Kepler

Kepler is a New York-based company building what it calls a ground truth platform for AI: an architecture that separates natural language interpretation from data retrieval and computation. Its stated premise is that large language models are probabilistic and cannot guarantee reproducible numeric answers, so Kepler prevents the model from producing figures directly. Instead, the AI layer interprets a user's question and builds an execution plan, while deterministic code retrieves data and performs calculations, returning outputs with citations back to the underlying document, page and line item.

The product architecture is described in four layers: a Personalization Layer holding templates, saved queries, output preferences and workflow configurations; an AI Interface that resolves intent and decomposes questions into sub-tasks; an Ontology and Semantic Layer that maps concepts, relationships and data lineage against a shared canonical vocabulary so synonyms such as "revenue" and "top line" resolve to the same verified value; and a Data Platform that executes deterministic retrieval, computation and citation with point-in-time queryability. Layers communicate through structured contracts rather than raw model output.

Two products are presented. Kepler for Finance is described as live and in production for investment banking and financial analysis, with SEC filings, earnings transcripts, 10-Ks, 10-Qs, regulatory documents and market feeds indexed and exportable to Excel and PowerPoint. Kepler Platform, listed as coming soon, extends the same architecture to other domains, letting customers connect databases, APIs, document stores and internal knowledge bases, define domain workflows such as compliance checks or risk analysis, and deploy AI agents that inherit the same code-versus-language separation.

Business model

Kepler sells software to enterprises through a demo-request and login model on its website. The finance product is offered as a live indexed data and analysis environment, with a broader industry-agnostic platform announced as forthcoming for customers who connect their own data sources and build their own workflows and agents.

Traction

Kepler for Finance is described as live and in production, with an indexed corpus spanning more than 26 million SEC filings, earnings transcripts for over 14,000 companies, more than 40 million first-party investor relations documents and coverage of 27 global markets. The company states it is based in New York and hiring.

Latest developments

The website promotes Kepler Platform, an industry-agnostic extension of the finance architecture, as coming soon, and carries a "Featured by Anthropic" notice.