Harvey
Harvey is an AI work platform for legal and professional services, provided by the AI company of the same name and described on its official site as “One Platform for Legal Work.” It serves law firms, in-house legal departments, and professional service networks. According to the official description, the platform covers complex legal research, bulk document analysis, contract review, and regulatory-change tracking, with organized solutions for litigation, transactional, and in-house teams. Modules and features change, so harvey.ai is the reference.
Product modules and boundaries
The modules listed on the official site include Agents (purpose-built agents executing legal work end to end), Vault (secure document storage and bulk analysis), Knowledge (research across legal, regulatory, and tax domains), Spaces (shared workspaces with governance and ethical walls), Command Center (analytics and benchmarking), Contract Intelligence (review and negotiation support), Horizon Scanning (regulatory tracking), Harvey Mobile, Ecosystem (integrations with trusted sources), Memory (preferences carried across the platform), and Harvey Academy (training).
Harvey is not a consumer tool: it is sold to institutions — law firms and the legal and professional-service teams of companies — and the site does not list public pricing; procurement goes through the official website. Outputs of legal work produced on the platform still require review by qualified professionals, which is the basic boundary of any legal AI tool.
Discussion in the show
In the chapter “Harvey：开源模型与产品壁垒” (Harvey: open-source models and product moats), Orange relays figures Harvey disclosed: the company used to pay model providers 1.5 dollars for every dollar of revenue, and recently trained its own model based on Kimi’s K3, turning gross margin positive for the first time. Yang Pan builds his argument on this: only a company at Harvey’s level — a stable business loop with the ability to evaluate its own model swaps — can cut costs by moving to open-source models; most startups that have not found product-market fit should not copy it. Guizang adds that legal is a closed, moated vertical — one of the few that can last and is relatively insulated from model churn.
These are participant relays of public business figures and judgments about industry structure; figures such as “1.5 dollars per dollar of revenue” have not been corroborated by the show or by Harvey’s official channels. The claim that Harvey trained a model on Kimi K3 comes from the show’s relay; background on Kimi is on Kimi. At the end of the same chapter, Orange says “cheap, easy, and effortless — Muse has shown everyone the direction”, turning the discussion toward the Muse personal agent. The episode 004 chapter in the Chinese transcript holds the context; an English transcript is not available.
Frequently asked questions
What is Harvey?
Harvey is an AI work platform for legal and professional services, offering legal research, document analysis, contract review, and agentic workflows for law firms, in-house teams, and professional service networks; see harvey.ai.
Is Harvey only for law firms?
The focus is legal work, but not only firms: the official solutions cover litigation, transactional, in-house departments, and professional service networks. There is no public self-serve entry for individual users.
How is Harvey priced?
Harvey is sold to institutions, and the official site does not list public pricing; sales and demos go through the website. The “gross margin” discussed in episode 004 is a company-level business figure, not what customers pay.
Did Harvey swap frontier models for an open-source one?
This is what episode 004 relays from Harvey’s disclosed information: Harvey trained its own model based on Kimi’s K3, which turned its gross margin positive. The claim comes from the show’s relay of public reporting; details belong to Harvey’s official disclosures.