How MagnusAI works
Three steps. The first time, then forever.
Pick a platform (phone, computer, web). MagnusAI signs in via MagnusID — one identity for the whole ecosystem.
Pick a permission tier (Cautious, Balanced, Autopilot, Custom). Toggle the per-capability matrix for each kind of action.
Voice or text. MagnusAI routes to the right ecosystem app, gates sensitive actions with consent, voice-confirms above your threshold.
Green-light consent architecture
Every permission is off by default. You choose what to grant, with caps and per-platform rules.
allow_callsallow_smsallow_social_postallow_purchaseallow_ad_spendallow_calendar_read / writeallow_contact_accessallow_locationallow_file_accessallow_app_installallow_data_shareallow_cred_giftingAgent-friendly APIs
Every ecosystem app registers capabilities via a JSON descriptor. MagnusAI consumes them natively — no manual integration work per new app.
{
"app_id": "quiky",
"capabilities": [{
"name": "create_listing",
"description": "List item on Quiky marketplace",
"input_schema": { ... },
"output_schema": { ... },
"required_permissions": ["allow_purchase"],
"consent_required": true,
"price_tier_available": ["free","pro","creator","business"]
}]
}Reversibility on every action
Before MagnusAI does something irreversible, it tells you: "This will post publicly and can't be unposted — confirm?" Scheduled actions stay cancellable until they fire. V1-F3.
More on safety →