Date Extraction Without Letting the Model Do Math
Extracts specified dates from text documents without performing mathematical operations.
Jev
LLMJev is TypeSafe AI's decision model. It does not write text: you send it a block of state and a set of typed questions, and it returns probabilities in under half a second. Three primitives cover everything: Choice picks one option from a list, Score places the state on a rubric, Noul answers yes or no with a probability. It is built to sit inside code as a fast, cheap judgment call before or after the expensive LLM.
How to use Jev
POST to https://api.typesafe.ai/v1/systemone with a bearer key, a model of jev-latest, a state that is a string, object or array, and a questions map. Each question is a Choice (up to 255 options), a Score (2 to 10 ordered levels) or a Noul (yes/no). Read the answers map in code and apply your own thresholds. Python and JavaScript SDKs and an agent skill for Claude Code and Codex are available.
Best for
Intent routing and triage in front of an LLM, input and output guardrails, reranking retrieval shortlists, RAG passage filtering, composite scoring of resumes or leads, moderation decisions, and any high-volume classification where a text model is too slow or too expensive.
How it behaves
Returns typed answers with a probability distribution and a confidence figure, never prose. Evaluates every question in a request in parallel, so ten questions cost about the same time as one. Reads dates and numbers as text, cannot count, and answers the question you wrote rather than the one you meant.
Tips
Ask one thing per question and put every question in a single request. Give Choice options a what, a not_for and examples when they could be confused. Describe Score levels as situations, not degrees. Keep arithmetic, counting and date logic in your code, and route on confidence: act above 0.9, confirm between 0.5 and 0.9, hand off below 0.5.
Pricing
Early access via waitlist. $0.042 per million input tokens, output is free. Multiple questions in one request share the same input cost.
Prompts for Jev
See all 19Resume Fit Scoring Across Four Dimensions
Generate a JSON-based scorecard for evaluating resumes across four dimensions in alignment with job descriptions.
Lead-to-Ideal-Customer Fit Score
Generate a fit score and appropriate routing for inbound leads based on provided criteria.
Semantic Lint Rules for Pull Request Diffs
Generate structured semantic lint rules for pull request diffs using AI classification. Produces JSON that triages code changes before deeper review.
Semantic Rerank of a Retrieval Shortlist
Generate a judgment on whether a candidate passage provides specific information needed by a query excerpt, producing a JSON output.
RAG Passage Filter Before the Context Window
Generate a filtering mechanism for passages based on a user query using specific criteria and thresholds. Produces a code snippet for integrating passage evaluation into existing systems.