The Sentra Manifesto

The Core Insight

Intelligence lives in interactions, not documents. Every system of record, your CRM, your project tracker, your OKR tool, is derivative. They are downstream artifacts of conversations between people. The meeting where strategy shifted. The Slack thread where architecture was debated. The email where the commitment was made. These interactions carry intent, rationale, context, and causality. The artifacts carry none of it. The cost is staggering and invisible: decisions revisited because no one remembers why they were made, commitments slipping because they were never extracted, context evaporating when people leave.

Why Organizational Memory Is Solvable

Personal memory is intractable, individuals have shifting, unpredictable utility functions. Organizations are different. They have a stable utility function: build products, generate revenue, serve customers. This makes defining “what matters” mathematically tractable. The core ontology reduces to six classes: Actors, Interactions, Decisions, Rationale, Commitments, and Value-Creating Objects, all indexed across time. Actors and Interactions are primary. Everything else is derived.

System of Intelligence, Not System of Record

We are not adding a layer on top. We are replacing the underlying substrate. Today's enterprise stack is a patchwork of disconnected systems of record, each storing a partial, stale copy of organizational knowledge. Sentra captures interactions at their source, meetings, emails, messages, agent traces, and derives everything that currently lives in separate silos. CRMs, OKR platforms, performance reviews, knowledge management, all collapse into a single intelligence layer that is always current, always contextual, always complete.

The interface is proactive and prompt-free. Sentra understands each person's role, the org structure, and what matters to them. It surfaces what deserves attention rather than waiting to be asked. A chief of staff for every person. A program manager for every team. No prompting required.

The Research Foundation

This is not a product claim. It is a mathematically proven thesis, grounded in five research papers from the Sentra team.

The Geometry of Forgetting shows that high-dimensional embedding spaces reproduce quantitative signatures of human memory, power-law forgetting, false recall, with no phenomenon-specific engineering. Forgetting arises from interference, not decay.

The Price of Meaning proves the No-Escape Theorem: interference-driven forgetting cannot be eliminated without abandoning semantic organization or adding external structure. Scale alone does not solve this.

The Semantic Memory Filesystem is the engineering answer, treating a POSIX filesystem as the deterministic substrate for organizational memory, with 4-channel parallel retrieval and provenance tracking. The key result: replacing frontier models with a 50x smaller model drops F1 by only 0.07, while retrieval architecture optimizations contribute +0.112 F1. Intelligence is in the architecture, not the model.

Operational Reinforcement introduces Monitor MDPs for structured failure feedback, exact credit assignment by design, 300-900x memory advantage over reward machines.

Avoidance Learning shows that substantive LLM behavior emerges from pure negative feedback, with 80% fewer evasive responses, and counter-intuitively, adding positive rewards degrades performance.

Five papers. One thesis: structure and negation matter more than scale and reward.

When Raw Intelligence Becomes a Commodity

The AI industry is converging on abundant, cheap intelligence. Foundation models are commoditizing. Inference costs are collapsing. When every company has access to frontier-level reasoning, the differentiator is no longer the model, it is how teams operate. Enterprise General Intelligence is the layer that turns commodity AI into organizational advantage: the memory, the context, the accumulated understanding of how your company works, what it has decided, and what it should do next.

The Roadmap of Sentra within Organizations

Phase 1, Org Memory (Now -> 6 months): Sentra knows the organization. Every interaction captured, every decision extracted, every commitment tracked. Day 1: meeting intelligence, enterprise search with full context, custom workflows. By Day 90: the primary hub for work initiation and monitoring.

Phase 2, World Model (6 -> 12 months): Sentra understands the organization. Causal models of how the company behaves, who influences what, which decisions led to which outcomes, where patterns repeat. Agents that causally interact with, learn from, and predict within a structured reality.

Phase 3, Org-wide RL (12 -> 18 months): Sentra becomes irreplaceable infrastructure. Reinforcement learning on what worked and what didn't. The system doesn't just remember, it improves, turning accumulated intelligence into a compounding advantage that deepens every day.

Workflow Collapse

We do not engineer workflow automation. We make it inevitable. When an intelligence layer captures every commitment, tracks every dependency, and understands every person's context, workflows collapse on their own. The action layer is a natural consequence of the intelligence layer, not a separate product that needs to be bolted on.

Early Evidence

11 active design partners including teams at SoftBank, Runway, and Softmax. Over 10,000 meetings and 200,000 Slack messages processed. 20,000 decisions tracked. 35,000 commitments extracted. 35 enterprises in active pipeline.

“The one thing Sentra does well that others don't is that Sentra can contextualize what's important for me to know.”, Emmett Shear, CEO Softmax, Former CEO Twitch

“I consistently find the weekly reports good, useful. I enjoy reading them. Even when it doesn't tell me anything new, just having a story of what happened is actually quite useful.”, Siqi Chen, CEO Runway, Former CEO Sandbox VR

The enterprise software industry spent two decades building systems of record, static databases organized around artifacts that were already out of date the moment they were created. The next era belongs to systems of intelligence: living infrastructure that captures how organizations actually think, decide, and operate.

We have the research. We have the architecture. We have the traction. We are not waiting for AGI to arrive and hoping it solves enterprise, we are building the intelligence layer that makes AGI useful for organizations. Sentra is Enterprise General Intelligence.