One gateway,
up to 255 devices
CNC machines, laser cutters and welders, robots, PLCs, and legacy machines with no network interface all connect. Data is normalized at the edge and exposed through five channels — HTTP, MQTT, MODBUS, direct database write, MCP — to business systems and AI tools.
Connectivity and data collection specifications
Current product capabilities, verifiable on site with a single gateway.
Connected by device type and normalized into one set of points; what each type returns is listed on this page
Up to 255 devices per gateway, sampled in parallel
Regular sampling configurable down to 100 ms, set per task
168 HTTP endpoints; 13 MCP tools
Read / upload / download / delete / back up / bulk send; tool offset read-write
Resumes after a network drop; cleaning, conversion and mapping done in the gateway
InfluxDB · MySQL · SQL Server · PostgreSQL
Per-user keys plus an authorized-API allow-list; high-risk write endpoints disabled by default
Modern and legacy equipment,
all connected by three paths
Equipment varies in age, interfaces and protocols. The gateway connects by direct protocol, PLC or IO and outputs one set of data definitions.
Protocol-connected machines
Controls with Ethernet and protocol support connect directly: state, program, position, feed, speed, load, tool number and alarms. Laser cutters and welders add laser power; robots return joint, base, user and world coordinates and alarm logs.
PLC equipment
Registered as PLC machines; points are read by generic and MODBUS area addressing, with state and counts reconstructed in the gateway.
Legacy manual machines
External PLCs, IO and sensors collect signals; gateway data tasks convert these signals into machine state and part counts.
Standardize data at the gateway
to simplify application integration
The same “running” comes from different signals and different rules on different machines. Normalization happens at the edge; upstream systems receive consistent data definitions.
- 01 · Clean Bulk read, pick points by index, rename tags
- 02 · Convert Unit conversion, bit-level flags, status-code mapping
- 03 · Map alarms Raw alarm numbers translated into readable descriptions via a mapping table
- 04 · Expose Standard points delivered through five channels
Five integration options
for different application needs
Gateway interfaces are available locally and remotely through Bivrost Cloud. Each channel provides queries and writes per its protocol; HTTP and MCP share one authentication scheme.
Planned MCP write tools and MQTT event notifications.
Write tools (monitoring thresholds, collection task control, machine registration) map one-to-one onto existing HTTP write endpoints; events push state changes, alarm raised/cleared and part-count increments over MQTT. They reuse the authorized-API allow-list, add write auditing and rate limits, and every write tool requires explicit confirmation. Delivered with a firmware update.
How devices connect,
and what data you get
The gateway connects devices by machine type and normalizes brands, vintages and interface conditions into one set of data points. Different models of the same brand return different data; the manual is the reference for what each system and model supports.
CNC machines
- Run state, alarms, part count, current tool number and program information
- Coordinates, program block, feed and spindle speed, load
- Machine time data for OEE and utilization
- Tool offset read and write, tool life
Read and write PLC addresses, plus macro variables and system parameters on supported models; read, upload, download and delete part programs, manage directories and select programs, with bulk send and backup.
Laser cutters and welders
- Run state, alarms and program information
- Coordinates and feed data
- Laser power (model-dependent)
- PLC data reads
Program transfer and directory management on some models.
Robots
- Run state, alarms and program information
- Joint, base, user and world coordinates
- PLC data reads
Alarm and operation logs, and PLC writes, depend on the model.
PLCs
- Standard protocols such as Modbus and OPC UA, or vendor protocols
- Address-based reads and writes: integer, float, string, boolean and more
- Custom tags and post-processing rules
Machine state, part count and current alarms can come from PLC task results, in the same form as CNC machines.
Legacy machines with no network interface
- An IO module reads stack lights, cycle-start and other digital signals, which the gateway converts into machine state
- Peripherals read through an external PLC
- Address-based reads where the builder supplies a point list
Those readings stand in for the native interface and still come out as standard part count, machine state and current alarms.
Which interfaces each system and model supports: Manual 11.4, Supported Devices. Device not listed there? Tell us the model →
If a gateway fails,
a standby takes over in about 15 seconds
Cluster failover (optional), also called hot standby: all gateways under one cloud platform share a standby pool (N+1). When a gateway fails as a whole, the platform assigns an idle standby to run under its identity; once the original is repaired and stable, it switches back automatically.
heartbeat timeout before a standby is assigned
switch-over, standby online under the original identity
stabilization window before automatic switch-back
Shared standby pool
A standby is not tied to one gateway. It is registered in the cloud platform’s pool and stands by for every gateway under that platform; all gateways are protected by default.
Takeover = identity switch
A standby does not collect while idle. On takeover it imports the failed gateway’s configuration snapshot and restarts under that identity; cloud and dashboard data sources follow automatically.
One takeover at a time
A standby covers one gateway at any moment. If every standby is busy, the platform alarms instead of switching, then takes over as soon as one is released.
Automatic switch-back
A repaired gateway comes back read-only with collection stopped, then switches back after the stabilization window; the standby returns to the pool.
Same-site N+1
The standby sits on the same shop-floor LAN and keeps collecting after takeover; interfaces and dashboards continue.
Off-site DR
The standby sits at another site and keeps serving configuration, queries and dashboards after takeover.
Multiple standbys
Register several standbys to cover several gateways failing at the same time.
Takeover scope, data replication and network setup are detailed in User manual 4.3 Cluster Failover (Optional)
140 machines. None left as “later”
140 machines across the plant, most of them old or non-standard: CNC machines connected by protocol, forging presses and furnaces registered as PLC machines, legacy manual machines reconstructed from IO loops. All three paths output the same points, and plant-wide OEE had a complete definition for the first time.
Bivrost Cloud
The gateway hands data to Bivrost Cloud for monitoring, analytics, asset management and program transfer.
Cloud product pageDNC Pro
Program file I/O goes through the gateway; versioning, approval and six-stage controlled deployment live in DNC Pro.
DNC Pro product pageAI applications
Bivrost Factory AI, developer agents and general-purpose AI assistants read through the same interfaces and MCP.
Developers page One gateway is enough to evaluate.
No line-wide retrofit
Share equipment models, quantities and application requirements to help confirm the trial configuration, pricing and shipping arrangements.
If no message appeared, copy the text below and send it to contact@bivrost.cn.