You run the coordinator and the relay on hardware you already own. We never see a packet, a key, or an account. Ten days free for the first group, up to ten devices, no card.
Ten days free when it opens · up to 10 devices · no card · macOS, Windows, Linux
The coordinator signs your network with a key generated on your machine that never leaves it. There is no account with us to be locked, and nothing of yours on our side to lose.
This is live — press something.
Your packets never touch us, so we could not meter them if we wanted to. What you buy is the product and the signed key that keeps your coordinator issuing maps — and every update, for as long as the key is current. Billed yearly, in euro, one invoice.
10 devices minimum · €480 / year floor · billed yearly
For a company that wants its own network. The coordinator and the relay run on machines you already own, so they cost nothing.
40 devices = €160 / month. A hosted mesh at €9 per user would bill the same company €180 / month — and hold the keys.
Unlimited machines per site · your technicians counted once
For a company that maintains other people’s machines. One price per site, whatever stands behind the maintenance node — and it is a line you pass straight on to that customer.
One site = €39 / month. Most service companies bill €75–150 per site for remote support, so a site pays for itself on the first ticket.
And no processor agreement, because we process nothing. EU company; personal data never leaves your own hardware.
Ten days of the whole product on your own hardware. If you license it, the mesh you built during the trial is the one that keeps running.
It signs the network map with an Ed25519 key that never leaves your machine. There is no account with us that could be locked, sold or subpoenaed.
Devices connect directly when they can. When a network refuses, the relay in the same installer carries the traffic, still inside your own walls.
The app, the agent, the coordinator and the relay ship together as one .pkg or .msi. Any Mac or PC can host the mesh.
A one-person company, a firm with staff on the road, a service company with a hundred customer sites. The software is identical; what differs is one file of rules saying who may reach what.
The machine under the desk becomes a name in your mesh. From the train, from a hotel, from the office, you open it as if you were sitting next to it. No port forwarding, no dynamic DNS, nothing exposed on the router.
The coordinator can live on that same machine. A phone, a laptop and the server is a complete mesh.
One gateway in the office advertises the office subnet. Staff get the printers, the file server and the internal tools wherever they are, and the tunnel follows them from home Wi-Fi to a hotspot without a reconnect.
Sign-in through your identity provider with MFA if you have one, or keep the mesh invite-only.
Each customer site gets a maintenance node that technicians reach on demand and that can never initiate a connection back. The isolation is structural, not a checkbox. Access is granted for a window, expires on its own, and every session lands in the audit trail.
The boxes that do the work are rarely in one place: one under a desk, two in a colo, one rented by the hour. On the mesh they are names with fixed addresses, reachable by the people in one group and by nobody else, and no training data, prompt or checkpoint travels through a third party on the way — because there is no third party in the path.
The same coordinator that signs the network also distributes the tooling: your inference harness, your eval runner, your own agent worker, signed by you and installed on the machines you aim them at.
Up to ten devices, every feature on, no card. Nothing is created on our side, so there is nothing to cancel.
There is no console to provision and no cloud tenant to create. You install, you invite, and you write down who may reach what.
Run the .pkg or .msi. The first machine becomes the coordinator and relay; it also gets the app and the agent like every other device. It sits in the menu bar.
Send an invite. The device receives its name, a virtual address in 100.64.0.0/10 and a network map signed with your Ed25519 key. It only trusts maps that carry that signature.
Rules are sentences: a group, an arrow, a target, a verb. The map carries the rules, so a device that is not named in one simply cannot see the other side. Nothing to audit later that was not there in the first place.
You hold the publisher key. Bridgewall has no way to publish to your devices, and no key of ours is trusted by them. The coordinator already knows every device and which group it belongs to, so it can also hand them software: you sign an artifact with your own key, aim it at a group, and each device checks the signature against the key pinned in your admin console before anything installs. Nothing installs that you did not sign, there is no public store in the path, and there is no installer link being forwarded around in chat. Revoke the key and every device stops accepting new artifacts from that moment. If you maintain machines across twenty customer sites, this is the part Tailscale leaves to your RMM.
Most internal tools travel by shared drive, a link in a chat, or a technician with a USB stick. In a Bridgewall network they travel the same path as everything else: signed, aimed at a group, and logged. Publishing is an upload and a version number; withdrawing is one switch, and the offer stops being served to every device at once.
The catalog is part of the product, not a separate service. It runs on your coordinator, from your machine, with your key.
The first app in the catalog is our own build worker, Auto-Mate. How that one works →
You download the same installer our customers run. Up to ten devices, no card, no feature switches. Nothing you set up is throwaway — a license key turns the trial mesh into the production one, same machine, same devices, no reinstall.
The question every engineer asks first, answered without hand-waving.
Every infrastructure product has this list. Most of them do not publish it.
Auto-Mate is a work engine, not a script runner. You draw a workflow as a graph — steps, conditions, the branch that fires when a step fails — put it on a clock or file work into it by hand, and agents running on machines you own carry out each step and have to prove they did before the next one starts. The recurring things a team does by hand every week become a flow that runs itself and reports back.
It installs the way any of your own apps would: signed by us, pinned by you, aimed at a group you choose. It needs a Claude subscription signed in on each machine that runs work — the work executes under that account, inside two ceilings you set, so it can never spend more of your quota than you allow and we never see any of it. €19 per worker machine per month, and Bridgewall behaves identically if you never install it.
Yes, and it has to be signed in on each machine that runs the worker. The worker executes under that account, so its usage comes out of your own windows and nobody else’s — that is what makes the two ceilings meaningful, and what keeps us out of the loop entirely. Without a subscription the worker will not claim work.
This applies to the add-on only. Bridgewall itself has nothing to do with Claude.
No. Only a key you generate and pin can. We have no publisher key on your mesh, and the catalog can be switched off entirely at install.
The same installer as everywhere else, with two more parts switched on: the coordinator, which holds the Ed25519 key and signs the network map, and the relay, which carries traffic between devices that cannot reach each other directly. It is a Mac or a PC that stays on. It does not need a public address; it needs to be reachable by the devices you invite.
No. Device traffic goes directly between your devices or through your own relay. The map is signed by your coordinator. We are not in the path and we do not hold a copy of your keys.
The trial key expires and the coordinator stops signing new maps. Devices only trust a current signed map, so they drop their tunnels within the hour. Nothing is deleted: names, addresses, rules and device enrolments stay on your coordinator. Enter a license key and the coordinator signs again; devices reconnect in seconds.
You get one reminder email on day 7.
Either way you like. A mesh can be invite-only, where the admin sends each device an invite from the app, or it can use your identity provider with SSO and MFA so that staff sign in with the account they already have. Both can be combined, for example SSO for staff and invites for a handful of contractors.
No. A maintenance node accepts connections and never initiates them; the map it receives does not contain your side as a target. That is enforced by the signed map on every device, not by a firewall rule you could forget. Access grants are time-boxed and every session is written to the audit trail.
A license key for your coordinator. €4 per device per month for your own staff, and €39 per customer site per month if you maintain other people’s machines. Billed yearly, ten devices minimum, and the figure is on the invoice before you spend a minute of the trial — see pricing.
Yes. The app catalog is part of the coordinator. You upload a build, sign it with your own publisher key, and choose the group it is offered to; the devices in that group see it in the Apps tab and verify the signature against a key pinned in your console before anything installs. Withdrawing an offer stops it being served everywhere at once.
Our own Auto-Mate Worker ships as an app in that catalog. It is an add-on, priced per worker machine, and it runs on your own Claude subscription under two usage ceilings you set — we never hold a key for it.
