Relationship beats unused compute
Most daily AI work is personal and practical. Continuity, trust, and usefulness can matter more than maximum benchmark capability.
resting
Investor overview · private preview
The relationship layer for personal AI.
Lumen separates the user’s identity, continuity, permissions, and trust from the model underneath—creating a personal intelligence that can improve without being replaced.
User-owned continuity. Governed action. Interchangeable compute.
Pre-seed conversation · founder-built prototype · claims separated by available, private-preview, and roadmap status
49 / 53Representative answer-quality checks passing
11,298Strict egress tests with zero violations
74% lessSystem prompt carried into each turn
Mac + iPhoneAuthenticated continuity across owner-controlled devices
The thesis
AI becomes more valuable when the relationship survives the model.
Frontier capability will continue to change. The durable product can be the user-owned layer that makes any capable model personal, governed, continuous, and useful.
Model independence compounds
Lumen can improve as models improve without forcing the user to abandon the intelligence, memory, and workflow they already built.
Trust expands the addressable role
Visible memory, bounded authority, and receipts create a path from conversation to meaningful work across software and future physical forms.
System architecture
Lumen is the identity. The model is an interchangeable brain.
This separation is the foundation for local-first deployments, user-selected compute, cross-device continuity, and eventually dedicated private pods.
Identity
The relationship, preferences, permissions, and continuity that belong to the user.
Brain
An interchangeable model selected for capability, privacy, latency, and available compute.
Body
Mac and iPhone applications today; glasses and robotics are future forms.
Constitution
Authority, evidence, privacy, and safety rules that persist when the model changes.
User authority→One governed route→Selected model + tools→Answer and receipt
Current engineering evidence
Progress measured by outcomes—not spectacle.
49 of 53 checks passing
The current collapse candidate scores 92.5% on its representative answer-quality battery.
Zero egress violations
The first strict run completed 11,298 tests without an unapproved outbound-data violation.
7,350 → 1,895 tokens
The One-Door collapse reduced the system prompt carried into each turn by 74% while maintaining the current quality result.
Mac + iPhone foundation
Native applications use an authenticated device bridge and explicit continuation across owner-controlled devices.
Status snapshot: September 30, 2026. The One-Door collapse remains in progress.
Defensibility
The compounding asset is trusted continuity.
A model can be swapped. A carefully governed history of collaboration, corrections, permissions, projects, and preferences becomes increasingly useful to its owner over time.
- 01
Relationship data stays user-controlled
Value grows without requiring an advertising or data-resale model.
- 02
Behavior improves across models
Better brains strengthen Lumen without resetting the user relationship.
- 03
One identity reaches new bodies
Desktop, phone, glasses, and future robotics share one governed behavioral language.
De-risking sequence
Prove transfer before scaling compute.
Complete One-Door collapse
Remove duplicate paths while preserving quality, privacy enforcement, and clear authority.
Independent family pilot
Deploy Lumen to a second owner and measure usefulness, correction, trust, and retention.
20–50 person private cohort
Validate repeatability, onboarding, unit economics, support burden, and daily engagement.
Pods, glasses, and owned models
Scale private compute, spatial interfaces, and model research only after the relationship layer proves itself.
Founder
Built by an electrician directing an AI engineering team.
Lucas Thomas runs Bolt Electric in Jacksonville, Florida. He began building the personal AI system he believed people deserved: useful, continuous, correctable, and accountable to its owner.
The unconventional advantage is direct product conviction paired with relentless AI-assisted execution. The current system spans native applications, local model infrastructure, memory, governed tools, business operations, and a working interaction prototype.
Talk with Lucas and Lumen →Private investor conversations
Help build personal AI that belongs to the people using it.
Architecture review, prototype demonstrations, pilot design, and funding conversations are welcome.