Improving Agent Accuracy With Harvey Graph
By Steven Van ·
Harvey's context layer maps US case law citations and per-Space matter data, so agents ask users for less background before starting work.
Harvey Graph is the name Harvey gives to the specialized graphs that organize legal and matter data and pass that structure to agents at run time. Harvey says the aim is for agents to start a task already knowing the matter, the documents, the precedents and the user's preferences, so people have to explain less up front. The company describes this in a blog post.
Why relationships matter
Harvey argues that legal work is relational. A credit agreement comes with drafts, amendments and side letters, and an agent has to know which version controls before it can answer. A diligence finding depends on a change-of-control clause, the counterparty it names and the deal timeline. Harvey says flat retrieval, which ranks chunks by similarity, finds passages rather than relationships, so agents end up re-discovering the same entities and cross-references on every task, spending time and tokens.
These connections are not stated in the documents. They have to be inferred, checked against the source and kept current, across millions of documents, under existing permissions and ethical walls.
What is in Harvey Graph today
- Knowledge graph: a structured map of US Case Law and the citations and relationships connecting it.
- Context graph: a per-Space map of how matters, data and documents from Vaults and integrations connect.
With the release of Harvey II this month, Spaces provide a closed, private graph of a matter's data and how it connects to the people who use it. For US Case Law, Harvey built the graph across the whole corpus: millions of opinions from every federal district and appellate court and all 50 states. It extracted who cites whom and how a case has been treated. When a question comes in, the Harvey agent retrieves authority, follows the citations, checks how those cases were treated, and checks its own work against the question.
How it is used
Users do not need to activate or manage the graph. The specialized graphs follow a Space's governance and security settings automatically. Harvey says users should notice the product asking less of them up front, with work that can be verified through citations.
Planned additions
Over the coming months, Harvey says it will launch precedent flags and cite checking for US case law, law evolution tracking across authorities, a new legal research surface for following these connections, and the ability for firms to define their own legal ontology on top of their documents in Harvey.
