Cinnamon, a new AI agent designed for household management, has launched with a focus on shared context and explicit user approval. Built by infrastructure veterans from Meta, Instagram, and Amazon, the platform allows users to delegate tasks like travel planning, inbox triage, and price monitoring to an autonomous agent that operates within strict privacy boundaries. The project, posted to Hacker News on November 6, 2026, emphasizes that no action is taken without the user’s explicit consent, a departure from many autonomous agents that act first and ask questions later.

The Architecture of Trust

The core differentiator for Cinnamon is its dual-space architecture. The platform separates tasks into a 'Household Space' visible to up to six invited members and a 'Personal Space' that remains strictly private. This design ensures that sensitive personal data, such as job searches or medical appointments, never leaks into the shared family context. The agent maintains plain-English memory facts that users can inspect, edit, or delete in one tap, rejecting the 'black-box' memory profiles common in other assistants. This transparency extends to data handling: the team claims no ads, no model training on user data, and OAuth tokens stored in vaults inaccessible to prompts.

Infrastructure Pedigree and Pricing

The founding team, led by CEO Michael, brings significant scale experience. Michael previously built Meta’s ML capacity forecasting framework for Instagram and Reality Labs and owned core streaming protocols for Amazon Prime Video. He also co-founded Advection, a CDN scaled to 1 billion monthly streams, and Monitoreal, an edge-AI video analytics company. This background informs Cinnamon’s enterprise-grade discipline in consumer AI. The pricing model reflects this robustness: a free trial with 25 credits is available, followed by a Personal Plan at $9.99/month, a Group Plan at $19.99/month for up to six members, and a Group Plan Max at $99.99/month for up to ten members. Notably, Group plans include private Personal Spaces for all members at no extra cost.

Key Takeaways

  • Cinnamon uses a 'cat eyes' UI metaphor to indicate when the agent is active (eyes open) versus idle or waiting for approval (eyes closed).
  • The agent never sends emails, modifies calendars, or spends money without explicit user approval, acting as a high-trust intermediary.
  • Memory is scoped strictly to either the shared household or the private personal space, with no cross-contamination allowed.
  • The platform is backed by founders with experience managing distributed systems at Meta, Amazon, and via their own startups Advection and Monitoreal.

The Bottom Line

Cinnamon represents a mature step toward agentic infrastructure, prioritizing trust and privacy over raw autonomy. By leveraging the founders' experience in high-availability systems, it offers a compelling model for household AI that respects boundaries while delivering utility.