I’m building a personal AI chief of staff that knows my psychology, goals, relationships and injects live context into every single interaction. Here’s the architecture.

Detailed Analysis

A Reddit user in the ClaudeAI community has published a detailed architectural breakdown of a self-described "Personal OS" system — a persistent, auto-injecting context framework built around the Claude API that they call "Master Chief." The system is designed to eliminate the repetitive burden of re-explaining personal context to an AI assistant by pre-loading a rich, continuously updated profile before every API call. The architecture relies on three main components: a Notion database serving as the long-term knowledge store, iOS Shortcuts acting as the real-time data pipeline, and Claude itself as both the reasoning engine and the memory curator that decides what new information is worth retaining after nightly debriefs.

The psychological layer of the system is particularly ambitious. Rather than simple preference tracking, the builder has constructed a 120-question profile synthesizing 17 distinct frameworks — including the Big Five personality model, Enneagram typology, Jungian shadow theory, attachment theory, and Kahneman's cognitive bias research. The explicit goal is not just personalization but metacognitive support: having Claude understand the user's blind spots, decision-making tendencies, and relational dynamics well enough to push back intelligently rather than simply validate. A dedicated "People directory" and "Active fronts" layer extend this into the social and situational domain, giving Claude real-time awareness of the user's relationships and ongoing projects. A compressed 800-word daily summary is injected on every call, with full Notion pages pulled selectively when deeper context is required — a two-tier retrieval architecture that balances token efficiency against depth of context.

The update loop the builder describes is one of the more technically considered elements of the design. Each nightly debrief feeds new information back to Claude, which extracts what is worth remembering, presents it for human approval, and then writes the approved updates back to Notion via API through iOS Shortcuts. This human-in-the-loop memory curation prevents context drift and information bloat while preserving user agency — a meaningful design choice distinguishing it from fully automated memory systems that can accumulate noise or errors over time. The real-time injection of location, calendar, weather, and time data adds a situational awareness layer that most conversational AI interactions entirely lack.

The project sits at the intersection of several accelerating trends in AI deployment: retrieval-augmented generation (RAG), personal knowledge management, and the emerging practice of "prompt engineering at the identity level." As large language models become more capable at reasoning, the limiting factor for their practical utility increasingly becomes contextual grounding rather than raw intelligence. Systems like this represent a DIY precursor to what AI companies are beginning to build natively — Anthropic's own memory and personalization features, for instance, are under active development, and competitors like OpenAI have introduced persistent memory in ChatGPT. The difference here is the degree of intentional psychological scaffolding and the user's explicit goal of decision support rather than task automation.

The broader significance of this architecture is that it treats the AI interface not as a productivity tool but as a cognitive partner — one that requires deep, structured knowledge of the human to function optimally. The framing around decision quality over productivity volume reflects a maturing understanding of where AI adds the most durable value: not in doing more things faster, but in helping humans reason better in high-stakes, context-dependent situations. The fact that this level of infrastructure is being built by individual developers using existing API access, consumer tools like Notion and iOS Shortcuts, and no proprietary infrastructure signals how rapidly the frontier of personal AI deployment is advancing outside of formal product development pipelines.