TypeSafe AI Launches Jev to Speed Up LangGraph Agents

TypeSafe AI has introduced Jev, a novel decision model designed specifically to handle structured judgment rather than text generation. Unlike traditional large language models that generate prose, Jev returns typed answers with probabilities based on the state and questions it is given. According to TypeSafe AI, this narrow focus allows Jev to run up to 200 times faster and 400 times cheaper than leading LLMs on routing and classification tasks.

To help developers orchestrate these rapid decisions, LangChain is highlighting how Jev integrates with LangGraph, its orchestration framework that sees over 60 million monthly downloads and is used by Fortune 50 companies. In a document review experiment for litigation, developers compared Jev against Sonnet. Jev completed the classification step five to six times faster than Sonnet, maintaining highly stable scores across 100 repeated test runs. The system used LangGraph to route documents, escalating to an LLM for redactions or to a human attorney when Jev flagged potential privilege issues.

For software practitioners, this integration enables a hybrid architecture where deterministic code maintains the workflow structure and Jev handles semantic branching. Browserbase applied this pattern to its Stagehand tool, using Jev to select interactive elements on a webpage. If Jev's confidence falls below a 0.7 threshold, the system falls back to a standard LLM. This approach reduced the median latency of the tool's action function from 1.97 seconds to just 0.46 seconds, representing a 4.3-fold speedup. This shift allows developers to prioritize low-cost models for basic decisions and escalate to frontier models only when necessary.