × 3 + annual ops hours × hourly rate × 3 + downtime cost × expected incidents. The first two are easy; the last two are where options actually differ.

Step 4 · By size

Scale changes what matters

3–10 people

Priority: do not tinker

No dedicated IT at this size. You want it to work after install and be diagnosable without help. A GUI console, local-language docs and one-command deployment matter more than expressive policy models.

10–50 people

Priority: access and revocation

Contractors, departures and multiple departments appear. You need per-person and per-device rules, one-click suspension, and a record of who changed what. Without ACL and audit, an incident is a matter of time.

50–200 people

Priority: audit and compliance

You must answer who could reach what, when, and who edited the rules. Retention period, log export and dual approval become procurement requirements rather than nice-to-haves.

Before launch

A 12-item checklist

Use it as the agenda for the go-live review. Each skipped item tends to come back as an incident within two weeks.

01

Subnet overlap

Do the office, stores and home networks share private ranges (usually 192.168.1.0/24)? Re-plan before launch day, not during it.

02

DNS plan

Will systems be reached by name or by IP? Decide before training staff; changing it later is a company-wide announcement.

03

Who approves devices

Name the approver and the person responsible for disabling departing staff. Without an owner, the account list only grows.

04

Offboarding runbook

Disable the account, revoke the device, check for shared credentials. Put it in the offboarding process, not in someone's memory.

05

Relay bandwidth

When hole punching fails, all traffic relays. Size the relay for worst-case concurrent users × per-user bandwidth, not for the average.

06

Log retention

Depends on what you must answer: internal audit, regulatory requirement, or just who changed a rule. Retention drives which tier you buy.

07

Backup and restore

Control-plane config, keys and device inventory must be recoverable. Scheduled exports stored separately is the cheapest insurance.

08

Upgrade window

Fix a maintenance window and announce it. Remote access is production dependency by then; it cannot be restarted at will.

09

Cross-platform check

Can Windows, Mac and phones all connect, and what happens when someone replaces a device? Mobile is usually the last thing tested.

10

Least-privilege baseline

Deny by default, open only the subnets and ports required. "Allow everything" to save time is handing over the internal network.

11

Rollback plan

If the experience is unacceptable, how do you go back? Keeping the old path running in parallel for a while is the cheapest option.

12

One measurable acceptance criterion

For example: "a travelling colleague opens the ERP and exports an invoice within one minute." Without one, you cannot say whether the rollout worked.

Misconceptions

Five judgements that cause rework

"We have a public IP, so we do not need a mesh"

A public IP answers whether something can be reached, not who may reach it, what they did, or how access is revoked. Exposing one system by port forwarding is reasonable; exposing ten turns into exposure management.

"Free software means it costs nothing"

Free refers to the licence, not the cost. Self-hosting costs a server and your hours; hosted free tiers become expensive on a curve that is usually steeper than expected.

"If hole punching works, we are done"

NAT types change with networks, ISPs and routers. What is direct today may relay next week, so plan the relay path alongside the direct one.

"Is a mesh just a public proxy"

No. An enterprise mesh connects your own devices into a private network so you can reach your own systems. It does not provide public internet access and does not change who your traffic egresses as. Using it as a general-purpose proxy fits neither the product design nor compliance requirements.

"Set it up once and forget it"

People, devices and networks change. Without a quarterly review you will have a handful of active accounts belonging to people who left.

Try one scenario, not a company-wide rebuild

Three devices: one in the office, one in your hand, one for the colleague who needs remote access most. Open one port. Run it for a week. That beats any selection document.