Deployment examples by situation

Pick the example that matches your number of factories, network and security rules. Examples can be combined: for instance, a multi-factory setup (example 2) may include one factory with an unreliable link (example 5).

The meaning of the icons and the principle that the field connects first are explained in The setup at a glance.

In the pictures, blue lines are connections the field makes to the centre, purple lines are people connecting through a web browser, and green dashed lines are verdicts delivered right on site.

Example 1. Start with a single factory#

When: you are introducing the platform and trying it on one or two lines in one factory.

• What to prepare: one server with a GPU, and one field PC per line. Staff PCs need only a web browser.
• Network: everything is on the same LAN, so there is nothing to open.
• Good to know: if the central server stops, training and the management screen stop. Verdicts in the field continue, though, and data piles up on the field PCs until the server is back.

Example 2. Run several factories from head office#

When: head office manages several factories on one screen.

• What to prepare: one central server at head office, and as many field PCs per factory as there are lines.
• Network: the field PCs only need to reach the central server's address at head office. Nothing inbound is opened at the factories.
• Good to know: you can send a different AI model to each factory and line. If something goes wrong, send the previous model again.

Example 3. Connect over the internet#

When: there is no private network between factories, or the central server sits in a data centre or cloud. It also applies when staff must open the management screen from outside the office.

• What to prepare: put an HTTPS security gateway (a CDN or reverse proxy) in front of the central server. The gateway's address is the only thing exposed.
• Network: field PCs use outbound connections only, so you do not change the factory router or firewall.
• Good to know: change the default password and signing key before exposing the server to the internet (Required security settings). The gateway has request size and time limits. Browser uploads and CLI bundle uploads send files in pieces, so they stay within the limits. Requests that are not split, such as pushing a large image, can be blocked (Reverse proxy caveats).

Example 4. A site with no internet (air-gapped)#

When: security rules cut the factory network off from the internet completely.

• What to prepare: the platform normally reaches the internet in a few places (install files, base files used for training and so on). Download them on a PC outside and carry them in as files. Bring in Edge SDK for the field PCs the same way.
• Network: once everything is in, it runs entirely inside the closed network.
• Good to know: Air-gapped installation lists what to fetch in advance. Upgrades go through the same transfer procedure.

Example 5. Sites with unreliable or slow links#

When: equipment moves from place to place, so there is no network while it works. For example, you tow a mobile control cabinet holding the field PC to the work site, work offline, and connect to the network on a set schedule to sync everything at once. Remote sites and mobile links (LTE) that drop often work the same way.

• What to prepare: a field PC with enough disk to hold data. By default it keeps up to 30 days and up to 50 GiB, so choose a connection schedule that keeps the data gathered between connections within that limit.
• Network: no connection is needed while working. Once the network is available, Edge SDK starts sending automatically.
• Good to know: on the management screen you set how each device sends data, for example send only at night, limit the transfer speed or pause while the field PC is busy. When storage fills up, the least important and oldest data is removed first, and the management screen shows how many items were dropped.

Example 6. Share one installation across teams or customers#

When: you install once and several business units or customers use it together.

• What to prepare: create one tenant (an isolated workspace) per team. Data, models and field devices all belong to a tenant.
• Network: a field PC connects only to its own team's space, because each device's token is tied to that team.
• Good to know: no team sees another team's data or devices. GPUs can go into a shared pool used by several teams, or be dedicated to one team (Core concepts).

Example 7. Add training servers#

When: training grows and GPUs run short, or runs wait too long to start.

• What to prepare: add GPU servers and attach them to the group that runs training and serving (the cluster). The default setup is a single server.
• Network: field PCs and staff keep using the single central server address.
• Good to know: training pods are placed on servers with free GPUs. The number of runs executing at once follows a server setting (default 1), so raise it when you add servers (Server settings). When you service a server, you can stop new work from going to it (Cluster).

Related pages#