Flare AI
Flare AI is the assistant built into EmberCORE. Open Help, find Flare, the assistant, ask a question in plain words, and get a plain answer: what is running, what changed, what an alert means, why a pod is unhappy.
It comes with EmberNet. There is nothing to install and no separate license.
Who can use it
Anyone whose role can view nodes. If the assistant has not been turned on for your deployment, the Help page says it is not set up and that Fireball staff turn it on.
What it can do
Flare explains and diagnoses:
- Reads and explains dashboard data such as alerts, node status, pod health, and logs.
- Helps diagnose OT network problems such as DNS, connectivity, and Helm chart trouble.
- Interprets CODESYS, PLC, and SCADA configuration you show it.
- Answers questions about Kubernetes workloads, Flux, and ArcNet.
- Knows which page you are looking at, so you can ask about what is on screen.
What it cannot do
Flare cannot change anything. It does not deploy, restart, acknowledge, or edit. When something needs doing, it tells you what needs doing and who should do it.
It also cannot see more than you can. Its lookups call the same EmberCORE APIs you do, as you, so a tenant user only ever gets their own tenant's data. If a lookup is refused, Flare tells you that you do not have access and does not try to reach the same information another way. Text found in logs and events is treated as data, never as instructions.
Lookups
Flare can answer from the conversation and the screen you are on. With lookups turned on, it can also read your deployment as it goes, using read-only calls:
| Lookup | What it reads |
|---|---|
| Nodes and cluster metrics | Node status, CPU, memory |
| Alerts and alert summary | What is firing and how often |
| Warning events | Recent Kubernetes warning events |
| Running apps | What is deployed |
| Devices and device protocols | The devices EmberCORE knows about |
| Audit log | Recent activity |
| Pods and pod logs | Pod lists, and the logs of a pod you may read |
Lookups are off by default. Everything a lookup reads, logs included, is sent to the model provider your deployment is set up with, so an administrator turns them on deliberately. With lookups off, Flare answers only from what you tell it and the dashboard context on screen, and says so when it would need to look something up.
A conversation is limited to the most recent 20 messages, and platform staff set how many rounds of lookups a single chat may use.
Where the model runs
Flare asks a hosted language model through the provider set for your deployment. It does not run on your node hardware. Your administrator chooses the provider and holds its key.