Managed agent platforms are now emerging from frontier labs and cloud providers.
So why build our own cloud agent platform at Harvey?
Because legal agents have three hard requirements: model flexibility, zero data retention, and cost control.
1) Model flexibility.
Lawyers can’t be locked into a single model provider. Client conflicts and confidentiality requirements can make a single model off-limits for a matter.
Instead, Harvey routes to the best model for a task. Our recent Legal Agent Benchmark (LAB) shows that different frontier models excel in different legal practice areas. The optimal choice is a mix of models across different types of legal work.
2) Zero data retention.
Lawyers work with client data that is often privileged, confidential, and subject to strict contractual controls.
Zero data retention means that data is never written to persistent storage by the model provider or agent runtime. The major managed agent platforms do not offer ZDR, so we had to build our own.
3) Cost control.
Legal agents run over large corpora and can involve many rounds of retrieval, reasoning, and tool use in a single workflow. If every step goes to the largest frontier model, cost runs out of control.
Owning the runtime lets Harvey match model capability to task complexity, routing to open-source models when it makes sense.
Empirically, we see 3-5x cost reductions versus a frontier-only approach.
More from our cofounder @gabepereyra:
Gabe Pereyra@gabepereyra
English
