Small model.
Big possibilities.

A browser extension evaluates social posts against custom rules using Jev and blurs matching content with a reveal control. It includes shareable rule packs and a local heuristic simulation when no API key is configured.

A RuneBench harness connects Jev to a bounded action catalog, a tick-based controller and a live dashboard. It explores game decisions and records benchmark runs; results depend on the test setup and skill.

The chatwoot-router worker uses Jev to classify incoming support messages by topic and type and choose an owner. The repository also contains a separate Discord relay; that relay is not the Jev-powered component.

Official n8n nodes bring Jev decisions into workflows. Evaluate returns typed Choice, Score and Noul answers; Route sends items to branches using an answer and confidence threshold, with a fallback output.

A command-line tool that sorts lines of text or entire documents by a plain-English criterion using pairwise comparisons from the Jev model, producing a Bradley-Terry scale with scores and standard errors.

An independent API-based audit of Jev that examines how confidence relates to observed correctness, including experiments with abstention. The published scripts and results are research evidence, not a guarantee for other datasets.

tripwire is an AI SDK middleware that judges every LLM response before the user sees it, running seven safety checks (PII leak, injection detection, tone, topic, hallucination risk, etc.) in a single ~100ms call to TypeSafe's Jev.

A UNIX filter that ingests JSONL security alerts and outputs typed decisions (close, escalate, contain, investigate) using TypeSafe Jev or any OpenAI-compatible local model. Policy routing is handled in ordinary Rust code, not prompts.

A command-line tool that uses Jev, a System One model, to score each line against a user-defined meaning and optionally combine meanings with AND/OR/NOT, enabling semantic search that does not rely on keywords.

Adds natural-language row filtering to MySQL using TypeSafe Jev. Developers describe a matching condition in plain language and use model judgments to select relevant records.

A pytest plugin that sends claims about model output to Jev and checks the returned probabilities against explicit thresholds. It supports multiple assertions per request and cached results for repeat runs.

An MCP tool that uses Jev to batch-classify files, webpages, or text snippets before they reach an agent's context, returning only relevance probabilities to reduce token waste.