AI Edge Assistant
From within the gateway screen, you can ask about edge data in your own language. Ask questions like "Are there any connections not currently being collected?" and the AI will select the necessary queries on its own and answer.
ai.enabled is false. To use it, enable it in Settings and enter the model address and key
(see How to enable).
What you can ask about
It covers what's inside the edge — edge status, OPC connections, tags, collection status. Equipment hierarchies or alarms in the upper platform are out of scope for this assistant.
| If you ask like this | What the AI uses |
|---|---|
| "Show me the connection list" | Connection list query |
| "How is connection 3 doing right now?" | Connection details, collection status |
| "What's the current value of this tag?" | Tag current value (multiple at once) |
| "Values have been weird for a while, show me the last hour" | Tag history |
| "Is it being sent to the upper level properly?" | Transmission queue status |
| "How's the gateway load right now?" | Monitoring metrics |
You can also attach photos — take a picture of an instrument or screen and ask about it.
🔴 Control requires "Approval"
Among the tools the AI can use, three change state.
| Tool | What it does |
|---|---|
start_opc | Starts a connection |
stop_opc | Stops a connection |
write_tag | Writes a value to a tag |
These three are not executed by the AI on its own. A prompt saying "Approve this control operation?" will appear, and it executes only after a person presses Approve. If you cancel, nothing happens.
If you leave the approval request up and press it much later, you'll see "Approval has expired or is invalid." In that case, ask again.
Tags whose client_access is R cannot be written even if the AI receives approval. The
path the assistant uses is the same as the existing API, so any guard in place there applies
likewise → Add tag
Query tools (lists, values, history, status) don't change state, so they run immediately, without approval.
How to enable
Configure it via app.properties → Web UI system configuration
| Setting | Default | What it is |
|---|---|---|
ai.enabled | false | On/Off |
ai.api.base_url | (empty) | Model server address — an on-prem gateway or local server |
ai.api.key | (empty) | That server's key |
ai.model | gpt-4o-mini | Model name |
ai.max_tool_steps | 8 | Maximum number of tool calls per question |
ai.timeout_ms | 30000 | Response wait time |
ai.bulk_read_limit | 50 | Upper limit for tags read at once |
"AI assistant is disabled. Specify ai.enabled and ai.api.base_url in Settings." — Both are required for it to work.
If ai.api.base_url points to the external internet, field data will leave your premises.
For air-gapped networks or sites where data export is blocked, point it to an on-prem gateway or a local model running inside the gateway. The key is not logged.
If it doesn't work
| Symptom | Cause |
|---|---|
| "Disabled" | ai.enabled is off, or ai.api.base_url is empty |
| "Could not respond" | Can't reach the model server — check address, key, firewall |
| "Too complex" | The question is too difficult to answer at once. Narrow the scope and ask again |
| "Approval has expired or is invalid" | The approval request is old. Ask again |
| Control rejected | The tag's client_access is R |