The AI assistant
There are three places the assistant appears, and they do different jobs.
Every question costs one AI assistant query from the monthly allowance — 25 a month on Free, 1,500 on Watch, 6,000 on Overwatch. Reading back a past conversation is free.
#The console AI
The ✦ AI button in the console, beside the search box. Ask in plain words and it answers from the live feeds, then opens the app that holds the detail.
"What's on fire in Greece." "Take me to where the sun is overhead." "Which ports are congested."
It is the fastest route into the product when you know what you want to know but not which app holds it.
#Per-app AI
Most apps carry a ✦ in their window header that swaps the manual controls for a conversation scoped to that app's own data.
The News AI summarises the last hour or keeps only what is critical. The Hazards AI filters to M5+ quakes or summarises by region. The Layers AI turns the whole map configuration on and off by description. Regions briefs a region from current sources.
Scoped means scoped: the Markets AI cannot see your Telegram, and the Hazards AI cannot run a DNS lookup.
#The lookup planner
Lookups' AI mode is the one that does more than talk. Give it a subject — a domain, an address, an email, a CVE — and it:
- Writes a plan — the sequence of lookups it intends to run, shown before it runs them.
- Runs them, each step marked as it completes: running, ok, or failed.
- Reads the results together and tells you what they add up to, with the raw response of every step one click away.
It can only reach the tools of the app it is running in, and only the tools your plan includes. It costs the AI meter for the conversation plus the normal lookup cost of each tool it actually runs — the plan is shown first partly so that is never a surprise.
#Picking a model
The header carries a model picker with three settings:
| Setting | Use it for |
|---|---|
| Fast | Quick questions, filtering, summarising a list |
| Agent | The default — chat and multi-step lookups |
| Smart | Harder reasoning, longer synthesis |
Beside it, a reasoning effort control, and a context readout showing how much of the conversation window is used. Long conversations eventually need a new one; the readout is the warning.
#Conversations are saved
Every thread is kept under /account → Sessions, split between general assistant chats and the lookup and Telegram sessions. You can reopen them, and you can delete them one by one. They also come out whole in an export.
#What to trust
The assistant reads live feeds and real lookup results, and it names its sources. It is still a language model:
- Region and news briefs are generated. Every one carries a line saying so and listing what it read. Verify anything that matters independently.
- It cannot see what you have not collected. Asked about something outside the feeds and your own findings, it should say so rather than fill the gap.
- The support assistant is a different thing. The one in the Support app answers questions about the product and hands over to a person for anything touching money, access or security. It is not the assistant described here.
Your questions and the data they touch are sent to the model provider to be answered. See Your data.