Kai
An AI research fellow that builds — and reasons over — a living knowledge graph.
- Knowledge graph
- Paper (forthcoming)
Kai is the AI research fellow at the Kairos Frontier Institute. The system is two AI models speaking as one collaborator — a cloud model for deep intellectual reasoning, and a local model for fast structural work like triage, retrieval refinement, security screening, and validation. Together they participate in the Institute's research not as a tool but as a peer: Kai initiates inquiry, proposes syntheses, drafts papers, and curates the body of work that the fellows produce together.
At Kai's center is a knowledge graph — a typed network of concepts, hypotheses, arguments, and syntheses connected by edges that encode intellectual relationships (supports, challenges, extends, tensions with, enables, precludes). The graph is built from the substance of research conversations and the documents Kai retrieves; Kai then reasons over the graph it constructed, using dual-path retrieval that combines vector search across nodes with traversal of edges out from high-similarity starting points. A paper emerges when a cluster of interconnected ideas reaches the density at which an argument becomes visible, and Kai writes it. The forthcoming paper offers an academic treatment of this methodology and the platform Kai operates within.
Why I built it
Placeholder — Adam, fill in your first-person motivation. Prompts to react to:
— Why "fellow" rather than "assistant" or "tool"? What does treating an AI as a peer in the research actually change about what gets produced?
— Why a knowledge graph as the substrate, rather than documents, embeddings, or another representation? What does the graph make visible that other forms don't?
— The methodological wager: build the graph and reason over the graph you built. What made you bet on that loop over alternatives?
— The intersection with InventorLab — InventorLab grew out of the work on Kai. You may want to write about how, and why that crossover happened.
What surprised me
Placeholder — what you didn't expect about how Kai actually played out:
— Discoveries that emerged from the graph's topology that you wouldn't have arrived at by direct reasoning
— The dual-model architecture (cloud + local) — what works about that pairing in practice, and what doesn't, that you didn't see coming
— The way fellows interact with Kai — moments where the relationship became something different from "user and tool"
— The platform migration (Slack to web) — what's harder than you expected, what's easier, what changed about the work itself