The shape
Three loops, running on top of one model layer.
Content compounds on the client's subdomain. Paid ads prune themselves nightly. Outbound speaks in the client's voice. All three feed the same attribution graph.
Content
Articles that compound
Marcus, Jordan, Priya. Friday batch, weekend review, Monday publish.
Learn more
Ads
Self-pruning campaigns
Nightly optimization, 7-day measurement, auto-rollback at -10% or -2σ.
Learn more
Outbound
LinkedIn + email, voiced
Voice model trained on day 1, fidelity within 60 days.
Learn more
The model layer
Elysia.
Self-hosted, three concurrent inference slots, the same instance across every loop.
Elysia diagram + capacity chart (3 concurrent slots, VRAM budget)
The agents
Eight specialists. One job each.
Marcus
Drafts long-form articles
Priya
Edits to client voice
Jordan
Reactions, opinions, takes
Samira
SEO meta + structure
Felix
Paid-ad copy + variants
Virgil
Outbound first-touch + reply triage
Echo
Voice-model training + fidelity scoring
Pulse
Attribution + reporting
ActivityFeedPanel showing all 8 agents working in parallel (port from /components/activity/ActivityFeedPanel.tsx)
The closed loop
Click to revenue, traced.
Attribution funnel with mocked numbers (port from /components/dashboard/InsightsSection.tsx, lines 189-270)
See the platform live