Early in the design process, I considered building two separate systems: one for professional artifacts and one for personal. The professional system would handle technical remediation, client management, and business operations. The personal system would handle life logistics, health tracking, and personal goals.
I abandoned that idea within a week. The reason is simple: the decisions that matter most are the ones that span both domains.
The Boundary Problem
When I am deciding whether to take on a new client engagement that requires travel, the relevant context is not purely professional. It includes my family commitments, my energy level based on the past month’s workload, upcoming personal obligations, and a pattern I have noticed over twenty years: I make worse technical decisions when I am overextended personally.
No professional-only system can surface that pattern. No personal-only system knows about the client engagement. The value is in the intersection, and the intersection requires a unified model.
This has privacy implications I take seriously. The unified corpus contains personal health information, financial data, family communications, and intimate reflections alongside client data, technical documentation, and business strategy. The security model for this system is not enterprise-grade. It is personal-vault-grade. Air-gapped, encrypted at rest, with no external API calls. Ever.
Context Windows for Life
Tessera maintains what I call “life context windows,” rolling summaries of different life domains that stay current as new artifacts are ingested. There is a health context window, a family context window, a financial context window, and multiple professional context windows organized by client, project, and domain.
When I ask Tessera a question, the relevant context windows are activated based on the query classification. A technical remediation question activates the client context, the technology context, and the decision history context. A scheduling question activates the family context, the health context, and the commitments context.
The context windows solve a fundamental problem with large language models: they can only attend to so much text at once. By pre-summarizing domains and activating only the relevant ones, Tessera can effectively reason across a much larger knowledge base than the model’s context window would normally allow.
The Integration Layer
Life context windows are updated incrementally as new artifacts arrive. An email from my doctor updates the health context. A client escalation email updates the client context and the remediation context. A calendar invite updates the commitments context and potentially the family context if it conflicts with personal obligations.
The update process is not just append. It is reconcile. If new information contradicts something in the context window, Tessera flags the contradiction for my review rather than silently overwriting. This is critical for maintaining trust. I need to know when Tessera’s understanding of my world changes, especially when the change involves a conflict between domains.
This is what makes Tessera a life assistant rather than a work assistant. It holds the full picture, and the full picture is what enables the kind of judgment support that no compartmentalized system can provide.