MENLO PARK, Calif. — Corepass, based on Sand Hill Road here, is working on three layers of artificial intelligence at once: its own models, agents that do work with them and a developer platform meant to let others build on both. It calls itself a full-stack AI company, and its plan starts with code.
“We start with coding, where work can be run, checked, and improved,” the company says on its about page. The reasoning is practical: software is a domain where an agent’s output can be tested, and where a wrong answer shows up as a failing check rather than a confident paragraph.
Three layers
The models are grouped into three lines. Scout covers lightweight specialists; its first 4-billion-parameter model has finished training on real codebases and is used for code search. Voyager and Horizon, a balanced model and a flagship reasoning model, are in development.
The agent is Remy, which writes code and checks its own work. The platform — a model API, an agent API, and tools and a runtime, around an agent harness that handles planning, context and task execution — is not yet open to outside developers. The company says access will come in stages.
- ModelsScout, Voyager and Horizon
- Agent harnessPlanning, context and task execution
- PlatformModel API, Agent API, tools and runtime
Industry tools
Alongside the developer products, Corepass sells software for lending and sales. LendingOS is a loan origination workspace for private lenders and brokers, with an underwriting tool, Proclara, and a sales workspace, SalesOS, built in. Wavus is a prospect intelligence tool that finds and enriches the prospects who feed it. The company says it has 25 years of proprietary data across deals, loans and markets, and that it is trusted by lenders, institutional investors and financial platforms.
“We start with coding, where work can be run, checked, and improved.”
How it says it works
Corepass names three principles: start with real tasks, keep work reviewable and use resources thoughtfully. The second shows up most clearly in Remy, which reports the checks it ran and what it did not verify, and in Proclara, which traces the reasoning behind each underwriting finding.
Hiring
The company is hiring for five roles: a forward deployed engineer, an LLM algorithm engineer, an LLM application engineer for agent systems, a product marketer and a go-to-market and partnerships specialist. The positions are in Menlo Park and the San Francisco Bay Area, five days a week in the office. Details are on the careers page.